02 - Vertex AI AutoML/02c - Vertex AI _ Pipelines - AutoML with clients (code) In automated pipeline.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "3cebabe4",
   "metadata": {},
   "source": [
    "# 02c - Vertex AI > Pipelines - AutoML\n",
    "\n",
    "Використовуйте конвеєри [Kubeflow](https://www.kubeflow.org/), які працюють на [Vertex AI Pipelines](https://cloud.google.com/vertex-ai/docs/pipelines/introduction), щоб організувати процес навчання користувацької моделі за допомогою AutoML Tabular і розгорнути його на кінцевій точці Vertex AI для отримання (онлайн і пакетних) прогнозів. Демонструється, як автоматизувати процеси (02a) або (02b) за допомогою конвеєрної оркестрації.\n",
    "\n",
    "### Покликання:\n",
    "-  [Vertex AI Pipelines](https://cloud.google.com/vertex-ai/docs/pipelines/build-pipeline#google-cloud-components)\n",
    "-  [Python Client for Vertex AI](https://googleapis.dev/python/aiplatform/latest/aiplatform.html)\n",
    "-  [Kubeflow Pipelines Components for Google Cloud](https://github.com/kubeflow/pipelines/tree/master/components/google-cloud)\n",
    "\n",
    "### Додаткові матеріали:\n",
    "-  Codelab: [Vertex AI Pipelines Introduction](https://codelabs.developers.google.com/vertex-mlmd-pipelines#0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "229cef71",
   "metadata": {},
   "source": [
    "---\n",
    "## Налаштування"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4b6b82e4",
   "metadata": {},
   "source": [
    "вхідні дані:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "63e4ffbd-4d77-4074-b84a-a9a10fe0b606",
   "metadata": {},
   "outputs": [],
   "source": [
    "project = !gcloud config get-value project\n",
    "PROJECT_ID = project[0]\n",
    "PROJECT_ID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b255bf32",
   "metadata": {},
   "outputs": [],
   "source": [
    "REGION = 'us-central1'\n",
    "DATANAME = 'fraud'\n",
    "NOTEBOOK = '02c'\n",
    "\n",
    "# Resources\n",
    "DEPLOY_COMPUTE = 'n1-standard-2'\n",
    "\n",
    "# Model Training\n",
    "VAR_TARGET = 'Class'\n",
    "VAR_OMIT = 'transaction_id' # add more variables to the string with space delimiters"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "00693f36",
   "metadata": {},
   "source": [
    "пакети:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d1e6b18c-6deb-41dc-a632-8b305705eeec",
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install -U google-cloud-pipeline-components -U -q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f3bf8620",
   "metadata": {},
   "outputs": [],
   "source": [
    "from google.cloud import aiplatform\n",
    "from datetime import datetime\n",
    "import kfp\n",
    "from kfp.v2 import compiler\n",
    "#import kfp.v2.dsl as dsl\n",
    "#import google_cloud_pipeline_components as gcc_aip\n",
    "from google_cloud_pipeline_components.v1.dataset import TabularDatasetCreateOp\n",
    "from google_cloud_pipeline_components.v1.automl.training_job import AutoMLTabularTrainingJobRunOp\n",
    "from google_cloud_pipeline_components.v1.endpoint import EndpointCreateOp, ModelDeployOp\n",
    "\n",
    "from google.cloud import bigquery\n",
    "from google.protobuf import json_format\n",
    "from google.protobuf.struct_pb2 import Value\n",
    "import json\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b01a4860",
   "metadata": {},
   "source": [
    "клієнти:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d0f88c50",
   "metadata": {},
   "outputs": [],
   "source": [
    "aiplatform.init(project=PROJECT_ID, location=REGION)\n",
    "bq = bigquery.Client()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d14fbbc",
   "metadata": {},
   "source": [
    "параметри:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f65b2223",
   "metadata": {},
   "outputs": [],
   "source": [
    "TIMESTAMP = datetime.now().strftime(\"%Y%m%d%H%M%S\")\n",
    "BUCKET = PROJECT_ID\n",
    "URI = f\"gs://{BUCKET}/{DATANAME}/models/{NOTEBOOK}\"\n",
    "DIR = f\"temp/{NOTEBOOK}\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2c68630e-4fe3-473d-abb3-16d39fb782e8",
   "metadata": {},
   "outputs": [],
   "source": [
    "SERVICE_ACCOUNT = !gcloud config list --format='value(core.account)' \n",
    "SERVICE_ACCOUNT = SERVICE_ACCOUNT[0]\n",
    "SERVICE_ACCOUNT"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d1d35908-9b69-40be-97dc-fb4fb0279709",
   "metadata": {},
   "source": [
    "Перелік поточних службових акаунтів:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a9f46306-bd55-425a-951a-fd2753807b91",
