notebooks/bqml-feature-engineering.ipynb

{
  "cells": [
   {
    "cell_type": "markdown",
    "id": "cd1ac1e8-1e73-467a-8118-ec02c1ec4cb4",
    "metadata": {},
    "source": [
     "![ga4](https://www.google-analytics.com/collect?v=2&tid=G-6VDTYWLKX6&cid=1&en=page_view&sid=1&dl=statmike%2Fvertex-ai-mlops%2F03+-+BigQuery+ML+%28BQML%29&dt=BQML+Feature+Engineering+-+Create+Model+With+Transpose.ipynb)\n",
     "\n",
     "# BigQuery ML (BQML) - Create Model With Inline Transpose\n",
     "\n",
     "This notebook will explore preparing data (preprocessing) for machine learning with BigQuery ML's [`CREATE MODEL` statement](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#create_model_syntax) as part of the inline [`TRANSFORM` clause](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#transform).  This provides the benefit of also automatically including the same preprocessing during serving with the [`ML.PREDICT` function](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-predict) and transportability by including the preprocessing within the [exported models](https://cloud.google.com/bigquery/docs/exporting-models) and models [directly registered to Vertex AI Model Registry](https://cloud.google.com/bigquery/docs/create_vertex). This prevents a common and complex source of [training-serving skew](https://developers.google.com/machine-learning/guides/rules-of-ml#training-serving_skew) in MLOps!\n",
     "\n",
     "This notebook will cover:\n",
     "- Setup: Setting up a datasource in BigQuery\n",
     "- Step 1: Training a Boosted Tree model with `TRANSFORM` clause\n",
     "- Step 2: Online Serving with Vertex AI Prediction Endpoints\n",
     "- Exporting the model to GCS\n",
     "\n",
     "**This content will accompany the blog post - TBD**\n",
     "\n",
     "---\n",
     "\n",
     "**BigQuery ML (BQML)**\n",
     "\n",
     "BigQuery Machine Learning allows you to use `SQL` to constuct an ML workflow.  This is a great leap in productivity and flexibility when the data source is BigQuery and users are already familiar with `SQL`. Using just `SQL` multiple techniques can be used for model training and even include hyperparameter tuning.  Predictions can be served directly in BigQuery which also include explainability. Models can be registered to Vertex AI model registry for online predictions on Vertex AI Endpoints.\n",
     "\n",
     "**If you know:**\n",
     "```SQL\n",
     "CREATE TABLE {table name}\n",
     "    AS\n",
     "        SELECT ...\n",
     "        FROM ...\n",
     "        WHERE ...\n",
     "```\n",
     "\n",
     "**Then you are ready to try:**\n",
     "```SQL\n",
     "CREATE MODEL {model name} \n",
     "    OPTIONS (\n",
     "        MODEL_TYPE = ...\n",
     "        {more options}\n",
     "    )\n",
     "    AS\n",
     "        SELECT ...\n",
     "        FROM ...\n",
     "        WHERE ...\n",
     "```\n",
     "\n",
     "**You can even use:**\n",
     "```SQL\n",
     "CREATE OR REPLACE MODEL {model name}...\n",
     "```\n",
     "\n",
     "And\n",
     "\n",
     "```SQL\n",
     "CREATE MODEL IF NOT EXISTS {model name}...\n",
     "```\n",
     "\n",
     "**Functions for your ML Workflow**\n",
     "\n",
     "Model Evaluation with `ML.EVALUTE`:\n",
     "```SQL\n",
     "SELECT *\n",
     "FROM ML.EVALUTE(\n",
     "    MODEL {model name},\n",
     "    (\n",
     "        SELECT ...\n",
     "        FROM ...\n",
     "        WHERE ...),\n",
     "    STRUCT({optional parameters})\n",
     ")\n",
     "```\n",
     "\n",
     "Prediction Using Models with `ML.PREDICT`:\n",
     "```SQL\n",
     "SELECT *\n",
     "FROM ML.PREDICT(\n",
     "    MODEL {model name},\n",
     "    (\n",
     "        SELECT ...\n",
     "        FROM ...\n",
     "        WHERE ...),\n",
     "    STRUCT({optional parameters})\n",
     ")\n",
     "```\n",
     "\n",
     "And [Many More](https://cloud.google.com/bigquery/docs/e2e-journey)!\n",
     "\n",
     "---\n",
     "\n",
     "**Feature Engineering**\n",
     "\n",
     "Feature engineering, or [preprocessing](https://cloud.google.com/bigquery/docs/preprocess-overview), is part of making data ready for machine learning.  BigQuery ML [manual feature preprocessing](https://cloud.google.com/bigquery/docs/manual-preprocessing) functions are available to make this process simple within BigQuery.  \n",
     "\n",
     "Each `CREATE MODEL ...` statement will do [automatic feature preprocessing](https://cloud.google.com/bigquery/docs/auto-preprocessing) by default.  It is also possible to include manual feature preprocessing in the [`CREATE MODEL` statement](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#create_model_syntax) as a [`TRANSFORM` clause](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#transform) where it will also become part of the model serving.  Many of these functions even accompany [exported models](https://cloud.google.com/bigquery/docs/exporting-models) and models [directly registered to Vertex AI Model Registry](https://cloud.google.com/bigquery/docs/create_vertex).\n",
     "\n",
     "```SQL\n",
     "CREATE MODEL {model name} AS\n",
     "    TRANSFORM (\n",
     "        ML.{function name}() OVER() as {name},\n",
     "        ...\n",
     "    )\n",
     "    OPTIONS (\n",
     "        MODEL_TYPE = ...\n",
     "        {more options}\n",
     "    )\n",
     "    AS\n",
     "        SELECT ...\n",
     "        FROM ...\n",
     "        WHERE ...\n",
     "```\n",
     "  \n",
     "---\n",
     "\n",
     "**Getting Started With BigQuery ML**\n",
     "<p align=\"center\" width=\"100%\">\n",
     "A great place to start exploring what model types are available and the functions to help create an ML workflow with each model type is this site:\n",
     "    <center>\n",
     "        <span style=\"font-size:xx-large;\">\n",
     "        <a href=\"https://cloud.google.com/bigquery/docs/e2e-journey\">\n",
     "            End-to-end user journey for each model\n",
     "        </a>\n",
     "        </span>\n",
     "    </center>\n",
     "\n",
     "Another great resourse for getting started is the \"What is BigQuery ML?\" starting page which include a model selection guide.\n",
     "    <center>\n",
     "        <span style=\"font-size:xx-large;\">\n",
     "        <a href=\"https://cloud.google.com/bigquery/docs/bqml-introduction\">\n",
     "        What is BigQuery ML?\n",
     "        </a>\n",
     "        </span>\n",
     "    </center>\n",
     "</p>\n",
     "\n",
     "---\n",
     "\n",
     "**Prerequisites:**\n",
     "\n",
     "None\n",
     "\n",
     "**Services Used:**\n",
     "- BigQuery\n",
     "- Vertex AI: Workbench (this notebook), Model Registry, Endpoints\n",
     "- GCS\n",
     "\n",
     "**Resources:**\n",
     "- [BigQuery ML (BQML) Overview](https://cloud.google.com/bigquery/docs/bqml-introduction)\n",
     "- [Overview of BQML methods and workflows](https://cloud.google.com/bigquery/docs/e2e-journey)\n",
     "- [BigQuery](https://cloud.google.com/bigquery)\n",
     "    - [Documentation:](https://cloud.google.com/bigquery/docs/query-overview)\n",
     "    - [API:](https://cloud.google.com/bigquery/docs/reference/libraries-overview)\n",
     "        - [Clients](https://cloud.google.com/bigquery/docs/reference/libraries)\n",
     "            - [Python SDK:](https://github.com/googleapis/python-bigquery)\n",
     "            - [Python Library Reference:](https://cloud.google.com/python/docs/reference/bigquery/latest)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "f2a5b95e-31c8-449d-bbcc-a34914c8a54c",
    "metadata": {},
    "source": [
     "---\n",
     "## Colab Setup\n",
     "\n",
     "To run this notebook in Colab click [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/statmike/vertex-ai-mlops/blob/main/03%20-%20BigQuery%20ML%20(BQML)/BQML%20Feature%20Engineering%20-%20Create%20Model%20With%20Transpose.ipynb) and run the cells in this section.  Otherwise, skip this section."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 475,
    "id": "83f5236f-7534-4db2-9478-12aae3167c3c",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Updated property [core/project].\n"
      ]
     }
    ],
    "source": [
     "PROJECT_ID = 'statmike-mlops-349915' # replace with project ID"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 473,
    "id": "72be7359-1aca-42e8-a391-7df5078e0453",
    "metadata": {},
    "outputs": [],
    "source": [
     "try:\n",
     "    import google.colab\n",
     "    !pip install --upgrade google-cloud-aiplatform -q\n",
     "    from google.colab import auth\n",
     "    auth.authenticate_user()\n",
     "    !gcloud config set project {PROJECT_ID}\n",
     "except Exception:\n",
     "    pass"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "5953317d-65b5-4bd3-9c0b-6492e41deb4b",
    "metadata": {},
    "source": [
     "**RESTART RUNTIME**\n",
     "\n",
     "The Next cell will restart the runtime by first stopping it and then Colab will automatically restart - you may need to dismiss a popup warning letting you know about this unexpected restart.  This restart makes the installs above available to the current session."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 474,
    "id": "22c89282-765e-48da-a0f2-49fb0dbdca8a",
    "metadata": {},
    "outputs": [],
    "source": [
     "import IPython\n",
     "app = IPython.Application.instance()\n",
     "app.kernel.do_shutdown(True)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "db2b4bb7-e0a9-401e-a5be-ea46911380ba",
    "metadata": {},
    "source": [
     "---\n",
     "## Environment Setup"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "5d24f412-b929-49fc-b75a-473d57f26b59",
    "metadata": {},
    "source": [
     "inputs:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 2,
    "id": "3a56d0a7-26f6-4e99-bf10-0094d41946d7",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'statmike-mlops-349915'"
       ]
      },
      "execution_count": 2,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "project = !gcloud config get-value project\n",
     "PROJECT_ID = project[0]\n",
     "PROJECT_ID"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 3,
    "id": "5b3e9dd1-a10b-4ecd-9663-001434cee8ab",
    "metadata": {},
    "outputs": [],
    "source": [
     "REGION = 'us-central1'\n",
     "EXPERIMENT = 'feature_engineering'\n",
     "SERIES = 'bqml'\n",
     "\n",
     "# source data\n",
     "BQ_PROJECT = PROJECT_ID\n",
     "BQ_DATASET = 'feature_engineering'\n",
     "# BQ_TABLE is created below"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "e5bbcebd-2a8f-4af8-8328-08287f95da77",
    "metadata": {},
    "source": [
     "packages:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 70,
    "id": "9a1ce719-dfb2-4e74-966d-d4691adb3b70",
    "metadata": {},
    "outputs": [],
    "source": [
     "from google.cloud import aiplatform\n",
     "from google.cloud import bigquery\n",
     "from google.cloud import storage\n",
     "\n",
     "from datetime import datetime\n",
     "import matplotlib.pyplot as plt"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "455c2b60-090a-4e32-aba2-4cea15e89352",
    "metadata": {},
    "source": [
     "clients:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 5,
    "id": "830224f8-d2ed-4f79-bfbb-9bd451d78712",
    "metadata": {},
    "outputs": [],
    "source": [
     "aiplatform.init(project = PROJECT_ID, location = REGION)\n",
     "bq = bigquery.Client(project = PROJECT_ID, location = 'US')\n",
     "gcs = storage.Client(project = PROJECT_ID)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "f3f4be47-eb05-47d7-8598-f17844f90181",
    "metadata": {},
    "source": [
     "---\n",
     "# Data Source: BigQuery"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "dfe5bb7b-d579-4c84-b43a-70070b4687b9",
    "metadata": {
     "tags": []
    },
    "source": [
     "## Create BigQuery Dataset\n",
     "\n",
     "Datasets are like folders, they hold items like tables, models, views.  Datasets are regional resources."
    ]
   },
   {
    "cell_type": "markdown",
    "id": "465a1a34-24d9-4699-9292-3aaff8b36a7d",
    "metadata": {},
    "source": [
     "Create the dataset if missing:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 7,
    "id": "4582e449-a75d-4453-9ab0-9b7df753384b",
    "metadata": {},
    "outputs": [],
    "source": [
     "ds = bigquery.Dataset(f\"{BQ_PROJECT}.{BQ_DATASET}\")\n",
     "ds.location = 'us'\n",
     "ds.labels = {'experiment': f'{EXPERIMENT}'}\n",
     "ds = bq.create_dataset(dataset = ds, exists_ok = True)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "824b6ade-0a55-467b-a0ce-7e34c76c724a",
    "metadata": {},
    "source": [
     "List BigQuery datasets in the project:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 8,
    "id": "f472ff6f-b336-4095-99bc-968a11499618",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "applied_forecasting\n",
       "applied_forecasting_pipeline_bqml_arimaplus\n",
       "applied_genai\n",
       "applied_genai_v2\n",
       "bqml\n",
       "bqml_remote_model_tutorial\n",
       "crossvalidation\n",
       "explained_columns\n",
       "feature_engineering\n",
       "forecasting_8_tournament\n",
       "fraud\n",
       "github_api\n",
       "model_deployment_monitoring_5876762107113897984\n"
      ]
     }
    ],
    "source": [
     "datasets = list(bq.list_datasets())\n",
     "for d in datasets:\n",
     "    print(d.dataset_id)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "1d319aa5-24b0-437e-93fd-ee285d2fff38",
    "metadata": {},
    "source": [
     "Review dataset attributes:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 9,
    "id": "0e125588-b2a3-4b95-b11b-f048fa5896d3",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'feature_engineering'"
       ]
      },
      "execution_count": 9,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "ds.dataset_id"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 10,
    "id": "bcba3390-3639-444a-a235-203bbbc2edab",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'statmike-mlops-349915'"
       ]
      },
      "execution_count": 10,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "ds.project"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 11,
    "id": "3902230b-e0d9-455d-bc91-23690cd9a3f3",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'statmike-mlops-349915:feature_engineering'"
       ]
      },
      "execution_count": 11,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "ds.full_dataset_id"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 12,
    "id": "bd0d52e3-d4d0-4faf-94d6-081f7ccdcc48",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'/projects/statmike-mlops-349915/datasets/feature_engineering'"
       ]
      },
      "execution_count": 12,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "ds.path"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 13,
    "id": "8d6c1555-2160-42e5-80dd-b5c2aa9c6ab0",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'US'"
       ]
      },
      "execution_count": 13,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "ds.location"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "ab4c8dcb-3dc5-40f1-a509-0828d3c0ef21",
    "metadata": {},
    "source": [
     "---\n",
     "## Create BigQuery Table\n",
     "\n",
     "Familiar with the [Great British Bake Off](https://en.wikipedia.org/wiki/The_Great_British_Bake_Off)? Or maybe just familiar with intricate nature of [baking bread](https://en.wikipedia.org/wiki/Bread)?  The code below generates different bread recipes along with multiple judges ratings of the resulting bread.  These are designed to be a lot like [Ciabatta](https://en.wikipedia.org/wiki/Ciabatta).\n",
     "\n",
     "This SQL makes heavy use of [GENERATE_ARRAY()](https://cloud.google.com/bigquery/docs/reference/standard-sql/arrays#generating_arrays_of_integers) to create sequences of numbers.  The [RAND() function](https://cloud.google.com/bigquery/docs/reference/standard-sql/mathematical_functions#rand) is the primary mechanism for generating random data but it is uniform `[0, 1)` only. To generate random normal data with MEAN=0 and SD=1 the [Box-Muller Transformation](https://en.wikipedia.org/wiki/Box%E2%80%93Muller_transform) is implemented in SQL using two `RAND()` values as inputs.  \n",
     "\n",
     "> Given two random uniform variable, $u_1$ and $u_2$, the following Box-Muller transformation generates two random normal(0, 1) variables:\n",
     ">\n",
     "> $\\sqrt{-2\\ln(u_1)}\\sin(2\\pi u_2)$ OR $\\sqrt{-2\\ln(u_1)}\\cos(2\\pi u_2)$\n",
     "\n",
     "Expressing these as SQL:\n",
     "```SQL\n",
     "# ACOS(-1) = PI\n",
     "SQRT(-2*LN(RAND()))*SIN(2*ACOS(-1)*RAND())\n",
     "\n",
     "SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND())\n",
     "```\n",
     "\n",
     "**Note** When simulating data with BigQuery and using [CTEs (WITH statement)](https://cloud.google.com/bigquery/docs/reference/standard-sql/query-syntax#with_clause) it is important to note that each CTE will be called once per reference.  That means that using the `RAND()` function could generate different data on each reference.\n"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 22,
    "id": "d4a73840-7e46-4f6b-9a3d-738ea2a1071d",
    "metadata": {
     "tags": []
    },
    "outputs": [],
    "source": [
     "seasons = 20\n",
     "rounds = 12\n",
     "contestants = 20\n",
     "\n",
     "n = seasons * rounds * contestants\n",
     "BQ_TABLE = 'bread'\n",
     "\n",
     "query = f\"\"\"\n",
     "CREATE OR REPLACE TABLE `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}` AS \n",
     "WITH\n",
     "    STEP_0 AS (\n",
     "        SELECT\n",
     "            season, round,\n",
     "            CONCAT('c_', contestant) as contestant_id,\n",
     "            ROW_NUMBER() OVER() recipe,\n",
     "            DATE_ADD(DATE_SUB(DATE_TRUNC(DATE(EXTRACT(YEAR FROM CURRENT_DATE()), 6, 1), WEEK(FRIDAY)), INTERVAL {seasons + 1} - season YEAR), INTERVAL round - 1 WEEK) as airdate,\n",
     "            season_adjuster + contestant_adjuster + round_adjuster + 2*RAND() as adjuster\n",
     "        FROM (\n",
     "          SELECT *\n",
     "          FROM (\n",
     "            SELECT season, SUM(adjuster/2) OVER(ORDER BY season) AS season_adjuster\n",
     "            FROM (SELECT ROW_NUMBER() OVER() season,RAND() as adjuster FROM UNNEST(GENERATE_ARRAY(1, {seasons})))\n",
     "          )\n",
     "          JOIN (\n",
     "            SELECT CEILING(contestant/{contestants}) as season, contestant, 4*adjuster AS contestant_adjuster\n",
     "            FROM (SELECT ROW_NUMBER() OVER() contestant, RAND() as adjuster FROM UNNEST(GENERATE_ARRAY(1, {seasons * contestants})))\n",
     "          ) USING(season)\n",
     "          CROSS JOIN (\n",
     "            SELECT round, SUM(adjuster/6) OVER(ORDER BY round) AS round_adjuster\n",
     "            FROM (SELECT ROW_NUMBER() OVER() round, RAND() as adjuster FROM UNNEST(GENERATE_ARRAY(1, {rounds})))\n",
     "          )\n",
     "          ORDER BY season, round, contestant\n",
     "        )\n",
     "    ),\n",
     "    STEP_1 AS (\n",
     "        SELECT\n",
     "            pos as recipe,\n",
     "            * EXCEPT(pos)\n",
     "        FROM (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 500+10*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS flourAmt)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(8+4*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS saltAmt) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(8+4*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS yeastAmt) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(5+1*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS mix1Time) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(2+3*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS mix1Speed) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(4+4*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS mix2Time) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(4+2*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS mix2Speed) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 330+20*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS water1Amt) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 110+20*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS water2Amt) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(40+10*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS waterTemp) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 100+5*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS proveTime) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(35+10*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS restTime) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT CAST(20+10*RAND() AS INT64) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS bakeTime) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 425+15*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS bakeTemp) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 60+10*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS ambTemp) USING(pos)\n",
     "        JOIN (SELECT *, ROW_NUMBER() OVER() pos FROM UNNEST(ARRAY((SELECT 50+10*SQRT(-2*LN(RAND()))*COS(2*ACOS(-1)*RAND()) FROM UNNEST(GENERATE_ARRAY(1, {n}))))) AS ambHumidity) USING(pos)\n",
     "    ),\n",
     "    STEP_2 AS (\n",
     "        SELECT *,\n",
     "            CASE \n",
     "                WHEN flourAmt-500 > 0 THEN 0.3*(flourAmt-500)\n",
     "                ELSE -0.3*(flourAmt-500)\n",
     "            END AS A,\n",
     "            CASE \n",
     "                WHEN saltAmt-10 > 0 THEN 2*(saltAmt-10)/2\n",
     "                ELSE -0.5*(saltAmt-10)/2\n",
     "            END AS B,\n",
     "            CASE \n",
     "                WHEN yeastAmt-10 > 0 THEN 5*(yeastAmt-10)/2\n",
     "                ELSE -5*(yeastAmt-10)/2\n",
     "            END AS C,\n",
     "            CASE \n",
     "                WHEN mix1Time-5 > 0 THEN 2*(mix1Time-5)\n",
     "                ELSE -2*(mix1Time-5)\n",
     "            END AS D,\n",
     "            CASE \n",
     "                WHEN mix1Speed-3.5 > 0 THEN 0.1*(mix1Speed-3.5)/1.5\n",
     "                ELSE -0.1*(mix1Speed-3.5)/1.5\n",
     "            END AS E,\n",
     "            CASE \n",
     "                WHEN mix2Time-6 > 0 THEN (mix2Time-6)/2\n",
     "                ELSE -2*(mix2Time-6)\n",
     "            END AS F,\n",
     "            CASE \n",
     "                WHEN mix2Speed-5 > 0 THEN 2*(mix2Speed-5)\n",
     "                ELSE -2*(mix2Speed-5)/2\n",
     "            END AS G,\n",
     "            CASE \n",
     "                WHEN water1Amt-330 > 0 THEN (water1Amt-330)/20\n",
     "                ELSE -1*(water1Amt-330)/20\n",
     "            END AS H,\n",
     "            CASE \n",
     "                WHEN water2Amt-110 > 0 THEN 6*(water2Amt-110)/20\n",
     "                ELSE -6*(water2Amt-110)/20\n",
     "            END AS I,\n",
     "            CASE \n",
     "                WHEN restTime-40 > 0 THEN 3*(restTime-40)/5\n",
     "                ELSE -3*(restTime-40)/5\n",
     "            END AS J,\n",
     "            CASE \n",
     "                WHEN bakeTime-25 > 0 THEN 6*(bakeTime-25)/5\n",
     "                ELSE -6*(bakeTime-25)/5\n",
     "            END AS K,\n",
     "            CASE \n",
     "                WHEN bakeTemp-425 > 0 THEN 15*(bakeTemp-425)/15\n",
     "                ELSE -10*(bakeTemp-425)/15\n",
     "            END AS L,\n",
     "            -1*((425-bakeTemp)*(bakeTime-25))/25 AS M\n",
     "        FROM STEP_1\n",
     "    ),\n",
     "    STEP_3 AS (\n",
     "        SELECT * EXCEPT(A, B, C, D, E, F, G, H, I ,J, K, L, M, adjuster),\n",
     "            CASE\n",
     "                WHEN 100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster < 0 THEN 0\n",
     "                WHEN 100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster > 100 THEN 100\n",
     "                ELSE ROUND(100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster)\n",
     "            END AS JUDGE_A,\n",
     "            CASE\n",
     "                WHEN 5-26*(90/69)+(90/69)*(100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster)-5 < 0 THEN 0\n",
     "                WHEN 5-26*(90/69)+(90/69)*(100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster)-5 > 100 THEN 100\n",
     "                ELSE ROUND(5-26*(90/69)+(90/69)*(100-(0+A+B+C+D+E+F+G+H+I+J+K+L+M)+adjuster)-5)\n",
     "            END AS JUDGE_B\n",
     "        FROM (\n",
     "            SELECT *\n",
     "            FROM (SELECT * FROM STEP_0)\n",
     "            JOIN (SELECT * FROM STEP_2) USING(recipe)\n",
     "        )\n",
     "    )\n",
     "SELECT *\n",
     "FROM STEP_3\n",
     "ORDER BY season, round, contestant_id\n",
     "\"\"\"\n",
     "#print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 15,
    "id": "9af563de-9263-4327-a179-df38d2e0f0d5",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "3.28"
       ]
      },
      "execution_count": 15,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "job = bq.query(query = query)\n",
     "job.result()\n",
     "(job.ended-job.started).total_seconds()"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 16,
    "id": "685a0258-e29d-487b-91ff-6d83cb5d8b16",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "0"
       ]
      },
      "execution_count": 16,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "job.total_bytes_processed"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "fe6afc36-66e6-4cbe-b454-f1a1ac8d9f80",
    "metadata": {},
    "source": [
     "### Check out this table in BigQuery Console:\n",
     "- Click: https://console.cloud.google.com/bigquery\n",
     "- Make sure project selected is the one from this notebook\n",
     "- Under Explore, expand this project and review the dataset and table"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 23,
    "id": "3f0547ca-ee3b-454f-a896-1448ac37a19e",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Direct Link To This Project In BigQuery:\n",
       "https://console.cloud.google.com/bigquery?project=statmike-mlops-349915&ws=!1m5!1m4!4m3!1sstatmike-mlops-349915!2sfeature_engineering!3sbread\n"
      ]
     }
    ],
    "source": [
     "print(f\"Direct Link To This Project In BigQuery:\\nhttps://console.cloud.google.com/bigquery?project={BQ_PROJECT}&ws=!1m5!1m4!4m3!1s{BQ_PROJECT}!2s{BQ_DATASET}!3s{BQ_TABLE}\")"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "f62b895e-601a-437f-a85f-faef600be711",
    "metadata": {},
    "source": [
     "---\n",
     "## Review Data From BigQuery\n",
     "Additional SQL queries could be used to review the data.  This section shows moving the table to a Pandas dataframe for local review in Python:\n",
     "\n",
     "> **Note:** <p>This query only selects a few columns.  This means BigQuery scans less data as it does not process the other columns.  </p>"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 18,
    "id": "8ffea105-a663-49ec-a064-0fab4fc2e2e6",
    "metadata": {},
    "outputs": [],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT season, JUDGE_A, JUDGE_B\n",
     "    FROM `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}`\n",
     "\"\"\"\n",
     "df = bq.query(query = query).to_dataframe()"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 19,
    "id": "9ab542ea-4447-47a8-97e4-7478060faf30",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT season, JUDGE_A, JUDGE_B\n",
       "    FROM `statmike-mlops-349915.feature_engineering.bread`\n",
       "\n"
      ]
     }
    ],
    "source": [
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 20,
    "id": "bb951d62-aaee-4ef2-bd5b-2cc85f1313a5",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>JUDGE_A</th>\n",
        "      <th>JUDGE_B</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>count</th>\n",
        "      <td>4800.000000</td>\n",
        "      <td>4800.000000</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>mean</th>\n",
        "      <td>77.627292</td>\n",
        "      <td>67.344583</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>std</th>\n",
        "      <td>10.069877</td>\n",
        "      <td>13.129074</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>min</th>\n",
        "      <td>28.000000</td>\n",
        "      <td>3.000000</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>25%</th>\n",
        "      <td>72.000000</td>\n",
        "      <td>60.000000</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>50%</th>\n",
        "      <td>79.000000</td>\n",
        "      <td>69.000000</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>75%</th>\n",
        "      <td>85.000000</td>\n",
        "      <td>77.000000</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>max</th>\n",
        "      <td>100.000000</td>\n",
        "      <td>100.000000</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "           JUDGE_A      JUDGE_B\n",
        "count  4800.000000  4800.000000\n",
        "mean     77.627292    67.344583\n",
        "std      10.069877    13.129074\n",
        "min      28.000000     3.000000\n",
        "25%      72.000000    60.000000\n",
        "50%      79.000000    69.000000\n",
        "75%      85.000000    77.000000\n",
        "max     100.000000   100.000000"
       ]
      },
      "execution_count": 20,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "df[['JUDGE_A', 'JUDGE_B']].describe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "7e314fa2-9be3-49fc-95e0-4880d9aced58",
    "metadata": {},
    "source": [
     "It looks like Judge A is not as harsh as Judge B!"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 21,
    "id": "a484cce9-9c82-4c1f-b2ed-819bf6301211",
    "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "/opt/conda/lib/python3.7/site-packages/ipykernel_launcher.py:1: FutureWarning: Indexing with multiple keys (implicitly converted to a tuple of keys) will be deprecated, use a list instead.\n",
       "  \"\"\"Entry point for launching an IPython kernel.\n"
      ]
     },
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead tr th {\n",
        "        text-align: left;\n",
        "    }\n",
        "\n",
        "    .dataframe thead tr:last-of-type th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr>\n",
        "      <th></th>\n",
        "      <th colspan=\"8\" halign=\"left\">JUDGE_A</th>\n",
        "      <th colspan=\"8\" halign=\"left\">JUDGE_B</th>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th></th>\n",
        "      <th>count</th>\n",
        "      <th>mean</th>\n",
        "      <th>std</th>\n",
        "      <th>min</th>\n",
        "      <th>25%</th>\n",
        "      <th>50%</th>\n",
        "      <th>75%</th>\n",
        "      <th>max</th>\n",
        "      <th>count</th>\n",
        "      <th>mean</th>\n",
        "      <th>std</th>\n",
        "      <th>min</th>\n",
        "      <th>25%</th>\n",
        "      <th>50%</th>\n",
        "      <th>75%</th>\n",
        "      <th>max</th>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>season</th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "      <th></th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>240.0</td>\n",
        "      <td>74.370833</td>\n",
        "      <td>10.395122</td>\n",
        "      <td>36.0</td>\n",