   "metadata": {},
   "outputs": [],
   "source": [
    "!gcloud projects get-iam-policy $PROJECT_ID --filter=\"bindings.members:$SERVICE_ACCOUNT\" --format='table(bindings.role)' --flatten=\"bindings[].members\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8343a71-3484-4352-9753-ec045cb6dfc5",
   "metadata": {},
   "source": [
    ">Примітка: Якщо у отриманому переліку відсутній [roles/storage.objectAdmin](https://cloud.google.com/storage/docs/access-control/iam-roles), тоді [поверніться до блокнота налаштувань](../00%20-%20Environment%20Setup.ipynb) і додайте цей дозвіл до облікового запису, дотримуючись наданих інструкцій."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "10f15187",
   "metadata": {},
   "source": [
    "оточення:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f7383876",
   "metadata": {},
   "outputs": [],
   "source": [
    "!rm -rf {DIR}\n",
    "!mkdir -p {DIR}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c08c3499",
   "metadata": {},
   "source": [
    "---\n",
    "## Визначення Pipeline (KFP) \n",
    "- Поточний шлях\n",
    "    - Створення набору даних Vertex AI з посилання на таблицю BigQuery\n",
    "    - Створення job для навчання табличних даних Vertex AI AutoML\n",
    "    - Створіть Endpoint та розгорніть модель\n",
    "    \n",
    "Використання [AI Platform Pipeline Components](https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-0.2.0/)\n",
    "- Особливо, [AutoMLTabularTrainingJobRunOp](https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-0.2.0/google_cloud_pipeline_components.aiplatform.html#google_cloud_pipeline_components.aiplatform.AutoMLTabularTrainingJobRunOp)\n",
    "\n",
    "Визначення Job:\n",
    "- Consider Weighting\n",
    "- Model Type\n",
    "- Optimization Objective"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "eb85abc1",
   "metadata": {},
   "outputs": [],
   "source": [
    "@kfp.dsl.pipeline(\n",
    "    name = f'kfp-{NOTEBOOK}-{DATANAME}-{TIMESTAMP}',\n",
    "    pipeline_root = URI+'/'+str(TIMESTAMP)+'/kfp/'\n",
    ")\n",
    "def pipeline(\n",
    "    project: str,\n",
    "    dataname: str,\n",
    "    display_name: str,\n",
    "    deploy_machine: str,\n",
    "    bq_source: str,\n",
    "    var_target: str,\n",
    "    var_omit: str,\n",
    "    features: dict,\n",
    "    labels: dict \n",
    "):\n",
    "    \n",
    "    # dataset\n",
    "    dataset = TabularDatasetCreateOp(\n",
    "        project = project,\n",
    "        display_name = display_name,\n",
    "        bq_source = bq_source,\n",
    "        labels = labels\n",
    "    )\n",
    "    \n",
    "    # training\n",
    "    model = AutoMLTabularTrainingJobRunOp(\n",
    "        project = project,\n",
    "        display_name = display_name,\n",
    "        optimization_prediction_type = \"classification\",\n",
    "        optimization_objective = \"maximize-au-prc\",\n",
    "        budget_milli_node_hours = 1000,\n",
    "        disable_early_stopping=False,\n",
    "        column_specs = features,\n",
    "        dataset = dataset.outputs['dataset'],\n",
    "        target_column = var_target,\n",
    "        predefined_split_column_name = 'splits',\n",
    "        labels = labels\n",
    "    )\n",
    "    \n",
    "    # Endpoint: створення\n",
    "    endpoint = EndpointCreateOp(\n",
    "        project = project,\n",
    "        display_name = display_name,\n",
    "        labels = labels\n",
    "    )\n",
    "    \n",
    "    # Endpoint: розгортання моделі\n",
    "    deployment = ModelDeployOp(\n",
    "        model = model.outputs[\"model\"],\n",
    "        endpoint = endpoint.outputs[\"endpoint\"],\n",
    "        dedicated_resources_min_replica_count = 1,\n",
    "        dedicated_resources_max_replica_count = 1,\n",
    "        traffic_split = {\"0\": 100},\n",
    "        dedicated_resources_machine_type= deploy_machine\n",
    "    )"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4bd7cda6",
   "metadata": {},
   "source": [
    "---\n",
    "## Компіляція Pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "db3e8ce2",
   "metadata": {},
   "outputs": [],
   "source": [
    "compiler.Compiler().compile(\n",
    "    pipeline_func = pipeline,\n",
    "    package_path = f\"{DIR}/{NOTEBOOK}.json\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "72d3a04e",
   "metadata": {},
   "source": [
    "Перемістіть скомпільовані файли Pipeline до GCS Bucket"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "671d4cbe",
   "metadata": {},
   "outputs": [],
   "source": [