        "      <td>69.75</td>\n",
        "      <td>76.0</td>\n",
        "      <td>81.25</td>\n",
        "      <td>94.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>63.066667</td>\n",
        "      <td>13.509478</td>\n",
        "      <td>13.0</td>\n",
        "      <td>56.75</td>\n",
        "      <td>65.0</td>\n",
        "      <td>72.00</td>\n",
        "      <td>88.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>240.0</td>\n",
        "      <td>74.295833</td>\n",
        "      <td>10.309437</td>\n",
        "      <td>36.0</td>\n",
        "      <td>68.75</td>\n",
        "      <td>76.0</td>\n",
        "      <td>82.00</td>\n",
        "      <td>95.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>62.970833</td>\n",
        "      <td>13.483726</td>\n",
        "      <td>13.0</td>\n",
        "      <td>55.00</td>\n",
        "      <td>65.0</td>\n",
        "      <td>73.00</td>\n",
        "      <td>90.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>240.0</td>\n",
        "      <td>75.075000</td>\n",
        "      <td>9.844109</td>\n",
        "      <td>35.0</td>\n",
        "      <td>70.00</td>\n",
        "      <td>77.0</td>\n",
        "      <td>82.00</td>\n",
        "      <td>94.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>64.037500</td>\n",
        "      <td>12.863406</td>\n",
        "      <td>11.0</td>\n",
        "      <td>57.75</td>\n",
        "      <td>66.0</td>\n",
        "      <td>73.25</td>\n",
        "      <td>89.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.125000</td>\n",
        "      <td>11.049574</td>\n",
        "      <td>35.0</td>\n",
        "      <td>71.00</td>\n",
        "      <td>78.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>95.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>65.400000</td>\n",
        "      <td>14.390780</td>\n",
        "      <td>12.0</td>\n",
        "      <td>58.00</td>\n",
        "      <td>68.0</td>\n",
        "      <td>75.25</td>\n",
        "      <td>91.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.570833</td>\n",
        "      <td>9.135153</td>\n",
        "      <td>45.0</td>\n",
        "      <td>71.75</td>\n",
        "      <td>77.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>96.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>65.979167</td>\n",
        "      <td>11.917586</td>\n",
        "      <td>24.0</td>\n",
        "      <td>59.00</td>\n",
        "      <td>67.0</td>\n",
        "      <td>75.00</td>\n",
        "      <td>92.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>240.0</td>\n",
        "      <td>77.933333</td>\n",
        "      <td>9.356924</td>\n",
        "      <td>31.0</td>\n",
        "      <td>73.00</td>\n",
        "      <td>79.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>94.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>67.725000</td>\n",
        "      <td>12.160339</td>\n",
        "      <td>7.0</td>\n",
        "      <td>61.75</td>\n",
        "      <td>70.0</td>\n",
        "      <td>76.00</td>\n",
        "      <td>89.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.900000</td>\n",
        "      <td>10.181121</td>\n",
        "      <td>46.0</td>\n",
        "      <td>71.00</td>\n",
        "      <td>78.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>97.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>66.366667</td>\n",
        "      <td>13.238674</td>\n",
        "      <td>26.0</td>\n",
        "      <td>59.00</td>\n",
        "      <td>68.0</td>\n",
        "      <td>76.00</td>\n",
        "      <td>93.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.504167</td>\n",
        "      <td>9.729822</td>\n",
        "      <td>28.0</td>\n",
        "      <td>72.00</td>\n",
        "      <td>77.5</td>\n",
        "      <td>83.00</td>\n",
        "      <td>96.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>65.854167</td>\n",
        "      <td>12.695648</td>\n",
        "      <td>3.0</td>\n",
        "      <td>60.00</td>\n",
        "      <td>67.0</td>\n",
        "      <td>74.00</td>\n",
        "      <td>91.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>240.0</td>\n",
        "      <td>77.954167</td>\n",
        "      <td>10.201833</td>\n",
        "      <td>48.0</td>\n",
        "      <td>72.00</td>\n",
        "      <td>80.0</td>\n",
        "      <td>85.00</td>\n",
        "      <td>100.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>67.808333</td>\n",
        "      <td>13.264478</td>\n",
        "      <td>28.0</td>\n",
        "      <td>60.00</td>\n",
        "      <td>70.0</td>\n",
        "      <td>77.00</td>\n",
        "      <td>97.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>10</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.370833</td>\n",
        "      <td>9.875797</td>\n",
        "      <td>38.0</td>\n",
        "      <td>70.00</td>\n",
        "      <td>77.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>96.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>65.704167</td>\n",
        "      <td>12.862972</td>\n",
        "      <td>16.0</td>\n",
        "      <td>57.00</td>\n",
        "      <td>67.0</td>\n",
        "      <td>75.00</td>\n",
        "      <td>91.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>11</th>\n",
        "      <td>240.0</td>\n",
        "      <td>77.358333</td>\n",
        "      <td>10.202384</td>\n",
        "      <td>41.0</td>\n",
        "      <td>72.00</td>\n",
        "      <td>79.5</td>\n",
        "      <td>85.00</td>\n",
        "      <td>95.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>66.987500</td>\n",
        "      <td>13.295320</td>\n",
        "      <td>19.0</td>\n",
        "      <td>60.00</td>\n",
        "      <td>70.0</td>\n",
        "      <td>76.00</td>\n",
        "      <td>90.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>12</th>\n",
        "      <td>240.0</td>\n",
        "      <td>79.345833</td>\n",
        "      <td>9.038804</td>\n",
        "      <td>49.0</td>\n",
        "      <td>73.00</td>\n",
        "      <td>80.0</td>\n",
        "      <td>86.00</td>\n",
        "      <td>97.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>69.575000</td>\n",
        "      <td>11.806078</td>\n",
        "      <td>30.0</td>\n",
        "      <td>61.00</td>\n",
        "      <td>71.0</td>\n",
        "      <td>78.25</td>\n",
        "      <td>92.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>13</th>\n",
        "      <td>240.0</td>\n",
        "      <td>77.950000</td>\n",
        "      <td>10.149386</td>\n",
        "      <td>35.0</td>\n",
        "      <td>72.00</td>\n",
        "      <td>79.0</td>\n",
        "      <td>86.00</td>\n",
        "      <td>97.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>67.775000</td>\n",
        "      <td>13.223513</td>\n",
        "      <td>12.0</td>\n",
        "      <td>60.00</td>\n",
        "      <td>69.0</td>\n",
        "      <td>78.00</td>\n",
        "      <td>92.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>14</th>\n",
        "      <td>240.0</td>\n",
        "      <td>76.816667</td>\n",
        "      <td>10.965891</td>\n",
        "      <td>39.0</td>\n",
        "      <td>71.00</td>\n",
        "      <td>79.0</td>\n",
        "      <td>84.00</td>\n",
        "      <td>99.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>66.345833</td>\n",
        "      <td>14.332030</td>\n",
        "      <td>16.0</td>\n",
        "      <td>59.00</td>\n",
        "      <td>69.5</td>\n",
        "      <td>76.00</td>\n",
        "      <td>96.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>15</th>\n",
        "      <td>240.0</td>\n",
        "      <td>78.812500</td>\n",
        "      <td>10.375495</td>\n",
        "      <td>31.0</td>\n",
        "      <td>73.00</td>\n",
        "      <td>80.5</td>\n",
        "      <td>86.00</td>\n",
        "      <td>100.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>68.870833</td>\n",
        "      <td>13.520014</td>\n",
        "      <td>6.0</td>\n",
        "      <td>62.00</td>\n",
        "      <td>71.0</td>\n",
        "      <td>79.00</td>\n",
        "      <td>96.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>16</th>\n",
        "      <td>240.0</td>\n",
        "      <td>79.270833</td>\n",
        "      <td>9.734422</td>\n",
        "      <td>31.0</td>\n",
        "      <td>74.00</td>\n",
        "      <td>80.0</td>\n",
        "      <td>86.00</td>\n",
        "      <td>100.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>69.525000</td>\n",
        "      <td>12.693004</td>\n",
        "      <td>7.0</td>\n",
        "      <td>62.75</td>\n",
        "      <td>71.0</td>\n",
        "      <td>78.00</td>\n",
        "      <td>97.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>17</th>\n",
        "      <td>240.0</td>\n",
        "      <td>80.325000</td>\n",
        "      <td>9.325057</td>\n",
        "      <td>50.0</td>\n",
        "      <td>75.00</td>\n",
        "      <td>82.0</td>\n",
        "      <td>87.00</td>\n",
        "      <td>98.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>70.850000</td>\n",
        "      <td>12.111511</td>\n",
        "      <td>32.0</td>\n",
        "      <td>64.00</td>\n",
        "      <td>73.0</td>\n",
        "      <td>79.00</td>\n",
        "      <td>94.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>18</th>\n",
        "      <td>240.0</td>\n",
        "      <td>80.000000</td>\n",
        "      <td>9.583371</td>\n",
        "      <td>52.0</td>\n",
        "      <td>75.00</td>\n",
        "      <td>81.0</td>\n",
        "      <td>86.00</td>\n",
        "      <td>99.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>70.379167</td>\n",
        "      <td>12.500125</td>\n",
        "      <td>33.0</td>\n",
        "      <td>64.00</td>\n",
        "      <td>72.0</td>\n",
        "      <td>79.00</td>\n",
        "      <td>96.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>19</th>\n",
        "      <td>240.0</td>\n",
        "      <td>79.504167</td>\n",
        "      <td>9.109744</td>\n",
        "      <td>48.0</td>\n",
        "      <td>74.00</td>\n",
        "      <td>81.0</td>\n",
        "      <td>86.00</td>\n",
        "      <td>100.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>69.825000</td>\n",
        "      <td>11.889774</td>\n",
        "      <td>28.0</td>\n",
        "      <td>62.75</td>\n",
        "      <td>72.0</td>\n",
        "      <td>78.00</td>\n",
        "      <td>97.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>20</th>\n",
        "      <td>240.0</td>\n",
        "      <td>81.062500</td>\n",
        "      <td>9.327617</td>\n",
        "      <td>44.0</td>\n",
        "      <td>75.00</td>\n",
        "      <td>82.0</td>\n",
        "      <td>88.00</td>\n",
        "      <td>100.0</td>\n",
        "      <td>240.0</td>\n",
        "      <td>71.845833</td>\n",
        "      <td>12.242203</td>\n",
        "      <td>23.0</td>\n",
        "      <td>64.00</td>\n",
        "      <td>73.0</td>\n",
        "      <td>80.00</td>\n",
        "      <td>100.0</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "       JUDGE_A                                                        JUDGE_B  \\\n",
        "         count       mean        std   min    25%   50%    75%    max   count   \n",
        "season                                                                          \n",
        "1        240.0  74.370833  10.395122  36.0  69.75  76.0  81.25   94.0   240.0   \n",
        "2        240.0  74.295833  10.309437  36.0  68.75  76.0  82.00   95.0   240.0   \n",
        "3        240.0  75.075000   9.844109  35.0  70.00  77.0  82.00   94.0   240.0   \n",
        "4        240.0  76.125000  11.049574  35.0  71.00  78.0  84.00   95.0   240.0   \n",
        "5        240.0  76.570833   9.135153  45.0  71.75  77.0  84.00   96.0   240.0   \n",
        "6        240.0  77.933333   9.356924  31.0  73.00  79.0  84.00   94.0   240.0   \n",
        "7        240.0  76.900000  10.181121  46.0  71.00  78.0  84.00   97.0   240.0   \n",
        "8        240.0  76.504167   9.729822  28.0  72.00  77.5  83.00   96.0   240.0   \n",
        "9        240.0  77.954167  10.201833  48.0  72.00  80.0  85.00  100.0   240.0   \n",
        "10       240.0  76.370833   9.875797  38.0  70.00  77.0  84.00   96.0   240.0   \n",
        "11       240.0  77.358333  10.202384  41.0  72.00  79.5  85.00   95.0   240.0   \n",
        "12       240.0  79.345833   9.038804  49.0  73.00  80.0  86.00   97.0   240.0   \n",
        "13       240.0  77.950000  10.149386  35.0  72.00  79.0  86.00   97.0   240.0   \n",
        "14       240.0  76.816667  10.965891  39.0  71.00  79.0  84.00   99.0   240.0   \n",
        "15       240.0  78.812500  10.375495  31.0  73.00  80.5  86.00  100.0   240.0   \n",
        "16       240.0  79.270833   9.734422  31.0  74.00  80.0  86.00  100.0   240.0   \n",
        "17       240.0  80.325000   9.325057  50.0  75.00  82.0  87.00   98.0   240.0   \n",
        "18       240.0  80.000000   9.583371  52.0  75.00  81.0  86.00   99.0   240.0   \n",
        "19       240.0  79.504167   9.109744  48.0  74.00  81.0  86.00  100.0   240.0   \n",
        "20       240.0  81.062500   9.327617  44.0  75.00  82.0  88.00  100.0   240.0   \n",
        "\n",
        "                                                               \n",
        "             mean        std   min    25%   50%    75%    max  \n",
        "season                                                         \n",
        "1       63.066667  13.509478  13.0  56.75  65.0  72.00   88.0  \n",
        "2       62.970833  13.483726  13.0  55.00  65.0  73.00   90.0  \n",
        "3       64.037500  12.863406  11.0  57.75  66.0  73.25   89.0  \n",
        "4       65.400000  14.390780  12.0  58.00  68.0  75.25   91.0  \n",
        "5       65.979167  11.917586  24.0  59.00  67.0  75.00   92.0  \n",
        "6       67.725000  12.160339   7.0  61.75  70.0  76.00   89.0  \n",
        "7       66.366667  13.238674  26.0  59.00  68.0  76.00   93.0  \n",
        "8       65.854167  12.695648   3.0  60.00  67.0  74.00   91.0  \n",
        "9       67.808333  13.264478  28.0  60.00  70.0  77.00   97.0  \n",
        "10      65.704167  12.862972  16.0  57.00  67.0  75.00   91.0  \n",
        "11      66.987500  13.295320  19.0  60.00  70.0  76.00   90.0  \n",
        "12      69.575000  11.806078  30.0  61.00  71.0  78.25   92.0  \n",
        "13      67.775000  13.223513  12.0  60.00  69.0  78.00   92.0  \n",
        "14      66.345833  14.332030  16.0  59.00  69.5  76.00   96.0  \n",
        "15      68.870833  13.520014   6.0  62.00  71.0  79.00   96.0  \n",
        "16      69.525000  12.693004   7.0  62.75  71.0  78.00   97.0  \n",
        "17      70.850000  12.111511  32.0  64.00  73.0  79.00   94.0  \n",
        "18      70.379167  12.500125  33.0  64.00  72.0  79.00   96.0  \n",
        "19      69.825000  11.889774  28.0  62.75  72.0  78.00   97.0  \n",
        "20      71.845833  12.242203  23.0  64.00  73.0  80.00  100.0  "
       ]
      },
      "execution_count": 21,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "df.groupby('season')['JUDGE_A', 'JUDGE_B'].describe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "fc1b85f7-7f9b-4308-9096-0164a164dccb",
    "metadata": {},
    "source": [
     "The summary over seasons seems to indicate the judges are gradually grading easier (higher scores) or the contestants are getter better!"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "6f38ae87-86c1-4dc5-a1d2-1abc959c22b7",
    "metadata": {},
    "source": [
     "---\n",
     "# GOAL: Predicting Judges Ratings\n",
     "\n",
     "Each row of the BigQuery table has all the parameters of a bread recipe and the rating for the resulting bake by two judges.  The goal of the this ML project is to predict the judges rating from the recipe and understand which parameters of the recipe where most important overall ([global explainability](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-global-explain)) and for each recipe ([local explainability](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-explain-predict)). BigQuery models include AI Explanation functions to assist with this [explainable AI](https://cloud.google.com/bigquery/docs/xai-overview)."
    ]
   },
   {
    "cell_type": "markdown",
    "id": "a8faa0b2-dc37-4f3a-8df3-b82d5c9295d5",
    "metadata": {},
    "source": [
     "---\n",
     "# Feature Engineering With Inline TRANSFORM Clause"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "cb67a9ea-402a-406b-8bee-28dbd1af5a35",
    "metadata": {},
    "source": [
     "## Step 1: Train Model\n",
     "\n",
     "Use BigQuery ML to train a boosted tree:\n",
     "- [Boosted Tree](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create-boosted-tree) with BigQuery ML (BQML)\n",
     "\n",
     "Feature preprocessing with the `TRANSFORM` clause:\n",
     "- [TRANSFORM](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#transform)\n",
     "- Approach:\n",
     "    - The data includes columns that group the recipes into `season`, `round`.  The can be good for identify and understanding the data but for predicting future rating these might not be as helpful.  The approach here uses the `TRANSFORM` clause to omit these and instead use parts of `airdate` to more generally understand changes over time.\n",
     "    - 'contestant_id' is a string the identifies the unique baker.  These are actually increment values that represent the bakers (contestants) over time.  To make this more general for use in a predictive model the `ML.LABEL_ENCODER` function is used to transform the column into a feature with integers.\n",
     "    - Use a variety of [Numerical functions](https://cloud.google.com/bigquery/docs/manual-preprocessing#numerical_functions) for regularizing data with numerical expressions.\n",
     "\n",
     "Register the model to Vertex AI Model Registry:\n",
     "- `MODEL_REGISTRY` [training options](https://cloud.google.com/bigquery/docs/create_vertex) to register the resulting model in the [Vertex AI Model Registry](https://cloud.google.com/vertex-ai/docs/model-registry/introduction)."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 24,
    "id": "b66e4328-543d-4296-a7da-d1297bee07bf",
    "metadata": {},
    "outputs": [],
    "source": [
     "TIMESTAMP = datetime.now().strftime(\"%Y%m%d%H%M%S\")\n",
     "RUN_NAME = f'run-{TIMESTAMP}'\n",
     "BQ_MODEL = f'{SERIES}_{EXPERIMENT}_transform'"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 27,
    "id": "ea86eef8-c9a9-4c37-86d7-183d634fa203",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "CREATE OR REPLACE MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`\n",
       "TRANSFORM (\n",
       "    JUDGE_A,\n",
       "    ML.LABEL_ENCODER(contestant_id) OVER() as contestant,\n",
       "    EXTRACT(YEAR FROM airdate) as year,\n",
       "    EXTRACT(ISOWEEK FROM airdate) as week,\n",
       "\n",
       "    ML.MIN_MAX_SCALER(flourAmt) OVER() as scale_flourAmt, \n",
       "    ML.ROBUST_SCALER(saltAmt) OVER() as scale_saltAmt,\n",
       "    ML.MAX_ABS_SCALER(yeastAmt) OVER() as scale_yeastAmt,\n",
       "    ML.STANDARD_SCALER(water1Amt) OVER() as scale_water1Amt,\n",
       "    ML.STANDARD_SCALER(water2Amt) OVER() as scale_water2Amt,\n",
       "\n",
       "    ML.STANDARD_SCALER(waterTemp) OVER() as scale_waterTemp,\n",
       "    ML.ROBUST_SCALER(bakeTemp) OVER() as scale_bakeTemp,\n",
       "    ML.MIN_MAX_SCALER(ambTemp) OVER() as scale_ambTemp,\n",
       "    ML.MAX_ABS_SCALER(ambHumidity) OVER() as scale_ambHumidity,\n",
       "\n",
       "    ML.ROBUST_SCALER(mix1Time) OVER() as scale_mix1Time,\n",
       "    ML.ROBUST_SCALER(mix2Time) OVER() as scale_mix2Time,\n",
       "    ML.ROBUST_SCALER(mix1Speed) OVER() as scale_mix1Speed,\n",
       "    ML.ROBUST_SCALER(mix2Speed) OVER() as scale_mix2Speed,\n",
       "    ML.STANDARD_SCALER(proveTime) OVER() as scale_proveTime,\n",
       "    ML.MAX_ABS_SCALER(restTime) OVER() as scale_restTime,\n",
       "    ML.MAX_ABS_SCALER(bakeTime) OVER() as scale_bakeTime\n",
       ")\n",
       "OPTIONS (\n",
       "        model_type = 'BOOSTED_TREE_REGRESSOR',\n",
       "        booster_type = 'GBTREE',\n",
       "        num_parallel_tree = 25,\n",
       "        early_stop = TRUE,\n",
       "        min_rel_progress = 0.01,\n",
       "        tree_method = 'HIST',\n",
       "        subsample = 0.85,\n",
       "        #model_type = 'LINEAR_REG',        \n",
       "        input_label_cols = ['JUDGE_A'],\n",
       "        enable_global_explain = TRUE,\n",
       "        data_split_method = 'AUTO_SPLIT',\n",
       "        l1_reg = 10,\n",
       "        l2_reg = 10,\n",
       "        MODEL_REGISTRY = 'VERTEX_AI',\n",
       "        VERTEX_AI_MODEL_ID = 'bqml_bqml_feature_engineering_transform',\n",
       "        VERTEX_AI_MODEL_VERSION_ALIASES = ['run-20230705114026']\n",
       "    ) AS\n",
       "SELECT *\n",
       "FROM `statmike-mlops-349915.feature_engineering.bread`\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "CREATE OR REPLACE MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`\n",
     "TRANSFORM (\n",
     "    JUDGE_A,\n",
     "    ML.LABEL_ENCODER(contestant_id) OVER() as contestant,\n",
     "    EXTRACT(YEAR FROM airdate) as year,\n",
     "    EXTRACT(ISOWEEK FROM airdate) as week,\n",
     "\n",
     "    ML.MIN_MAX_SCALER(flourAmt) OVER() as scale_flourAmt, \n",
     "    ML.ROBUST_SCALER(saltAmt) OVER() as scale_saltAmt,\n",
     "    ML.MAX_ABS_SCALER(yeastAmt) OVER() as scale_yeastAmt,\n",
     "    ML.STANDARD_SCALER(water1Amt) OVER() as scale_water1Amt,\n",
     "    ML.STANDARD_SCALER(water2Amt) OVER() as scale_water2Amt,\n",
     "\n",
     "    ML.STANDARD_SCALER(waterTemp) OVER() as scale_waterTemp,\n",
     "    ML.ROBUST_SCALER(bakeTemp) OVER() as scale_bakeTemp,\n",
     "    ML.MIN_MAX_SCALER(ambTemp) OVER() as scale_ambTemp,\n",
     "    ML.MAX_ABS_SCALER(ambHumidity) OVER() as scale_ambHumidity,\n",
     "\n",
     "    ML.ROBUST_SCALER(mix1Time) OVER() as scale_mix1Time,\n",
     "    ML.ROBUST_SCALER(mix2Time) OVER() as scale_mix2Time,\n",
     "    ML.ROBUST_SCALER(mix1Speed) OVER() as scale_mix1Speed,\n",
     "    ML.ROBUST_SCALER(mix2Speed) OVER() as scale_mix2Speed,\n",
     "    ML.STANDARD_SCALER(proveTime) OVER() as scale_proveTime,\n",
     "    ML.MAX_ABS_SCALER(restTime) OVER() as scale_restTime,\n",
     "    ML.MAX_ABS_SCALER(bakeTime) OVER() as scale_bakeTime\n",
     ")\n",
     "OPTIONS (\n",
     "        model_type = 'BOOSTED_TREE_REGRESSOR',\n",
     "        booster_type = 'GBTREE',\n",
     "        num_parallel_tree = 25,\n",
     "        early_stop = TRUE,\n",
     "        min_rel_progress = 0.01,\n",
     "        tree_method = 'HIST',\n",
     "        subsample = 0.85,        \n",
     "        input_label_cols = ['JUDGE_A'],\n",
     "        enable_global_explain = TRUE,\n",
     "        data_split_method = 'AUTO_SPLIT',\n",
     "        l1_reg = 10,\n",
     "        l2_reg = 10,\n",
     "        MODEL_REGISTRY = 'VERTEX_AI',\n",
     "        VERTEX_AI_MODEL_ID = 'bqml_{BQ_MODEL}',\n",
     "        VERTEX_AI_MODEL_VERSION_ALIASES = ['{RUN_NAME}']\n",
     "    ) AS\n",
     "SELECT *\n",
     "FROM `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}`\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 28,
    "id": "810adc92-f1a7-4676-ab23-da88375dc1e7",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "531.572"
       ]
      },
      "execution_count": 28,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "job = bq.query(query = query)\n",
     "job.result()\n",
     "(job.ended-job.started).total_seconds()"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 29,
    "id": "eab37574-0613-4781-9aba-57a6ff95cd25",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "27788576190"
       ]
      },
      "execution_count": 29,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "job.total_bytes_processed"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "7cdf70fe-4d2c-4943-8d11-054d813bc8ee",
    "metadata": {},
    "source": [
     "Add labels to the model in BigQuery:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 30,
    "id": "f39f906c-de1b-4a16-be8d-e8aa2f21ff10",
    "metadata": {},
    "outputs": [],
    "source": [
     "bqml_model = bq.get_model(f'{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}')\n",
     "bqml_model.labels = {'series' : f'{SERIES}', 'experiment' : f'{EXPERIMENT}'}\n",
     "bqml_model = bq.update_model(bqml_model, ['labels'])"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "fe0b9a98-9317-4adb-914b-bfbe6ec81b86",
    "metadata": {},
    "source": [
     "### Feature and Training Review"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "37cb556f-be13-4985-a96d-fdb6de4ac9da",
    "metadata": {},
    "source": [
     "Review the Model Inputs, the feature information before preprocessing, with [ML.FEATURE_INFO](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-feature):"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 31,
    "id": "39e25006-13e6-42dd-8f97-ccd19da7bd85",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT *\n",
       "    FROM ML.FEATURE_INFO(MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`)\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT *\n",
     "    FROM ML.FEATURE_INFO(MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`)\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 32,
    "id": "3d9c7d4b-7198-49e1-93c9-7e00a57b0d04",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>input</th>\n",
        "      <th>min</th>\n",
        "      <th>max</th>\n",
        "      <th>mean</th>\n",
        "      <th>median</th>\n",
        "      <th>stddev</th>\n",
        "      <th>category_count</th>\n",
        "      <th>null_count</th>\n",
        "      <th>dimension</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>mix2Time</td>\n",
        "      <td>4.000000</td>\n",
        "      <td>8.000000</td>\n",
        "      <td>5.992685</td>\n",
        "      <td>6.000000</td>\n",
        "      <td>1.233598</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>water1Amt</td>\n",
        "      <td>249.055782</td>\n",
        "      <td>403.247837</td>\n",
        "      <td>330.043420</td>\n",
        "      <td>330.044068</td>\n",
        "      <td>20.219386</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>bakeTemp</td>\n",
        "      <td>370.462086</td>\n",
        "      <td>478.768391</td>\n",
        "      <td>425.368663</td>\n",
        "      <td>425.387130</td>\n",
        "      <td>15.143971</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>bakeTime</td>\n",
        "      <td>20.000000</td>\n",
        "      <td>30.000000</td>\n",
        "      <td>24.938088</td>\n",
        "      <td>25.000000</td>\n",
        "      <td>2.924194</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>proveTime</td>\n",
        "      <td>81.828180</td>\n",
        "      <td>117.068216</td>\n",
        "      <td>99.852456</td>\n",
        "      <td>99.821503</td>\n",
        "      <td>4.910245</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>mix1Speed</td>\n",
        "      <td>2.000000</td>\n",
        "      <td>5.000000</td>\n",
        "      <td>3.491379</td>\n",
        "      <td>3.000000</td>\n",
        "      <td>0.961326</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>water2Amt</td>\n",
        "      <td>44.858814</td>\n",
        "      <td>196.973103</td>\n",
        "      <td>109.679854</td>\n",
        "      <td>109.669713</td>\n",
        "      <td>19.945945</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>airdate</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>240</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>waterTemp</td>\n",
        "      <td>40.000000</td>\n",
        "      <td>50.000000</td>\n",
        "      <td>44.986416</td>\n",
        "      <td>45.000000</td>\n",
        "      <td>2.877033</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>ambHumidity</td>\n",
        "      <td>15.513824</td>\n",
        "      <td>88.402399</td>\n",
        "      <td>49.873886</td>\n",
        "      <td>50.167885</td>\n",
        "      <td>10.177659</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>10</th>\n",
        "      <td>yeastAmt</td>\n",
        "      <td>8.000000</td>\n",
        "      <td>12.000000</td>\n",
        "      <td>9.963427</td>\n",
        "      <td>10.000000</td>\n",
        "      <td>1.202827</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>11</th>\n",
        "      <td>flourAmt</td>\n",
        "      <td>464.847670</td>\n",
        "      <td>539.094446</td>\n",
        "      <td>499.910467</td>\n",
        "      <td>499.901335</td>\n",
        "      <td>9.889363</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>12</th>\n",
        "      <td>mix2Speed</td>\n",
        "      <td>4.000000</td>\n",
        "      <td>6.000000</td>\n",
        "      <td>4.998171</td>\n",
        "      <td>5.000000</td>\n",
        "      <td>0.710331</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>13</th>\n",
        "      <td>restTime</td>\n",
        "      <td>35.000000</td>\n",
        "      <td>45.000000</td>\n",
        "      <td>40.004702</td>\n",
        "      <td>40.000000</td>\n",
        "      <td>2.932919</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>14</th>\n",
        "      <td>ambTemp</td>\n",
        "      <td>27.328280</td>\n",
        "      <td>101.615204</td>\n",
        "      <td>60.073862</td>\n",
        "      <td>59.989266</td>\n",
        "      <td>9.860562</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>15</th>\n",
        "      <td>saltAmt</td>\n",
        "      <td>8.000000</td>\n",
        "      <td>12.000000</td>\n",
        "      <td>9.975444</td>\n",
        "      <td>10.000000</td>\n",
        "      <td>1.242452</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>16</th>\n",
        "      <td>mix1Time</td>\n",
        "      <td>2.000000</td>\n",
        "      <td>8.000000</td>\n",
        "      <td>5.015674</td>\n",
        "      <td>5.000000</td>\n",
        "      <td>1.037456</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>17</th>\n",
        "      <td>contestant_id</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>NaN</td>\n",
        "      <td>400</td>\n",
        "      <td>0</td>\n",
        "      <td>&lt;NA&gt;</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "            input         min         max        mean      median     stddev  \\\n",
        "0        mix2Time    4.000000    8.000000    5.992685    6.000000   1.233598   \n",
        "1       water1Amt  249.055782  403.247837  330.043420  330.044068  20.219386   \n",
        "2        bakeTemp  370.462086  478.768391  425.368663  425.387130  15.143971   \n",
        "3        bakeTime   20.000000   30.000000   24.938088   25.000000   2.924194   \n",
        "4       proveTime   81.828180  117.068216   99.852456   99.821503   4.910245   \n",
        "5       mix1Speed    2.000000    5.000000    3.491379    3.000000   0.961326   \n",
        "6       water2Amt   44.858814  196.973103  109.679854  109.669713  19.945945   \n",
        "7         airdate         NaN         NaN         NaN         NaN        NaN   \n",
        "8       waterTemp   40.000000   50.000000   44.986416   45.000000   2.877033   \n",
        "9     ambHumidity   15.513824   88.402399   49.873886   50.167885  10.177659   \n",
        "10       yeastAmt    8.000000   12.000000    9.963427   10.000000   1.202827   \n",
        "11       flourAmt  464.847670  539.094446  499.910467  499.901335   9.889363   \n",
        "12      mix2Speed    4.000000    6.000000    4.998171    5.000000   0.710331   \n",
        "13       restTime   35.000000   45.000000   40.004702   40.000000   2.932919   \n",
        "14        ambTemp   27.328280  101.615204   60.073862   59.989266   9.860562   \n",
        "15        saltAmt    8.000000   12.000000    9.975444   10.000000   1.242452   \n",
        "16       mix1Time    2.000000    8.000000    5.015674    5.000000   1.037456   \n",
        "17  contestant_id         NaN         NaN         NaN         NaN        NaN   \n",
        "\n",
        "    category_count  null_count  dimension  \n",
        "0             <NA>           0       <NA>  \n",
        "1             <NA>           0       <NA>  \n",
        "2             <NA>           0       <NA>  \n",
        "3             <NA>           0       <NA>  \n",
        "4             <NA>           0       <NA>  \n",
        "5             <NA>           0       <NA>  \n",
        "6             <NA>           0       <NA>  \n",
        "7              240           0       <NA>  \n",
        "8             <NA>           0       <NA>  \n",
        "9             <NA>           0       <NA>  \n",
        "10            <NA>           0       <NA>  \n",
        "11            <NA>           0       <NA>  \n",
        "12            <NA>           0       <NA>  \n",
        "13            <NA>           0       <NA>  \n",
        "14            <NA>           0       <NA>  \n",
        "15            <NA>           0       <NA>  \n",
        "16            <NA>           0       <NA>  \n",
        "17             400           0       <NA>  "
       ]
      },
      "execution_count": 32,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "bq.query(query = query).to_dataframe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "5505f13f-bb13-479d-975d-5d291564117b",
    "metadata": {},
    "source": [
     "Review the iterations from training with [ML.TRAINING_INFO](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-train):"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 33,
    "id": "00ef993b-0e37-4955-b8dc-be011fd55334",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT *\n",
       "    FROM ML.TRAINING_INFO(MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`)\n",
       "    ORDER BY iteration\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT *\n",
     "    FROM ML.TRAINING_INFO(MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`)\n",
     "    ORDER BY iteration\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 34,
    "id": "b6c3520b-c8d3-4da9-a01e-667196703db5",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>training_run</th>\n",
        "      <th>iteration</th>\n",
        "      <th>loss</th>\n",
        "      <th>eval_loss</th>\n",
        "      <th>learning_rate</th>\n",
        "      <th>duration_ms</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>0</td>\n",
        "      <td>1</td>\n",
        "      <td>54.81380</td>\n",
        "      <td>55.47980</td>\n",
        "      <td>0.3</td>\n",
        "      <td>91952</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>0</td>\n",
        "      <td>2</td>\n",
        "      <td>38.84060</td>\n",
        "      <td>39.59560</td>\n",
        "      <td>0.3</td>\n",
        "      <td>52</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>0</td>\n",
        "      <td>3</td>\n",
        "      <td>27.70890</td>\n",
        "      <td>28.48400</td>\n",
        "      <td>0.3</td>\n",
        "      <td>66</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>0</td>\n",
        "      <td>4</td>\n",
        "      <td>19.94960</td>\n",
        "      <td>20.79310</td>\n",
        "      <td>0.3</td>\n",
        "      <td>70</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>0</td>\n",
        "      <td>5</td>\n",
        "      <td>14.57790</td>\n",
        "      <td>15.48580</td>\n",
        "      <td>0.3</td>\n",
        "      <td>63</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>0</td>\n",
        "      <td>6</td>\n",
        "      <td>10.86560</td>\n",
        "      <td>11.76090</td>\n",
        "      <td>0.3</td>\n",
        "      <td>110</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>0</td>\n",
        "      <td>7</td>\n",
        "      <td>8.28401</td>\n",
        "      <td>9.38575</td>\n",
        "      <td>0.3</td>\n",
        "      <td>69</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>0</td>\n",
        "      <td>8</td>\n",
        "      <td>6.53435</td>\n",
        "      <td>7.63817</td>\n",
        "      <td>0.3</td>\n",
        "      <td>60</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>0</td>\n",
        "      <td>9</td>\n",
        "      <td>5.29837</td>\n",
        "      <td>6.39582</td>\n",
        "      <td>0.3</td>\n",
        "      <td>15670</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>0</td>\n",
        "      <td>10</td>\n",
        "      <td>4.46699</td>\n",
        "      <td>5.53626</td>\n",
        "      <td>0.3</td>\n",
        "      <td>57</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>10</th>\n",
        "      <td>0</td>\n",
        "      <td>11</td>\n",
        "      <td>3.88552</td>\n",
        "      <td>4.92203</td>\n",
        "      <td>0.3</td>\n",
        "      <td>52</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>11</th>\n",
        "      <td>0</td>\n",
        "      <td>12</td>\n",
        "      <td>3.45866</td>\n",
        "      <td>4.47603</td>\n",
        "      <td>0.3</td>\n",
        "      <td>77</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>12</th>\n",
        "      <td>0</td>\n",
        "      <td>13</td>\n",
        "      <td>3.14020</td>\n",
        "      <td>4.14678</td>\n",
        "      <td>0.3</td>\n",
        "      <td>58</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>13</th>\n",
        "      <td>0</td>\n",
        "      <td>14</td>\n",
        "      <td>2.90640</td>\n",
        "      <td>3.89820</td>\n",
        "      <td>0.3</td>\n",
        "      <td>77</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>14</th>\n",
        "      <td>0</td>\n",
        "      <td>15</td>\n",
        "      <td>2.71175</td>\n",
        "      <td>3.70595</td>\n",
        "      <td>0.3</td>\n",
        "      <td>54</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>15</th>\n",
        "      <td>0</td>\n",
        "      <td>16</td>\n",
        "      <td>2.55734</td>\n",
        "      <td>3.56235</td>\n",
        "      <td>0.3</td>\n",
        "      <td>53</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>16</th>\n",
        "      <td>0</td>\n",
        "      <td>17</td>\n",
        "      <td>2.41951</td>\n",
        "      <td>3.44475</td>\n",
        "      <td>0.3</td>\n",
        "      <td>58</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>17</th>\n",
        "      <td>0</td>\n",
        "      <td>18</td>\n",