    "!gsutil cp {DIR}/{NOTEBOOK}.json {URI}/{TIMESTAMP}/kfp/"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45241ff1",
   "metadata": {},
   "source": [
    "---\n",
    "## Створення Vertex AI Pipeline Job"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9fed3319-fac1-45bb-9483-e02cfd4e0e3e",
   "metadata": {},
   "source": [
    "Отримати словник ознак для вхідних даних конвеєра:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0ad629ff-3843-4e54-8a80-27bd7bb20cb7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# отримання назв змінних\n",
    "query = f\"SELECT * FROM {DATANAME}.INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME = '{DATANAME}_prepped'\"\n",
    "schema = bq.query(query).to_dataframe()\n",
    "OMIT = VAR_OMIT.split() + [VAR_TARGET, 'splits']\n",
    "features = schema[~schema.column_name.isin(OMIT)].column_name.tolist()\n",
    "features = dict.fromkeys(features, 'auto')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0adf1a54-05f1-48f9-99e5-3de1c8f40bd5",
   "metadata": {},
   "source": [
    "Запуск pipeline:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f5dd3e4c",
   "metadata": {},
   "outputs": [],
   "source": [
    "pipeline = aiplatform.PipelineJob(\n",
    "    display_name = f'{NOTEBOOK}_{DATANAME}_{TIMESTAMP}',\n",
    "    template_path = f\"{URI}/{TIMESTAMP}/kfp/{NOTEBOOK}.json\",\n",
    "    parameter_values = {\n",
    "        \"project\" : PROJECT_ID,\n",
    "        \"dataname\" : DATANAME,\n",
    "        \"display_name\" : f'{NOTEBOOK}_{DATANAME}_{TIMESTAMP}',\n",
    "        \"deploy_machine\" : DEPLOY_COMPUTE,\n",
    "        \"bq_source\" : f'bq://{PROJECT_ID}.{DATANAME}.{DATANAME}_prepped',\n",
    "        \"var_target\" : VAR_TARGET,\n",
    "        \"var_omit\" : VAR_OMIT,\n",
    "        \"features\" : features,\n",
    "        \"labels\" : {'notebook': NOTEBOOK}       \n",
    "    },\n",
    "    labels = {'notebook': NOTEBOOK},\n",
    "    enable_caching=False\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5ee521df",
   "metadata": {
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "response = pipeline.run(\n",
    "    service_account = SERVICE_ACCOUNT\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6e4c8e67-bceb-4629-a246-6a6f18a1c530",
   "metadata": {},
   "source": [
    "Візуальне представлення pipeline можна переглянути в консолі:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bd262fe9-0693-446a-be88-779e15fb0809",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(f\"Review the Pipeline as it runs here:\\nhttps://console.cloud.google.com/vertex-ai/locations/{REGION}/pipelines/runs/{pipeline.resource_name.split('/')[-1]}?project={PROJECT_ID}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9d982aa6-2999-47f6-92a0-01510050e681",
   "metadata": {},
   "source": [
    "Отримати інформацію про pipeline:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "29ff7495",
   "metadata": {},
   "outputs": [],
   "source": [
    "aiplatform.get_pipeline_df(pipeline = f'kfp-{NOTEBOOK}-{DATANAME}-{TIMESTAMP}')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0523ffaa-9dfb-41cb-b7b5-6cc6ddae1613",
   "metadata": {
    "tags": []
   },
   "source": [
    "---\n",
    "## Оцінювання\n",
    "\n",
    "Для отримання додаткової інформації перегляньте [цю сторінку] (https://cloud.google.com/vertex-ai/docs/training/evaluating-automl-models)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "27b1c3f1",
   "metadata": {},
   "source": [
    "Отримання моделі:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b6634c0d",
   "metadata": {},
   "outputs": [],
   "source": [
    "models = aiplatform.Model.list(filter=f'labels.notebook={NOTEBOOK}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a4f7b61b-ad0f-487b-8dc6-ac8ed12e6afb",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = models[0]\n",
    "model.resource_name"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "71227e04-8bb0-46d5-862e-3fddfe0361ae",
   "metadata": {},
   "source": [
    "Повертає агреговані метрики оцінки моделі для моделі в цілому.  \n",
    "\n",
    "Обидва способи:\n",
    "- Спочатку використовуйте `model.list_model_evaluations()` для отримання ідентифікатора оцінки, потім використовуйте `model.get_model_evaluation(evaluation_id = )` для ідентифікатора оцінки\n",