        "      <td>2.30183</td>\n",
        "      <td>3.34597</td>\n",
        "      <td>0.3</td>\n",
        "      <td>56</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>18</th>\n",
        "      <td>0</td>\n",
        "      <td>19</td>\n",
        "      <td>2.19767</td>\n",
        "      <td>3.25959</td>\n",
        "      <td>0.3</td>\n",
        "      <td>62</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>19</th>\n",
        "      <td>0</td>\n",
        "      <td>20</td>\n",
        "      <td>2.10157</td>\n",
        "      <td>3.19111</td>\n",
        "      <td>0.3</td>\n",
        "      <td>55</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "    training_run  iteration      loss  eval_loss  learning_rate  duration_ms\n",
        "0              0          1  54.81380   55.47980            0.3        91952\n",
        "1              0          2  38.84060   39.59560            0.3           52\n",
        "2              0          3  27.70890   28.48400            0.3           66\n",
        "3              0          4  19.94960   20.79310            0.3           70\n",
        "4              0          5  14.57790   15.48580            0.3           63\n",
        "5              0          6  10.86560   11.76090            0.3          110\n",
        "6              0          7   8.28401    9.38575            0.3           69\n",
        "7              0          8   6.53435    7.63817            0.3           60\n",
        "8              0          9   5.29837    6.39582            0.3        15670\n",
        "9              0         10   4.46699    5.53626            0.3           57\n",
        "10             0         11   3.88552    4.92203            0.3           52\n",
        "11             0         12   3.45866    4.47603            0.3           77\n",
        "12             0         13   3.14020    4.14678            0.3           58\n",
        "13             0         14   2.90640    3.89820            0.3           77\n",
        "14             0         15   2.71175    3.70595            0.3           54\n",
        "15             0         16   2.55734    3.56235            0.3           53\n",
        "16             0         17   2.41951    3.44475            0.3           58\n",
        "17             0         18   2.30183    3.34597            0.3           56\n",
        "18             0         19   2.19767    3.25959            0.3           62\n",
        "19             0         20   2.10157    3.19111            0.3           55"
       ]
      },
      "execution_count": 34,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "history = bq.query(query = query).to_dataframe()\n",
     "history"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "8545fa26-6489-499c-b6f0-800687e84025",
    "metadata": {},
    "source": [
     "Plot the loss history:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 35,
    "id": "3199388c-0e5a-4bb3-8a4c-db2377f4a117",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "<AxesSubplot:xlabel='iteration'>"
       ]
      },
      "execution_count": 35,
      "metadata": {},
      "output_type": "execute_result"
     },
     {
      "data": {
       "image/png": "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\n",
       "text/plain": [
        "<Figure size 1000x500 with 1 Axes>"
       ]
      },
      "metadata": {},
      "output_type": "display_data"
     }
    ],
    "source": [
     "history.plot(x = 'iteration', y = ['loss','eval_loss'], figsize = (10, 5), grid = True)"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "dc139095-ddda-4a76-a94e-ec4078fd2564",
    "metadata": {},
    "source": [
     "### Model Metrics Review\n",
     "\n",
     "Review the model evaluation statistics with [ML.EVALUATE](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-evaluate):"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 36,
    "id": "4e0a904f-76a0-4e23-bdfe-bab341c3348e",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT * FROM ML.EVALUATE(MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`)\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT * FROM ML.EVALUATE(MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`)\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 37,
    "id": "6227af2f-5849-4a49-973a-d5a233a6c94b",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>mean_absolute_error</th>\n",
        "      <th>mean_squared_error</th>\n",
        "      <th>mean_squared_log_error</th>\n",
        "      <th>median_absolute_error</th>\n",
        "      <th>r2_score</th>\n",
        "      <th>explained_variance</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>2.242513</td>\n",
        "      <td>8.197958</td>\n",
        "      <td>0.001505</td>\n",
        "      <td>1.882072</td>\n",
        "      <td>0.916807</td>\n",
        "      <td>0.917342</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "   mean_absolute_error  mean_squared_error  mean_squared_log_error  \\\n",
        "0             2.242513            8.197958                0.001505   \n",
        "\n",
        "   median_absolute_error  r2_score  explained_variance  \n",
        "0               1.882072  0.916807            0.917342  "
       ]
      },
      "execution_count": 37,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "bq.query(query = query).to_dataframe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "490ec674-524e-4547-8d67-e468647752b0",
    "metadata": {},
    "source": [
     "### Feature Importance\n",
     "\n",
     "Retrieve the [XGBoost feature importances](https://xgboost.readthedocs.io/en/latest/python/python_api.html#xgboost.XGBRegressor.feature_importances_) for the model using [ML.FEATURE_IMPORTANCE](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-importance).\n",
     "- importance_weight: the number of splits the feature was used in across all trees\n",
     "- importance_gain: the improvement in accuracy due to splits using the feature\n",
     "- importance_cover: the number of rows of data covered by splits using the feature"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 38,
    "id": "6ee976b5-6bc1-4c82-818c-9e34d62790d5",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT *\n",
       "    FROM ML.FEATURE_IMPORTANCE (MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`)\n",
       "    ORDER BY importance_gain DESC\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT *\n",
     "    FROM ML.FEATURE_IMPORTANCE (MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`)\n",
     "    ORDER BY importance_gain DESC\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 39,
    "id": "153489af-58dc-48b7-a7f9-9fce646df468",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>feature</th>\n",
        "      <th>importance_weight</th>\n",
        "      <th>importance_gain</th>\n",
        "      <th>importance_cover</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>scale_bakeTemp</td>\n",
        "      <td>3077</td>\n",
        "      <td>1723.055303</td>\n",
        "      <td>648.608060</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>scale_water2Amt</td>\n",
        "      <td>2763</td>\n",
        "      <td>450.440461</td>\n",
        "      <td>489.240680</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>year</td>\n",
        "      <td>1170</td>\n",
        "      <td>312.603170</td>\n",
        "      <td>618.773504</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>scale_mix2Time</td>\n",
        "      <td>565</td>\n",
        "      <td>295.250026</td>\n",
        "      <td>715.058407</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>scale_bakeTime</td>\n",
        "      <td>1687</td>\n",
        "      <td>292.783790</td>\n",
        "      <td>437.933610</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>scale_yeastAmt</td>\n",
        "      <td>1101</td>\n",
        "      <td>254.193778</td>\n",
        "      <td>636.986376</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>scale_mix1Time</td>\n",
        "      <td>1114</td>\n",
        "      <td>189.146947</td>\n",
        "      <td>538.284560</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>scale_flourAmt</td>\n",
        "      <td>2393</td>\n",
        "      <td>154.002105</td>\n",
        "      <td>338.239866</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>scale_mix2Speed</td>\n",
        "      <td>373</td>\n",
        "      <td>134.481077</td>\n",
        "      <td>554.182306</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>scale_restTime</td>\n",
        "      <td>717</td>\n",
        "      <td>85.717913</td>\n",
        "      <td>397.451883</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>10</th>\n",
        "      <td>scale_saltAmt</td>\n",
        "      <td>541</td>\n",
        "      <td>75.637670</td>\n",
        "      <td>374.133087</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>11</th>\n",
        "      <td>scale_water1Amt</td>\n",
        "      <td>1030</td>\n",
        "      <td>43.398274</td>\n",
        "      <td>168.664078</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>12</th>\n",
        "      <td>contestant</td>\n",
        "      <td>1092</td>\n",
        "      <td>20.246916</td>\n",
        "      <td>78.205128</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>13</th>\n",
        "      <td>scale_proveTime</td>\n",
        "      <td>718</td>\n",
        "      <td>19.166594</td>\n",
        "      <td>80.030641</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>14</th>\n",
        "      <td>scale_ambTemp</td>\n",
        "      <td>824</td>\n",
        "      <td>17.944245</td>\n",
        "      <td>76.337379</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>15</th>\n",
        "      <td>scale_ambHumidity</td>\n",
        "      <td>735</td>\n",
        "      <td>15.327752</td>\n",
        "      <td>83.742857</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>16</th>\n",
        "      <td>week</td>\n",
        "      <td>675</td>\n",
        "      <td>14.019638</td>\n",
        "      <td>73.524444</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>17</th>\n",
        "      <td>scale_waterTemp</td>\n",
        "      <td>353</td>\n",
        "      <td>13.845814</td>\n",
        "      <td>54.195467</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>18</th>\n",
        "      <td>scale_mix1Speed</td>\n",
        "      <td>174</td>\n",
        "      <td>12.294526</td>\n",
        "      <td>63.586207</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "              feature  importance_weight  importance_gain  importance_cover\n",
        "0      scale_bakeTemp               3077      1723.055303        648.608060\n",
        "1     scale_water2Amt               2763       450.440461        489.240680\n",
        "2                year               1170       312.603170        618.773504\n",
        "3      scale_mix2Time                565       295.250026        715.058407\n",
        "4      scale_bakeTime               1687       292.783790        437.933610\n",
        "5      scale_yeastAmt               1101       254.193778        636.986376\n",
        "6      scale_mix1Time               1114       189.146947        538.284560\n",
        "7      scale_flourAmt               2393       154.002105        338.239866\n",
        "8     scale_mix2Speed                373       134.481077        554.182306\n",
        "9      scale_restTime                717        85.717913        397.451883\n",
        "10      scale_saltAmt                541        75.637670        374.133087\n",
        "11    scale_water1Amt               1030        43.398274        168.664078\n",
        "12         contestant               1092        20.246916         78.205128\n",
        "13    scale_proveTime                718        19.166594         80.030641\n",
        "14      scale_ambTemp                824        17.944245         76.337379\n",
        "15  scale_ambHumidity                735        15.327752         83.742857\n",
        "16               week                675        14.019638         73.524444\n",
        "17    scale_waterTemp                353        13.845814         54.195467\n",
        "18    scale_mix1Speed                174        12.294526         63.586207"
       ]
      },
      "execution_count": 39,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "feature_importance = bq.query(query = query).to_dataframe()\n",
     "feature_importance"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "884ebbbf-1847-4aef-8d84-e10dd63aa0b5",
    "metadata": {},
    "source": [
     "### Global Explainability - What is important to Judges?\n",
     "\n",
     "Retrieve the influence of each feature on the model with [ML.GLOBAL_EXPLAIN](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-global-explain).  "
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 40,
    "id": "6068d2c1-5323-4870-a925-c7413627485c",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT *\n",
       "    FROM ML.GLOBAL_EXPLAIN (MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`)\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT *\n",
     "    FROM ML.GLOBAL_EXPLAIN (MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`)\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 41,
    "id": "7ab6413a-7069-457e-b043-9d897d5f35d6",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>feature</th>\n",
        "      <th>attribution</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>scale_bakeTemp</td>\n",
        "      <td>6.024916</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>scale_water2Amt</td>\n",
        "      <td>2.544323</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>year</td>\n",
        "      <td>1.244881</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>scale_bakeTime</td>\n",
        "      <td>1.025809</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>scale_flourAmt</td>\n",
        "      <td>0.973105</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>scale_mix2Time</td>\n",
        "      <td>0.769810</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>scale_yeastAmt</td>\n",
        "      <td>0.755667</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>scale_mix1Time</td>\n",
        "      <td>0.597188</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>scale_mix2Speed</td>\n",
        "      <td>0.268663</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>scale_restTime</td>\n",
        "      <td>0.230816</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>10</th>\n",
        "      <td>scale_saltAmt</td>\n",
        "      <td>0.219477</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>11</th>\n",
        "      <td>scale_water1Amt</td>\n",
        "      <td>0.121160</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>12</th>\n",
        "      <td>contestant</td>\n",
        "      <td>0.035856</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>13</th>\n",
        "      <td>week</td>\n",
        "      <td>0.027385</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>14</th>\n",
        "      <td>scale_ambTemp</td>\n",
        "      <td>0.020811</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>15</th>\n",
        "      <td>scale_ambHumidity</td>\n",
        "      <td>0.013113</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>16</th>\n",
        "      <td>scale_proveTime</td>\n",
        "      <td>0.010081</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>17</th>\n",
        "      <td>scale_mix1Speed</td>\n",
        "      <td>0.007083</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>18</th>\n",
        "      <td>scale_waterTemp</td>\n",
        "      <td>0.005962</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "              feature  attribution\n",
        "0      scale_bakeTemp     6.024916\n",
        "1     scale_water2Amt     2.544323\n",
        "2                year     1.244881\n",
        "3      scale_bakeTime     1.025809\n",
        "4      scale_flourAmt     0.973105\n",
        "5      scale_mix2Time     0.769810\n",
        "6      scale_yeastAmt     0.755667\n",
        "7      scale_mix1Time     0.597188\n",
        "8     scale_mix2Speed     0.268663\n",
        "9      scale_restTime     0.230816\n",
        "10      scale_saltAmt     0.219477\n",
        "11    scale_water1Amt     0.121160\n",
        "12         contestant     0.035856\n",
        "13               week     0.027385\n",
        "14      scale_ambTemp     0.020811\n",
        "15  scale_ambHumidity     0.013113\n",
        "16    scale_proveTime     0.010081\n",
        "17    scale_mix1Speed     0.007083\n",
        "18    scale_waterTemp     0.005962"
       ]
      },
      "execution_count": 41,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "gexplain = bq.query(query = query).to_dataframe()\n",
     "gexplain"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "ee3a401d-1484-441c-8134-9266a51e8dc9",
    "metadata": {},
    "source": [
     "Plot the attribution:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 42,
    "id": "29389f1a-4926-4be6-9d36-d3794afb3b4e",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "image/png": "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\n",
       "text/plain": [
        "<Figure size 1500x800 with 1 Axes>"
       ]
      },
      "metadata": {},
      "output_type": "display_data"
     }
    ],
    "source": [
     "plt.figure(figsize=(15,8))\n",
     "plt.barh(gexplain['feature'], gexplain['attribution'])\n",
     "plt.title('Global Influence')\n",
     "plt.xlabel('Attribution')\n",
     "#plt.ylabel('Feature')\n",
     "plt.show()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "13bc9512-723b-4807-82dc-515b0c6f0569",
    "metadata": {},
    "source": [
     "### Predictions With BigQuery ML (BQML)\n",
     "\n",
     "Create a pandas dataframe with retrieved predictions for the test data in the table using [ML.PREDICT](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-predict):"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 43,
    "id": "025797c5-7a21-49f9-b536-2e08fa7f434a",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT Recipe, JUDGE_A, predicted_JUDGE_A, JUDGE_A-predicted_JUDGE_A as Error\n",
       "    FROM ML.PREDICT (MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`,(\n",
       "        SELECT *\n",
       "        FROM `statmike-mlops-349915.feature_engineering.bread`\n",
       "        WHERE Recipe <= 10)\n",
       "    )\n",
       "    ORDER BY Recipe\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT Recipe, JUDGE_A, predicted_JUDGE_A, JUDGE_A-predicted_JUDGE_A as Error\n",
     "    FROM ML.PREDICT (MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`,(\n",
     "        SELECT *\n",
     "        FROM `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}`\n",
     "        WHERE Recipe <= 10)\n",
     "    )\n",
     "    ORDER BY Recipe\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 44,
    "id": "d290ac39-0cee-4fd2-b2c3-14c2cf8cde22",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>Recipe</th>\n",
        "      <th>JUDGE_A</th>\n",
        "      <th>predicted_JUDGE_A</th>\n",
        "      <th>Error</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>1</td>\n",
        "      <td>75.0</td>\n",
        "      <td>73.526794</td>\n",
        "      <td>1.473206</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>2</td>\n",
        "      <td>86.0</td>\n",
        "      <td>85.937080</td>\n",
        "      <td>0.062920</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>3</td>\n",
        "      <td>87.0</td>\n",
        "      <td>86.481888</td>\n",
        "      <td>0.518112</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>4</td>\n",
        "      <td>81.0</td>\n",
        "      <td>82.495926</td>\n",
        "      <td>-1.495926</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>5</td>\n",
        "      <td>83.0</td>\n",
        "      <td>82.583466</td>\n",
        "      <td>0.416534</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>6</td>\n",
        "      <td>76.0</td>\n",
        "      <td>78.951942</td>\n",
        "      <td>-2.951942</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>7</td>\n",
        "      <td>79.0</td>\n",
        "      <td>79.647018</td>\n",
        "      <td>-0.647018</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>8</td>\n",
        "      <td>84.0</td>\n",
        "      <td>84.974556</td>\n",
        "      <td>-0.974556</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>9</td>\n",
        "      <td>77.0</td>\n",
        "      <td>82.447281</td>\n",
        "      <td>-5.447281</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>10</td>\n",
        "      <td>84.0</td>\n",
        "      <td>82.935287</td>\n",
        "      <td>1.064713</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "   Recipe  JUDGE_A  predicted_JUDGE_A     Error\n",
        "0       1     75.0          73.526794  1.473206\n",
        "1       2     86.0          85.937080  0.062920\n",
        "2       3     87.0          86.481888  0.518112\n",
        "3       4     81.0          82.495926 -1.495926\n",
        "4       5     83.0          82.583466  0.416534\n",
        "5       6     76.0          78.951942 -2.951942\n",
        "6       7     79.0          79.647018 -0.647018\n",
        "7       8     84.0          84.974556 -0.974556\n",
        "8       9     77.0          82.447281 -5.447281\n",
        "9      10     84.0          82.935287  1.064713"
       ]
      },
      "execution_count": 44,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "bq.query(query = query).to_dataframe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "869cce8d-51ca-4ee7-8c34-f6b28944f9d6",
    "metadata": {},
    "source": [
     "### Local Explainability - What impacted the judges rating?\n",
     "\n",
     "Retrieve feature attribution values to interpret each features contribution to a prediction relative to the baseline prediction using [ML.EXPLAIN_PREDICT](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-explain-predict)."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 45,
    "id": "91fe8acb-d8b3-4e37-bad8-c0eba2173ad9",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    SELECT *\n",
       "    FROM ML.EXPLAIN_PREDICT (MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`,(\n",
       "        SELECT *\n",
       "        FROM `statmike-mlops-349915.feature_engineering.bread`\n",
       "        WHERE Recipe <= 10)\n",
       "    )\n",
       "    ORDER BY Recipe\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT *\n",
     "    FROM ML.EXPLAIN_PREDICT (MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`,(\n",
     "        SELECT *\n",
     "        FROM `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}`\n",
     "        WHERE Recipe <= 10)\n",
     "    )\n",
     "    ORDER BY Recipe\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 46,
    "id": "e423d524-f912-443f-b0ad-26c96fb8ec46",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>predicted_JUDGE_A</th>\n",
        "      <th>top_feature_attributions</th>\n",
        "      <th>baseline_prediction_value</th>\n",
        "      <th>prediction_value</th>\n",
        "      <th>approximation_error</th>\n",
        "      <th>recipe</th>\n",
        "      <th>season</th>\n",
        "      <th>round</th>\n",
        "      <th>contestant_id</th>\n",
        "      <th>airdate</th>\n",
        "      <th>...</th>\n",
        "      <th>water2Amt</th>\n",
        "      <th>waterTemp</th>\n",
        "      <th>proveTime</th>\n",
        "      <th>restTime</th>\n",
        "      <th>bakeTime</th>\n",
        "      <th>bakeTemp</th>\n",
        "      <th>ambTemp</th>\n",
        "      <th>ambHumidity</th>\n",
        "      <th>JUDGE_A</th>\n",
        "      <th>JUDGE_B</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>73.526794</td>\n",
        "      <td>[{'feature': 'year', 'attribution': -2.2365446...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>73.526747</td>\n",
        "      <td>0.0</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_1</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>98.612837</td>\n",
        "      <td>46</td>\n",
        "      <td>105.673044</td>\n",
        "      <td>44</td>\n",
        "      <td>28</td>\n",
        "      <td>435.393493</td>\n",
        "      <td>51.279961</td>\n",
        "      <td>61.443331</td>\n",
        "      <td>75.0</td>\n",
        "      <td>63.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>85.937080</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>85.936985</td>\n",
        "      <td>0.0</td>\n",
        "      <td>2</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_2</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>111.708531</td>\n",
        "      <td>44</td>\n",
        "      <td>93.831810</td>\n",
        "      <td>42</td>\n",
        "      <td>21</td>\n",
        "      <td>423.411066</td>\n",
        "      <td>75.692056</td>\n",
        "      <td>35.545136</td>\n",
        "      <td>86.0</td>\n",
        "      <td>78.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>86.481888</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>86.481768</td>\n",
        "      <td>0.0</td>\n",
        "      <td>3</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_3</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>101.911841</td>\n",
        "      <td>48</td>\n",
        "      <td>104.969749</td>\n",
        "      <td>40</td>\n",
        "      <td>20</td>\n",
        "      <td>426.042110</td>\n",
        "      <td>59.630283</td>\n",
        "      <td>63.139995</td>\n",
        "      <td>87.0</td>\n",
        "      <td>80.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>82.495926</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>82.495880</td>\n",
        "      <td>0.0</td>\n",
        "      <td>4</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_4</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>131.296973</td>\n",
        "      <td>45</td>\n",
        "      <td>100.003988</td>\n",
        "      <td>39</td>\n",
        "      <td>28</td>\n",
        "      <td>425.303916</td>\n",
        "      <td>63.791167</td>\n",
        "      <td>69.601895</td>\n",
        "      <td>81.0</td>\n",
        "      <td>72.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>82.583466</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>82.583417</td>\n",
        "      <td>0.0</td>\n",
        "      <td>5</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_5</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>100.470180</td>\n",
        "      <td>45</td>\n",
        "      <td>96.943759</td>\n",
        "      <td>38</td>\n",
        "      <td>25</td>\n",
        "      <td>433.129239</td>\n",
        "      <td>57.452172</td>\n",
        "      <td>56.456663</td>\n",
        "      <td>83.0</td>\n",
        "      <td>75.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>78.951942</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>78.951858</td>\n",
        "      <td>0.0</td>\n",
        "      <td>6</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_6</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>130.229546</td>\n",
        "      <td>41</td>\n",
        "      <td>102.368168</td>\n",
        "      <td>41</td>\n",
        "      <td>21</td>\n",
        "      <td>428.770877</td>\n",
        "      <td>62.391203</td>\n",
        "      <td>66.493702</td>\n",
        "      <td>76.0</td>\n",
        "      <td>65.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>79.647018</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>79.647017</td>\n",
        "      <td>0.0</td>\n",
        "      <td>7</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_7</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>85.706252</td>\n",
        "      <td>50</td>\n",
        "      <td>106.748456</td>\n",
        "      <td>42</td>\n",
        "      <td>24</td>\n",
        "      <td>429.830463</td>\n",
        "      <td>62.704889</td>\n",
        "      <td>50.332733</td>\n",
        "      <td>79.0</td>\n",
        "      <td>69.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>84.974556</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>84.974490</td>\n",
        "      <td>0.0</td>\n",
        "      <td>8</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_8</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>99.990115</td>\n",
        "      <td>49</td>\n",
        "      <td>105.191873</td>\n",
        "      <td>35</td>\n",
        "      <td>25</td>\n",
        "      <td>423.358189</td>\n",
        "      <td>67.455814</td>\n",
        "      <td>47.565266</td>\n",
        "      <td>84.0</td>\n",
        "      <td>75.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>82.447281</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>82.447260</td>\n",
        "      <td>0.0</td>\n",
        "      <td>9</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_9</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>101.542559</td>\n",
        "      <td>46</td>\n",
        "      <td>94.640184</td>\n",
        "      <td>42</td>\n",
        "      <td>25</td>\n",
        "      <td>428.768843</td>\n",
        "      <td>66.431606</td>\n",
        "      <td>32.277536</td>\n",
        "      <td>77.0</td>\n",
        "      <td>66.0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>82.935287</td>\n",
        "      <td>[{'feature': 'scale_bakeTemp', 'attribution': ...</td>\n",
        "      <td>77.436806</td>\n",
        "      <td>82.935196</td>\n",
        "      <td>0.0</td>\n",
        "      <td>10</td>\n",
        "      <td>1</td>\n",
        "      <td>1</td>\n",
        "      <td>c_10</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>...</td>\n",
        "      <td>94.687304</td>\n",
        "      <td>46</td>\n",
        "      <td>96.571240</td>\n",
        "      <td>39</td>\n",
        "      <td>25</td>\n",
        "      <td>424.924214</td>\n",
        "      <td>43.868904</td>\n",
        "      <td>58.789972</td>\n",
        "      <td>84.0</td>\n",
        "      <td>76.0</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "<p>10 rows × 28 columns</p>\n",
        "</div>"
       ],
       "text/plain": [
        "   predicted_JUDGE_A                           top_feature_attributions  \\\n",
        "0          73.526794  [{'feature': 'year', 'attribution': -2.2365446...   \n",
        "1          85.937080  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "2          86.481888  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "3          82.495926  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "4          82.583466  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "5          78.951942  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "6          79.647018  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "7          84.974556  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "8          82.447281  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "9          82.935287  [{'feature': 'scale_bakeTemp', 'attribution': ...   \n",
        "\n",
        "   baseline_prediction_value  prediction_value  approximation_error  recipe  \\\n",
        "0                  77.436806         73.526747                  0.0       1   \n",
        "1                  77.436806         85.936985                  0.0       2   \n",
        "2                  77.436806         86.481768                  0.0       3   \n",
        "3                  77.436806         82.495880                  0.0       4   \n",
        "4                  77.436806         82.583417                  0.0       5   \n",
        "5                  77.436806         78.951858                  0.0       6   \n",
        "6                  77.436806         79.647017                  0.0       7   \n",
        "7                  77.436806         84.974490                  0.0       8   \n",
        "8                  77.436806         82.447260                  0.0       9   \n",
        "9                  77.436806         82.935196                  0.0      10   \n",
        "\n",
        "   season  round contestant_id     airdate  ...   water2Amt  waterTemp  \\\n",
        "0       1      1           c_1  2003-05-26  ...   98.612837         46   \n",
        "1       1      1           c_2  2003-05-26  ...  111.708531         44   \n",
        "2       1      1           c_3  2003-05-26  ...  101.911841         48   \n",
        "3       1      1           c_4  2003-05-26  ...  131.296973         45   \n",
        "4       1      1           c_5  2003-05-26  ...  100.470180         45   \n",
        "5       1      1           c_6  2003-05-26  ...  130.229546         41   \n",
        "6       1      1           c_7  2003-05-26  ...   85.706252         50   \n",
        "7       1      1           c_8  2003-05-26  ...   99.990115         49   \n",
        "8       1      1           c_9  2003-05-26  ...  101.542559         46   \n",
        "9       1      1          c_10  2003-05-26  ...   94.687304         46   \n",
        "\n",
        "    proveTime  restTime  bakeTime    bakeTemp    ambTemp  ambHumidity  \\\n",
        "0  105.673044        44        28  435.393493  51.279961    61.443331   \n",
        "1   93.831810        42        21  423.411066  75.692056    35.545136   \n",
        "2  104.969749        40        20  426.042110  59.630283    63.139995   \n",
        "3  100.003988        39        28  425.303916  63.791167    69.601895   \n",
        "4   96.943759        38        25  433.129239  57.452172    56.456663   \n",
        "5  102.368168        41        21  428.770877  62.391203    66.493702   \n",
        "6  106.748456        42        24  429.830463  62.704889    50.332733   \n",
        "7  105.191873        35        25  423.358189  67.455814    47.565266   \n",
        "8   94.640184        42        25  428.768843  66.431606    32.277536   \n",
        "9   96.571240        39        25  424.924214  43.868904    58.789972   \n",
        "\n",
        "   JUDGE_A  JUDGE_B  \n",
        "0     75.0     63.0  \n",
        "1     86.0     78.0  \n",
        "2     87.0     80.0  \n",
        "3     81.0     72.0  \n",
        "4     83.0     75.0  \n",
        "5     76.0     65.0  \n",
        "6     79.0     69.0  \n",
        "7     84.0     75.0  \n",
        "8     77.0     66.0  \n",
        "9     84.0     76.0  \n",
        "\n",
        "[10 rows x 28 columns]"
       ]
      },
      "execution_count": 46,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "explain = bq.query(query = query).to_dataframe()\n",
     "explain"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "dad68c50-4f3f-492a-a06a-0f48605ba448",
    "metadata": {},
    "source": [
     "Create a plot of feature attributions for a single prediction:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 47,
    "id": "bba5463c-8a94-4baa-9d88-c2fe5419c383",
    "metadata": {},
    "outputs": [],
    "source": [
     "recipe = 1\n",
     "explanation = explain.iloc[recipe-1]['top_feature_attributions']"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 48,
    "id": "46a76ebc-cdad-4db0-9e14-ea25b2de56d9",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "image/png": "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\n",
       "text/plain": [
        "<Figure size 1500x500 with 1 Axes>"
       ]
      },
      "metadata": {},
      "output_type": "display_data"
     }
    ],
    "source": [
     "features = []\n",
     "scores = []\n",
     "for k in explanation:\n",
     "    features.append(k['feature'])\n",
     "    scores.append(k['attribution'])\n",
     "features = [x for _, x in sorted(zip(scores, features))]\n",
     "scores = sorted(scores)\n",
     "\n",
     "plt.figure(figsize=(15,5))\n",
     "plt.barh(features, scores)\n",
     "plt.title(f'Feature Attribution for Prediction - Recipe {recipe}')\n",
     "plt.show()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "010d39e1-60cd-4275-9fd5-90978aecbc3a",
    "metadata": {},
    "source": [
     "### Check out this model in the BigQuery Console:\n",
     "- Make sure project selected is the one from this notebook\n",
     "- Under Explore, expand this project and dataset\n",
     "- Expand Models and select the model create here"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 49,
    "id": "e0855079-2994-4f92-852e-c10de2b8bd9b",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Direct link to the model in BigQuery:\n",
       "https://console.cloud.google.com/bigquery?project=statmike-mlops-349915&ws=!1m5!1m4!5m3!1sstatmike-mlops-349915!2sfeature_engineering!3sbqml_feature_engineering_transform\n"
      ]
     }
    ],
    "source": [
     "print(f'Direct link to the model in BigQuery:\\nhttps://console.cloud.google.com/bigquery?project={PROJECT_ID}&ws=!1m5!1m4!5m3!1s{PROJECT_ID}!2s{BQ_DATASET}!3s{BQ_MODEL}')"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "a545999c-1bee-4d5c-ab43-c4ca63be0d7d",
    "metadata": {},
    "source": [
     "### Check out this model in the Vertex AI Console:\n",
     "- Make sure project selected is the one from this notebook\n",
     "- Make sure the region is `europe-west4` (default when BQ is `EU` multiregion)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 50,
    "id": "a3801c09-f8e4-4dda-871a-2289cbf428d6",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "View the model in the Vertex AI Console:\n",
       "https://console.cloud.google.com/vertex-ai/locations/us-central1/models/bqml_bqml_feature_engineering_transform?project=statmike-mlops-349915\n"
      ]
     }
    ],
    "source": [
     "print(f\"View the model in the Vertex AI Console:\\nhttps://console.cloud.google.com/vertex-ai/locations/{REGION}/models/bqml_{BQ_MODEL}?project={PROJECT_ID}\")"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "5d28ab3a-e919-4847-96b3-3361ae4255d9",
    "metadata": {},
    "source": [
     "---\n",
     "## Step 2: Serving\n",
     "\n",
     "**Notes on Serving With Manual Feature Engineering**\n",
     "\n",
     "When this model was trained it also registered to Vertex AI Model Registry.  With the inline `TRANSFORM` clause the resulting model is also directly servable even though it is actually 2 models, a preprocessing model that represents the `TRANSFORM` clause, and a model that serves responses from the inputs shaped by the preprocessing.  When deployed the models and pipelined together on the same endpoint.  As long as the `TRANSFORM` statement used [supported SQL functions](https://cloud.google.com/bigquery/docs/exporting-models#export-transform-functions) this will work."