    "- Або скористайтеся `.get_model_evaluation()` і отримаєте першу оцінку моделі"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b7b5d7e0-6ef5-4787-a5d0-953848318989",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation = model.get_model_evaluation().to_dict() # get first evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "35848c5d-d9af-4109-abd5-f093adba2acc",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1763abbd-d2f4-430a-9a03-9e5af413b82a",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation['metrics'].keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a562dcf6-02af-4185-96be-5b8b77a5cfd7",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation['metrics']['auPrc']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "45bd1c4b-00a3-4b2c-95a0-bb837307f8a0",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation['metrics']['confidenceMetrics'][3]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "39aa9551-2e4a-4673-b0d5-1ddec1233e0e",
   "metadata": {},
   "source": [
    "Перегляньте кілька показників. Також порівняйте їх з результатами з консолі."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bedd8a0e-cb68-4c45-bb58-c8622b9b0a96",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(f\"Переглянути цю модель в консолі:\\nhttps://console.cloud.google.com/vertex-ai/locations/{REGION}/models/{model.name}/versions/{model.version_id}/evaluations/{evaluation['name'].split('/')[-1]}?project={PROJECT_ID}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8dd6e04f-c0ec-4ad1-a8be-460cddcb0267",
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation['metrics']['auPrc']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "edd9892b-eeac-4a64-aa9b-68a7a45a3e76",
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(len(evaluation['metrics']['confusionMatrix']['annotationSpecs'])):\n",
    "    print('True Label = ', evaluation['metrics']['confusionMatrix']['annotationSpecs'][i]['displayName'], ' has Predicted labels = ', evaluation['metrics']['confusionMatrix']['rows'][i])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2266afe6-be3f-4565-afd9-cb839682df2c",
   "metadata": {},
   "source": [
    "Для моделей з мітками ви також можете отримати метрики оцінювання для кожного зрізу моделі, використовуючи версію .gapic api:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "703ce330-8b74-44ff-a71f-f7670a1d924f",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_client = aiplatform.gapic.ModelServiceClient(\n",
    "    client_options = {\n",
    "        'api_endpoint' : f'{REGION}-aiplatform.googleapis.com'\n",
    "    }\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "045a6f0d-ab29-4825-a29e-335f5a1727e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "slices = model_client.list_model_evaluation_slices(parent = evaluation['name'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4b7b5c5a-d43c-4c77-8806-fa92f6438e19",
   "metadata": {},
   "outputs": [],
   "source": [
    "for slice in slices:\n",
    "    print('Label = ', slice.slice_.value, 'has auPrc = ', slice.metrics['auPrc'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3829782e",
   "metadata": {},
   "source": [
    "---\n",
    "## Прогнозування"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0269e25d",
   "metadata": {},
   "source": [
    "### Вхідні дані для прогнозування"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5bda8542",
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = bq.query(\n",
    "    query = f\"\"\"\n",
    "        SELECT * EXCEPT({VAR_TARGET}, splits, {VAR_OMIT})\n",
    "        FROM {DATANAME}.{DATANAME}_prepped\n",
    "        WHERE splits='TEST'\n",
    "        LIMIT 10\n",
    "    \"\"\"\n",
    ").to_dataframe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bb5a77b7",
   "metadata": {},
   "outputs": [],
   "source": [
    "pred.head(4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4c32ed6b",
   "metadata": {},
   "outputs": [],
   "source": [
    "pred['Time'] = pred['Time'].astype(str)\n",
    "newobs = pred.to_dict(orient='records')\n",
    "newobs[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c2fde74e",
   "metadata": {},
   "source": [
    "Наступні комірки отримують модель з кінцевої точки та її схеми:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d438383e",
   "metadata": {},
   "outputs": [],
   "source": [
    "instances = [json_format.ParseDict(newob, Value()) for newob in newobs]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec61f206",
   "metadata": {},
   "source": [
    "### Отримання прогнозів: Python Client"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "116fefd9-e7c7-4971-8e6b-d630469f2330",
   "metadata": {},
   "outputs": [],
   "source": [
    "aiplatform.Endpoint.list(filter=f'labels.notebook={NOTEBOOK}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "279cf479",