    ]
   },
   {
    "cell_type": "markdown",
    "id": "97ff12fa-70c4-4f29-acf3-95a3b35990bb",
    "metadata": {},
    "source": [
     "### Vertex AI Model Registry - Get Models\n",
     "\n",
     "When these modes were trained in BigQuery ML they included options for registration in Vertex AI Model Registry.  This section shows how to retrieve the information from the model unsing Vetex AI.\n",
     "\n",
     "References:\n",
     "- [aiplatform.Model()](https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.Model)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 53,
    "id": "cb1a70f4-9dfa-4a6e-81ca-521840a28107",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'projects/1026793852137/locations/us-central1/models/bqml_bqml_feature_engineering_transform'"
       ]
      },
      "execution_count": 53,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "vertex_model = aiplatform.Model(model_name = f'bqml_{BQ_MODEL}')\n",
     "vertex_model.resource_name"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 54,
    "id": "61e981d2-dc91-462f-9ec6-e3085070dd91",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "{'name': 'projects/1026793852137/locations/us-central1/models/bqml_bqml_feature_engineering_transform@1/evaluations/5578997075363758080',\n",
        " 'metricsSchemaUri': 'gs://google-cloud-aiplatform/schema/modelevaluation/regression_metrics_1.0.0.yaml',\n",
        " 'metrics': {'meanAbsoluteError': 2.2425125,\n",
        "  'rootMeanSquaredError': 2.8632076,\n",
        "  'rSquared': 0.9168072,\n",
        "  'rootMeanSquaredLogError': 0.03879608},\n",
        " 'createTime': '2023-07-05T12:05:51.543719Z'}"
       ]
      },
      "execution_count": 54,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "aiplatform.ModelEvaluation(\n",
     "    evaluation_name = vertex_model.get_model_evaluation().resource_name\n",
     ").to_dict()"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 56,
    "id": "df736369-1aea-4faf-abb3-64aca3432b40",
    "metadata": {},
    "outputs": [],
    "source": [
     "# note: this is blank since the model is in two parts which are managed in a single container at deployment, see export below:\n",
     "vertex_model.uri "
    ]
   },
   {
    "cell_type": "markdown",
    "id": "1a9c2875-c7d7-4bcc-b5ee-e0d1064c12ce",
    "metadata": {},
    "source": [
     "### Vertex AI Endpoint - Create or Retrieve\n",
     "\n",
     "References:\n",
     "- [aiplatform.Endpoint()](https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.Endpoint)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 57,
    "id": "611804b9-d9cb-48db-a17c-f6351e6de40e",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Creating Endpoint\n",
       "Create Endpoint backing LRO: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944/operations/7776347870650695680\n",
       "Endpoint created. Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n",
       "To use this Endpoint in another session:\n",
       "endpoint = aiplatform.Endpoint('projects/1026793852137/locations/us-central1/endpoints/6287935475437010944')\n",
       "Endpoint Created: bqml_feature_engineering\n"
      ]
     }
    ],
    "source": [
     "if not aiplatform.Endpoint.list(filter = f\"display_name={SERIES}_{EXPERIMENT}\"):\n",
     "    endpoint = aiplatform.Endpoint.create(\n",
     "        display_name = f\"{SERIES}_{EXPERIMENT}\",\n",
     "        labels = {'series' : f\"{SERIES}\"}\n",
     "    )\n",
     "    print(f\"Endpoint Created: {endpoint.display_name}\")\n",
     "else: endpoint = aiplatform.Endpoint.list(filter = f\"display_name={SERIES}_{EXPERIMENT}_transform\")[0]"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 58,
    "id": "a9dd6318-b1e5-48f3-a3ba-8961133a0c93",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "'bqml_feature_engineering'"
       ]
      },
      "execution_count": 58,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "endpoint.display_name"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 59,
    "id": "4ee05c26-5f23-44bf-8420-95b4980ea361",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "{}"
       ]
      },
      "execution_count": 59,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "endpoint.traffic_split"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 60,
    "id": "d083c647-7352-43c3-ba9a-aae68bffe7fc",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "[]"
       ]
      },
      "execution_count": 60,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "deployed_models = endpoint.list_models()\n",
     "deployed_models"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "1f8019dc-a2a6-422d-8c5a-cb811b1d9bb7",
    "metadata": {},
    "source": [
     "### Deploy Model To Vertex AI Endpoint\n",
     "\n",
     "**NOTE: This can take 15+ minutes to complete.**\n",
     "\n",
     "Reference:\n",
     "- [aiplatform.Endpoint.deploy()](https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.Endpoint#google_cloud_aiplatform_Endpoint_deploy)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 61,
    "id": "a85770f5-980a-449e-aaa1-cdea65361208",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Deploying Model projects/1026793852137/locations/us-central1/models/bqml_bqml_feature_engineering_transform to Endpoint : projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n",
       "Deploy Endpoint model backing LRO: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944/operations/5776749636098195456\n",
       "Endpoint model deployed. Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     }
    ],
    "source": [
     "endpoint.deploy(\n",
     "    model = vertex_model,\n",
     "    deployed_model_display_name = vertex_model.display_name,\n",
     "    traffic_percentage = 100,\n",
     "    machine_type = 'n1-standard-2',\n",
     "    min_replica_count = 1,\n",
     "    max_replica_count = 1\n",
     ")"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 62,
    "id": "abf3aa99-9ce6-4ff1-94fa-3c6473ce3e5d",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "[id: \"7768764332795494400\"\n",
        " model: \"projects/1026793852137/locations/us-central1/models/bqml_bqml_feature_engineering_transform\"\n",
        " display_name: \"bqml_bqml_feature_engineering_transform\"\n",
        " create_time {\n",
        "   seconds: 1688559727\n",
        "   nanos: 579630000\n",
        " }\n",
        " dedicated_resources {\n",
        "   machine_spec {\n",
        "     machine_type: \"n1-standard-2\"\n",
        "   }\n",
        "   min_replica_count: 1\n",
        "   max_replica_count: 1\n",
        " }\n",
        " explanation_spec {\n",
        "   parameters {\n",
        "     sampled_shapley_attribution {\n",
        "       path_count: 5\n",
        "     }\n",
        "   }\n",
        "   metadata {\n",
        "     inputs {\n",
        "       key: \"airdate\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"ambHumidity\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"ambTemp\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"bakeTemp\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"bakeTime\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"contestant_id\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"flourAmt\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"mix1Speed\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"mix1Time\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"mix2Speed\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"mix2Time\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"proveTime\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"restTime\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"saltAmt\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"water1Amt\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"water2Amt\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"waterTemp\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     inputs {\n",
        "       key: \"yeastAmt\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "     outputs {\n",
        "       key: \"predictions\"\n",
        "       value {\n",
        "       }\n",
        "     }\n",
        "   }\n",
        " }\n",
        " disable_container_logging: true\n",
        " model_version_id: \"1\"]"
       ]
      },
      "execution_count": 62,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "endpoint.list_models()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "1f55bf78-c880-40e9-9dc0-76d00a2aa356",
    "metadata": {},
    "source": [
     "### Retrieve Records For Prediction\n",
     "The first 10 recipes.\n",
     "\n",
     "**Notes:**\n",
     "- Exclude columns not used in the model.  While BigQuery ML.PREDICT can handle parsing these out, the model hosted on Vertex AI Prediction Endpoints has an expected input signature of just the features used in the model.\n",
     "- The feature `airdate` is a date formated feature.  In the model this is stored as a string and needs to be converted for Vertex AI Prediction.\n",
     "- To check the expected features and format of features you can also export the model and used the TensorFlow saved model CLI to review the signature.  This is demonstrated in the `Model Exports` section below."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 136,
    "id": "ea503e1e-87a9-4982-9dc6-be194e34ec13",
    "metadata": {},
    "outputs": [],
    "source": [
     "n = 10\n",
     "pred = bq.query(\n",
     "    query = f\"\"\"\n",
     "        SELECT * EXCEPT(JUDGE_A, JUDGE_B, season, round, recipe) # exclude columns not used in model\n",
     "        FROM {BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}\n",
     "        WHERE Recipe <= 10\n",
     "        ORDER BY Recipe\n",
     "    \"\"\"\n",
     ").to_dataframe()\n",
     "pred['airdate'] = pred['airdate'].astype(str)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 137,
    "id": "3979eec5-fedf-4d17-bd62-152b1bbfcb93",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>contestant_id</th>\n",
        "      <th>airdate</th>\n",
        "      <th>flourAmt</th>\n",
        "      <th>saltAmt</th>\n",
        "      <th>yeastAmt</th>\n",
        "      <th>mix1Time</th>\n",
        "      <th>mix1Speed</th>\n",
        "      <th>mix2Time</th>\n",
        "      <th>mix2Speed</th>\n",
        "      <th>water1Amt</th>\n",
        "      <th>water2Amt</th>\n",
        "      <th>waterTemp</th>\n",
        "      <th>proveTime</th>\n",
        "      <th>restTime</th>\n",
        "      <th>bakeTime</th>\n",
        "      <th>bakeTemp</th>\n",
        "      <th>ambTemp</th>\n",
        "      <th>ambHumidity</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>c_1</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>484.289865</td>\n",
        "      <td>9</td>\n",
        "      <td>10</td>\n",
        "      <td>5</td>\n",
        "      <td>3</td>\n",
        "      <td>5</td>\n",
        "      <td>5</td>\n",
        "      <td>311.663494</td>\n",
        "      <td>98.612837</td>\n",
        "      <td>46</td>\n",
        "      <td>105.673044</td>\n",
        "      <td>44</td>\n",
        "      <td>28</td>\n",
        "      <td>435.393493</td>\n",
        "      <td>51.279961</td>\n",
        "      <td>61.443331</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>c_2</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>501.259590</td>\n",
        "      <td>9</td>\n",
        "      <td>10</td>\n",
        "      <td>3</td>\n",
        "      <td>4</td>\n",
        "      <td>5</td>\n",
        "      <td>4</td>\n",
        "      <td>315.472473</td>\n",
        "      <td>111.708531</td>\n",
        "      <td>44</td>\n",
        "      <td>93.831810</td>\n",
        "      <td>42</td>\n",
        "      <td>21</td>\n",
        "      <td>423.411066</td>\n",
        "      <td>75.692056</td>\n",
        "      <td>35.545136</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>c_3</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>495.194450</td>\n",
        "      <td>8</td>\n",
        "      <td>11</td>\n",
        "      <td>5</td>\n",
        "      <td>4</td>\n",
        "      <td>8</td>\n",
        "      <td>4</td>\n",
        "      <td>324.477611</td>\n",
        "      <td>101.911841</td>\n",
        "      <td>48</td>\n",
        "      <td>104.969749</td>\n",
        "      <td>40</td>\n",
        "      <td>20</td>\n",
        "      <td>426.042110</td>\n",
        "      <td>59.630283</td>\n",
        "      <td>63.139995</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>c_4</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>498.689824</td>\n",
        "      <td>9</td>\n",
        "      <td>12</td>\n",
        "      <td>4</td>\n",
        "      <td>4</td>\n",
        "      <td>8</td>\n",
        "      <td>5</td>\n",
        "      <td>321.759002</td>\n",
        "      <td>131.296973</td>\n",
        "      <td>45</td>\n",
        "      <td>100.003988</td>\n",
        "      <td>39</td>\n",
        "      <td>28</td>\n",
        "      <td>425.303916</td>\n",
        "      <td>63.791167</td>\n",
        "      <td>69.601895</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>c_5</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>502.184684</td>\n",
        "      <td>10</td>\n",
        "      <td>9</td>\n",
        "      <td>4</td>\n",
        "      <td>3</td>\n",
        "      <td>5</td>\n",
        "      <td>5</td>\n",
        "      <td>297.358523</td>\n",
        "      <td>100.470180</td>\n",
        "      <td>45</td>\n",
        "      <td>96.943759</td>\n",
        "      <td>38</td>\n",
        "      <td>25</td>\n",
        "      <td>433.129239</td>\n",
        "      <td>57.452172</td>\n",
        "      <td>56.456663</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>c_6</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>500.544560</td>\n",
        "      <td>12</td>\n",
        "      <td>11</td>\n",
        "      <td>4</td>\n",
        "      <td>2</td>\n",
        "      <td>5</td>\n",
        "      <td>6</td>\n",
        "      <td>288.291749</td>\n",
        "      <td>130.229546</td>\n",
        "      <td>41</td>\n",
        "      <td>102.368168</td>\n",
        "      <td>41</td>\n",
        "      <td>21</td>\n",
        "      <td>428.770877</td>\n",
        "      <td>62.391203</td>\n",
        "      <td>66.493702</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>c_7</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>508.734730</td>\n",
        "      <td>11</td>\n",
        "      <td>9</td>\n",
        "      <td>4</td>\n",
        "      <td>4</td>\n",
        "      <td>6</td>\n",
        "      <td>5</td>\n",
        "      <td>341.572993</td>\n",
        "      <td>85.706252</td>\n",
        "      <td>50</td>\n",
        "      <td>106.748456</td>\n",
        "      <td>42</td>\n",
        "      <td>24</td>\n",
        "      <td>429.830463</td>\n",
        "      <td>62.704889</td>\n",
        "      <td>50.332733</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>c_8</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>495.788757</td>\n",
        "      <td>9</td>\n",
        "      <td>12</td>\n",
        "      <td>3</td>\n",
        "      <td>2</td>\n",
        "      <td>7</td>\n",
        "      <td>5</td>\n",
        "      <td>335.026274</td>\n",
        "      <td>99.990115</td>\n",
        "      <td>49</td>\n",
        "      <td>105.191873</td>\n",
        "      <td>35</td>\n",
        "      <td>25</td>\n",
        "      <td>423.358189</td>\n",
        "      <td>67.455814</td>\n",
        "      <td>47.565266</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>c_9</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>516.618334</td>\n",
        "      <td>12</td>\n",
        "      <td>11</td>\n",
        "      <td>3</td>\n",
        "      <td>4</td>\n",
        "      <td>8</td>\n",
        "      <td>6</td>\n",
        "      <td>310.179809</td>\n",
        "      <td>101.542559</td>\n",
        "      <td>46</td>\n",
        "      <td>94.640184</td>\n",
        "      <td>42</td>\n",
        "      <td>25</td>\n",
        "      <td>428.768843</td>\n",
        "      <td>66.431606</td>\n",
        "      <td>32.277536</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>c_10</td>\n",
        "      <td>2003-05-26</td>\n",
        "      <td>482.890706</td>\n",
        "      <td>8</td>\n",
        "      <td>8</td>\n",
        "      <td>6</td>\n",
        "      <td>2</td>\n",
        "      <td>5</td>\n",
        "      <td>5</td>\n",
        "      <td>325.829584</td>\n",
        "      <td>94.687304</td>\n",
        "      <td>46</td>\n",
        "      <td>96.571240</td>\n",
        "      <td>39</td>\n",
        "      <td>25</td>\n",
        "      <td>424.924214</td>\n",
        "      <td>43.868904</td>\n",
        "      <td>58.789972</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "  contestant_id     airdate    flourAmt  saltAmt  yeastAmt  mix1Time  \\\n",
        "0           c_1  2003-05-26  484.289865        9        10         5   \n",
        "1           c_2  2003-05-26  501.259590        9        10         3   \n",
        "2           c_3  2003-05-26  495.194450        8        11         5   \n",
        "3           c_4  2003-05-26  498.689824        9        12         4   \n",
        "4           c_5  2003-05-26  502.184684       10         9         4   \n",
        "5           c_6  2003-05-26  500.544560       12        11         4   \n",
        "6           c_7  2003-05-26  508.734730       11         9         4   \n",
        "7           c_8  2003-05-26  495.788757        9        12         3   \n",
        "8           c_9  2003-05-26  516.618334       12        11         3   \n",
        "9          c_10  2003-05-26  482.890706        8         8         6   \n",
        "\n",
        "   mix1Speed  mix2Time  mix2Speed   water1Amt   water2Amt  waterTemp  \\\n",
        "0          3         5          5  311.663494   98.612837         46   \n",
        "1          4         5          4  315.472473  111.708531         44   \n",
        "2          4         8          4  324.477611  101.911841         48   \n",
        "3          4         8          5  321.759002  131.296973         45   \n",
        "4          3         5          5  297.358523  100.470180         45   \n",
        "5          2         5          6  288.291749  130.229546         41   \n",
        "6          4         6          5  341.572993   85.706252         50   \n",
        "7          2         7          5  335.026274   99.990115         49   \n",
        "8          4         8          6  310.179809  101.542559         46   \n",
        "9          2         5          5  325.829584   94.687304         46   \n",
        "\n",
        "    proveTime  restTime  bakeTime    bakeTemp    ambTemp  ambHumidity  \n",
        "0  105.673044        44        28  435.393493  51.279961    61.443331  \n",
        "1   93.831810        42        21  423.411066  75.692056    35.545136  \n",
        "2  104.969749        40        20  426.042110  59.630283    63.139995  \n",
        "3  100.003988        39        28  425.303916  63.791167    69.601895  \n",
        "4   96.943759        38        25  433.129239  57.452172    56.456663  \n",
        "5  102.368168        41        21  428.770877  62.391203    66.493702  \n",
        "6  106.748456        42        24  429.830463  62.704889    50.332733  \n",
        "7  105.191873        35        25  423.358189  67.455814    47.565266  \n",
        "8   94.640184        42        25  428.768843  66.431606    32.277536  \n",
        "9   96.571240        39        25  424.924214  43.868904    58.789972  "
       ]
      },
      "execution_count": 137,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "pred"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "6023aef5-7018-4e2d-8ba7-760af775e3da",
    "metadata": {},
    "source": [
     "Shape as instances: dictionaries of key:value pairs for only features used in model"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 138,
    "id": "194575d1-dd85-4ef4-86fc-07a84f8b0f18",
    "metadata": {},
    "outputs": [],
    "source": [
     "newobs = pred.to_dict(orient='records')"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 139,
    "id": "7e33ec86-2eb6-4b9c-9d9d-2f67047d7546",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "{'contestant_id': 'c_1',\n",
        " 'airdate': '2003-05-26',\n",
        " 'flourAmt': 484.28986452656386,\n",
        " 'saltAmt': 9,\n",
        " 'yeastAmt': 10,\n",
        " 'mix1Time': 5,\n",
        " 'mix1Speed': 3,\n",
        " 'mix2Time': 5,\n",
        " 'mix2Speed': 5,\n",
        " 'water1Amt': 311.66349401065276,\n",
        " 'water2Amt': 98.61283742264706,\n",
        " 'waterTemp': 46,\n",