   "metadata": {},
   "outputs": [],
   "source": [
    "endpoint = aiplatform.Endpoint.list(filter=f'labels.notebook={NOTEBOOK}')[0]\n",
    "endpoint.display_name"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c7c298d5",
   "metadata": {},
   "outputs": [],
   "source": [
    "prediction = endpoint.predict(instances = instances) # or instances = newobs\n",
    "prediction.predictions[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1323d595",
   "metadata": {},
   "outputs": [],
   "source": [
    "prediction.predictions[0]['classes'][np.argmax(prediction.predictions[0]['scores'])]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f68781f2",
   "metadata": {},
   "source": [
    "### Отримання прогнозів: REST"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f6b3a5b9",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(f'{DIR}/request.json','w') as file:\n",
    "    file.write(json.dumps({\"instances\": [newobs[0]]}))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6e7b1b91",
   "metadata": {},
   "outputs": [],
   "source": [
    "!curl -X POST \\\n",
    "-H \"Authorization: Bearer \"$(gcloud auth application-default print-access-token) \\\n",
    "-H \"Content-Type: application/json; charset=utf-8\" \\\n",
    "-d @{DIR}/request.json \\\n",
    "https://{REGION}-aiplatform.googleapis.com/v1/{endpoint.resource_name}:predict"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7b428c9d",
   "metadata": {},
   "source": [
    "### Отримання прогнозів: gcloud (CLI)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d3523017",
   "metadata": {},
   "outputs": [],
   "source": [
    "!gcloud beta ai endpoints predict {endpoint.name.rsplit('/',1)[-1]} --region={REGION} --json-request={DIR}/request.json"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d0d825b",
   "metadata": {},
   "source": [
    "---\n",
    "## Інтерпретація\n",
    "Гайд\n",
    "- https://cloud.google.com/vertex-ai/docs/predictions/interpreting-results-automl#tabular"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "80280fc8",
   "metadata": {},
   "outputs": [],
   "source": [
    "explanation = endpoint.explain(instances = instances)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6b2a6912",
   "metadata": {},
   "outputs": [],
   "source": [
    "explanation.predictions[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a862b22a",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(\"attribution:\")\n",
    "print(\"baseline output\",explanation.explanations[0].attributions[0].baseline_output_value)\n",
    "print(\"instance output\",explanation.explanations[0].attributions[0].instance_output_value)\n",
    "print(\"output_index\",explanation.explanations[0].attributions[0].output_index)\n",
    "print(\"output display value\",explanation.explanations[0].attributions[0].output_display_name)\n",
    "print(\"approximation error\",explanation.explanations[0].attributions[0].approximation_error)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6afdc889-9620-4e5c-adf9-5ecf2238ad75",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "features = []\n",
    "scores = []\n",
    "for k in explanation.explanations[0].attributions[0].feature_attributions:\n",
    "    features.append(k)\n",
    "    scores.append(explanation.explanations[0].attributions[0].feature_attributions[k])\n",
    "features = [x for _, x in sorted(zip(scores, features))]\n",
    "scores = sorted(scores)\n",
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(9, 9)\n",
    "ax.barh(features, scores)\n",
    "fig.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1a12e906",
   "metadata": {},
   "source": [
    "---\n",
    "## Batch Predictions:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7145bae3",
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = aiplatform.BatchPredictionJob.create(\n",
    "    job_display_name = f'{NOTEBOOK}_{DATANAME}_{TIMESTAMP}',\n",
    "    model_name = endpoint.list_models()[0].model,\n",
    "    instances_format = \"bigquery\",\n",
    "    predictions_format = \"bigquery\",\n",
    "    bigquery_source = f'bq://{PROJECT_ID}.{DATANAME}.{DATANAME}_prepped',\n",
    "    bigquery_destination_prefix = f\"{PROJECT_ID}\",\n",
    "    generate_explanation=True,\n",
    "    labels = {'notebook':f'{NOTEBOOK}'}\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4022b5bd",
   "metadata": {},
   "source": [
    "---\n",
    "## Видалити ресурси\n",
    "див. зошит [_Cleanup](_Cleanup.ipynb)"
   ]
  }
 ],
 "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
}