        " 'proveTime': 105.67304373851782,\n",
        " 'restTime': 44,\n",
        " 'bakeTime': 28,\n",
        " 'bakeTemp': 435.39349280229476,\n",
        " 'ambTemp': 51.27996072412186,\n",
        " 'ambHumidity': 61.44333141984406}"
       ]
      },
      "execution_count": 139,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "newobs[0]"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "34ab0e25-cbfc-42ea-9499-78ca854b6a5e",
    "metadata": {},
    "source": [
     "### Get Predictions: Python Client\n",
     "\n",
     "Reference:\n",
     "- [aiplatform.Endpoint.predict()](https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.Endpoint#google_cloud_aiplatform_Endpoint_predict)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 140,
    "id": "3f733727-e1a8-411a-a7ad-ce881888347d",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "[73.52679443359375,\n",
        " 85.93708038330078,\n",
        " 86.48188781738281,\n",
        " 82.49593353271484,\n",
        " 82.58346557617188,\n",
        " 78.95195007324219,\n",
        " 79.64701843261719,\n",
        " 84.97455596923828,\n",
        " 82.44728088378906,\n",
        " 82.93528747558594]"
       ]
      },
      "execution_count": 140,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "prediction = endpoint.predict(instances = newobs[0:])\n",
     "prediction.predictions"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "99c4d80f-9a85-47b2-b78e-f79982f7325b",
    "metadata": {},
    "source": [
     "#### Compare to BigQuery ML.PREDICT Results:"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 141,
    "id": "f49bcd76-5f71-47db-a818-e9e7950265b1",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
        "    .dataframe tbody tr th:only-of-type {\n",
        "        vertical-align: middle;\n",
        "    }\n",
        "\n",
        "    .dataframe tbody tr th {\n",
        "        vertical-align: top;\n",
        "    }\n",
        "\n",
        "    .dataframe thead th {\n",
        "        text-align: right;\n",
        "    }\n",
        "</style>\n",
        "<table border=\"1\" class=\"dataframe\">\n",
        "  <thead>\n",
        "    <tr style=\"text-align: right;\">\n",
        "      <th></th>\n",
        "      <th>Recipe</th>\n",
        "      <th>JUDGE_A</th>\n",
        "      <th>predicted_JUDGE_A</th>\n",
        "      <th>Error</th>\n",
        "    </tr>\n",
        "  </thead>\n",
        "  <tbody>\n",
        "    <tr>\n",
        "      <th>0</th>\n",
        "      <td>1</td>\n",
        "      <td>75.0</td>\n",
        "      <td>73.526794</td>\n",
        "      <td>1.473206</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>2</td>\n",
        "      <td>86.0</td>\n",
        "      <td>85.937080</td>\n",
        "      <td>0.062920</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>3</td>\n",
        "      <td>87.0</td>\n",
        "      <td>86.481888</td>\n",
        "      <td>0.518112</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>4</td>\n",
        "      <td>81.0</td>\n",
        "      <td>82.495926</td>\n",
        "      <td>-1.495926</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>5</td>\n",
        "      <td>83.0</td>\n",
        "      <td>82.583466</td>\n",
        "      <td>0.416534</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>5</th>\n",
        "      <td>6</td>\n",
        "      <td>76.0</td>\n",
        "      <td>78.951942</td>\n",
        "      <td>-2.951942</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>6</th>\n",
        "      <td>7</td>\n",
        "      <td>79.0</td>\n",
        "      <td>79.647018</td>\n",
        "      <td>-0.647018</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>7</th>\n",
        "      <td>8</td>\n",
        "      <td>84.0</td>\n",
        "      <td>84.974556</td>\n",
        "      <td>-0.974556</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>8</th>\n",
        "      <td>9</td>\n",
        "      <td>77.0</td>\n",
        "      <td>82.447281</td>\n",
        "      <td>-5.447281</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>9</th>\n",
        "      <td>10</td>\n",
        "      <td>84.0</td>\n",
        "      <td>82.935287</td>\n",
        "      <td>1.064713</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
        "</div>"
       ],
       "text/plain": [
        "   Recipe  JUDGE_A  predicted_JUDGE_A     Error\n",
        "0       1     75.0          73.526794  1.473206\n",
        "1       2     86.0          85.937080  0.062920\n",
        "2       3     87.0          86.481888  0.518112\n",
        "3       4     81.0          82.495926 -1.495926\n",
        "4       5     83.0          82.583466  0.416534\n",
        "5       6     76.0          78.951942 -2.951942\n",
        "6       7     79.0          79.647018 -0.647018\n",
        "7       8     84.0          84.974556 -0.974556\n",
        "8       9     77.0          82.447281 -5.447281\n",
        "9      10     84.0          82.935287  1.064713"
       ]
      },
      "execution_count": 141,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    SELECT Recipe, JUDGE_A, predicted_JUDGE_A, JUDGE_A-predicted_JUDGE_A as Error\n",
     "    FROM ML.PREDICT (MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`,(\n",
     "        SELECT *\n",
     "        FROM `{BQ_PROJECT}.{BQ_DATASET}.{BQ_TABLE}`\n",
     "        WHERE Recipe <= 10)\n",
     "    )\n",
     "    ORDER BY Recipe\n",
     "\"\"\"\n",
     "bq.query(query = query).to_dataframe()"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "de8c3b90-8b38-413b-b7a9-d1c9474d9ccc",
    "metadata": {},
    "source": [
     "---\n",
     "# (Optional) Model Exports\n",
     "\n",
     "The process above used BigQuery ML to train a model and directly register it in Vertex AI's Model Registry.  This enabled simple deployment to a Vertex AI Prediction Endpoint.\n",
     "\n",
     "There may still be times when exporting the model for portability is helpful like:\n",
     "- [model co-hosting with deployment resource pools](https://cloud.google.com/vertex-ai/docs/predictions/model-co-hosting)\n",
     "- workflows with multiple models using [NVIDIA Triton on Vertex AI Prediction](https://cloud.google.com/vertex-ai/docs/predictions/using-nvidia-triton)\n"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "d19addbb-1972-4968-902d-9af4b90ee9ac",
    "metadata": {},
    "source": [
     "### Manually export models from BigQuery\n",
     "\n",
     "The models trained in BQML can be exported from BigQuery to GCS:\n",
     "\n",
     "References:\n",
     "- [Exporting Models](https://cloud.google.com/bigquery/docs/exporting-models)\n",
     "- [`EXPORT MODEL` statement](https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-export-model)\n"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 142,
    "id": "a1579ce3-5d90-4cf5-b1fd-1f4778f0c399",
    "metadata": {},
    "outputs": [],
    "source": [
     "BUCKET = f'{PROJECT_ID}-{REGION}-bqml-exports'"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 143,
    "id": "e2c7ca22-6695-4f2b-a130-4bb14ce9dc9b",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Bucket already exist: statmike-mlops-349915-us-central1-bqml-exports\n"
      ]
     }
    ],
    "source": [
     "if not gcs.lookup_bucket(BUCKET):\n",
     "    bucketDef = gcs.bucket(BUCKET)\n",
     "    bucketDef = gcs.create_bucket(bucketDef, project=PROJECT_ID, location=REGION)\n",
     "    print(f'Created Bucket: {gcs.lookup_bucket(BUCKET).name}')\n",
     "else:\n",
     "    bucketDef = gcs.bucket(BUCKET)\n",
     "    print(f'Bucket already exist: {bucketDef.name}')"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 144,
    "id": "f6ac2bbd-b58c-40b0-945b-727d1a2a335d",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Review the storage bucket in the console here:\n",
       "https://console.cloud.google.com/storage/browser/statmike-mlops-349915-us-central1-bqml-exports;tab=objects&project=statmike-mlops-349915\n"
      ]
     }
    ],
    "source": [
     "print(f'Review the storage bucket in the console here:\\nhttps://console.cloud.google.com/storage/browser/{BUCKET};tab=objects&project={PROJECT_ID}')"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 145,
    "id": "a6f54204-2c53-4560-8fdc-e867d36bd3a5",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "\n",
       "    EXPORT MODEL `statmike-mlops-349915.feature_engineering.bqml_feature_engineering_transform`\n",
       "        OPTIONS (URI = 'gs://statmike-mlops-349915-us-central1-bqml-exports/bqml/model');\n",
       "\n"
      ]
     }
    ],
    "source": [
     "query = f\"\"\"\n",
     "    EXPORT MODEL `{BQ_PROJECT}.{BQ_DATASET}.{BQ_MODEL}`\n",
     "        OPTIONS (URI = 'gs://{BUCKET}/{SERIES}/model');\n",
     "\"\"\"\n",
     "print(query)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 146,
    "id": "91e56b78-1452-440f-ae77-438407f7cb78",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "38.912"
       ]
      },
      "execution_count": 146,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "export = bq.query(query = query)\n",
     "export.result()\n",
     "(export.ended - export.started).total_seconds()"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 147,
    "id": "79436a10-5020-4914-81a0-92945e702577",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
        "[<Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/, 1688563304427607>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/, 1688563304743435>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/assets/model_metadata.json, 1688564779912946>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/explanation_metadata.json, 1688564779500036>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/main.py, 1688564779767363>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/model.bst, 1688564779320162>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/, 1688563305221371>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/assets/, 1688563306204677>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/assets/dict.txt, 1688564776859452>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/fingerprint.pb, 1688564776987701>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/saved_model.pb, 1688564777106577>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/variables/, 1688563305887107>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/variables/variables.data-00000-of-00001, 1688564777243066>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/transform/variables/variables.index, 1688564777457010>,\n",
        " <Blob: statmike-mlops-349915-us-central1-bqml-exports, bqml/model/xgboost_predictor-0.1.tar.gz, 1688564779641692>]"
       ]
      },
      "execution_count": 147,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
     "list(bucketDef.list_blobs(prefix = f'{SERIES}'))"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "9a022c3f-aaa4-4cbb-ab38-bed2d7ab2eb1",
    "metadata": {},
    "source": [
     "**Observe**\n",
     "\n",
     "The model with a `TRANSFORM` clause has a `/model` folder as well as a `/model/transform` that each contain a model save file.  "
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 148,
    "id": "b8e30429-3de8-41d7-a5d7-e9f6977600c9",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Review the models in the storage bucket with the console here:\n",
       "https://console.cloud.google.com/storage/browser/statmike-mlops-349915-us-central1-bqml-exports/bqml;tab=objects&project=statmike-mlops-349915\n"
      ]
     }
    ],
    "source": [
     "print(f'Review the models in the storage bucket with the console here:\\nhttps://console.cloud.google.com/storage/browser/{BUCKET}/{SERIES};tab=objects&project={PROJECT_ID}')"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "cdeee2f2-5765-4747-8112-2244d021bcda",
    "metadata": {},
    "source": [
     "**Review the Models Serving Signature**\n",
     "\n",
     "To understand the features and formats expected by the model as inputs use the [TensorFlow SavedModel CLI](https://www.tensorflow.org/guide/saved_model#details_of_the_savedmodel_command_line_interface) to review the serving signature."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 154,
    "id": "e89664da-1434-4b7a-bc08-3df61324a10e",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "2023-07-05 13:52:39.533343: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA\n",
       "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
       "2023-07-05 13:52:42.923733: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda/lib64:/usr/local/nccl2/lib:/usr/local/cuda/extras/CUPTI/lib64\n",
       "2023-07-05 13:52:42.923909: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda/lib64:/usr/local/nccl2/lib:/usr/local/cuda/extras/CUPTI/lib64\n",
       "2023-07-05 13:52:42.923937: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.\n",
       "The given SavedModel SignatureDef contains the following input(s):\n",
       "  inputs['airdate'] tensor_info:\n",
       "      dtype: DT_STRING\n",
       "      shape: (-1)\n",
       "      name: serving_default_airdate:0\n",
       "  inputs['ambHumidity'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_ambHumidity:0\n",
       "  inputs['ambTemp'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_ambTemp:0\n",
       "  inputs['bakeTemp'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_bakeTemp:0\n",
       "  inputs['bakeTime'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_bakeTime:0\n",
       "  inputs['contestant_id'] tensor_info:\n",
       "      dtype: DT_STRING\n",
       "      shape: (-1)\n",
       "      name: serving_default_contestant_id:0\n",
       "  inputs['flourAmt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_flourAmt:0\n",
       "  inputs['mix1Speed'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_mix1Speed:0\n",
       "  inputs['mix1Time'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_mix1Time:0\n",
       "  inputs['mix2Speed'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_mix2Speed:0\n",
       "  inputs['mix2Time'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_mix2Time:0\n",
       "  inputs['proveTime'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_proveTime:0\n",
       "  inputs['restTime'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_restTime:0\n",
       "  inputs['saltAmt'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_saltAmt:0\n",
       "  inputs['water1Amt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_water1Amt:0\n",
       "  inputs['water2Amt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: serving_default_water2Amt:0\n",
       "  inputs['waterTemp'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_waterTemp:0\n",
       "  inputs['yeastAmt'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: serving_default_yeastAmt:0\n",
       "The given SavedModel SignatureDef contains the following output(s):\n",
       "  outputs['contestant'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:0\n",
       "  outputs['scale_ambHumidity'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:1\n",
       "  outputs['scale_ambTemp'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:2\n",
       "  outputs['scale_bakeTemp'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:3\n",
       "  outputs['scale_bakeTime'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:4\n",
       "  outputs['scale_flourAmt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:5\n",
       "  outputs['scale_mix1Speed'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:6\n",
       "  outputs['scale_mix1Time'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:7\n",
       "  outputs['scale_mix2Speed'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:8\n",
       "  outputs['scale_mix2Time'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:9\n",
       "  outputs['scale_proveTime'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:10\n",
       "  outputs['scale_restTime'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:11\n",
       "  outputs['scale_saltAmt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:12\n",
       "  outputs['scale_water1Amt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:13\n",
       "  outputs['scale_water2Amt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:14\n",
       "  outputs['scale_waterTemp'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:15\n",
       "  outputs['scale_yeastAmt'] tensor_info:\n",
       "      dtype: DT_DOUBLE\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:16\n",
       "  outputs['week'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:17\n",
       "  outputs['year'] tensor_info:\n",
       "      dtype: DT_INT64\n",
       "      shape: (-1)\n",
       "      name: StatefulPartitionedCall:18\n",
       "Method name is: tensorflow/serving/predict\n"
      ]
     }
    ],
    "source": [
     "!saved_model_cli show --dir gs://{BUCKET}/{SERIES}/model/transform --signature_def serving_default --tag_set serve"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "66983c40-9307-4abb-a84a-91ae1044d5ef",
    "metadata": {},
    "source": [
     "---\n",
     "# Remove Resources\n",
     "It is important to remove the resources created for this demonstration to prevent ongoing cost associated with storage (GCS, BigQuery) and Vertex AI Endpoints.  The code below removes all the resources created by this notebook.\n",
     "- gcs bucket delete (contains model exports)\n",
     "- endpoints (need to undeploy any models in use)\n",
     "- bq_model (delete these first, before the dataset.  This also removes them from Vertex AI)\n",
     "- bq_dataset"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 155,
    "id": "0280b64e-39e9-4202-a34d-b5cb6e831575",
    "metadata": {},
    "outputs": [],
    "source": [
     "# delete gcs bucket\n",
     "bucketDef.delete(force = True)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 156,
    "id": "20fb4823-fd55-404e-b4ae-0aa4d0a86b01",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Undeploying Endpoint model: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.models:Undeploying Endpoint model: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Undeploy Endpoint model backing LRO: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944/operations/8407274030947631104\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.models:Undeploy Endpoint model backing LRO: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944/operations/8407274030947631104\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Endpoint model undeployed. Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.models:Endpoint model undeployed. Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Deleting Endpoint : projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.base:Deleting Endpoint : projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Delete Endpoint  backing LRO: projects/1026793852137/locations/us-central1/operations/1919838767720431616\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.base:Delete Endpoint  backing LRO: projects/1026793852137/locations/us-central1/operations/1919838767720431616\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Endpoint deleted. . Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
       "INFO:google.cloud.aiplatform.base:Endpoint deleted. . Resource name: projects/1026793852137/locations/us-central1/endpoints/6287935475437010944\n"
      ]
     }
    ],
    "source": [
     "# delete endpoint, use `force` incase models are deployed or undeploy them first\n",
     "endpoint.delete(force = True)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 157,
    "id": "4faa91fe-a63b-4616-abf5-af837a4579df",
    "metadata": {},
    "outputs": [],
    "source": [
     "# delete models created in BQ with BQML - this also removes them from Vertex AI Model Registry\n",
     "#bq.delete_model(bqml_model)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 158,
    "id": "1cb4ba7d-ff75-484f-8a78-d308fca6c297",
    "metadata": {},
    "outputs": [],
    "source": [
     "# remove the BQ dataset created for this notebook\n",
     "#bq.delete_dataset(ds, delete_contents = True)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "id": "dbf7f397-5539-47a7-833a-6d70ec57260a",
    "metadata": {},
    "outputs": [],
    "source": []
   }
  ],
  "metadata": {
   "environment": {
    "kernel": "python3",
    "name": "tf2-gpu.2-11.m104",
    "type": "gcloud",
    "uri": "gcr.io/deeplearning-platform-release/tf2-gpu.2-11:m104"
   },
   "kernelspec": {
    "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {
     "name": "ipython",
     "version": 3
    },
    "file_extension": ".py",
    "mimetype": "text/x-python",
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
    "version": "3.7.12"
   }
  },
  "nbformat": 4,
  "nbformat_minor": 5
 }