analytics-zoo-ncf-movielens.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "2021-06-27-analytics-zoo-ncf-movielens.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "authorship_tag": "ABX9TyMaJVQP5DiI+UgE9DsX8cGJ"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vWAyqrazbVkU"
      },
      "source": [
        "# Analytics Zoo Recommendation Part 1\n",
        "> Applying NCF on Movielens using Analytics Zoo library\n",
        "\n",
        "- toc: true\n",
        "- badges: true\n",
        "- comments: true\n",
        "- categories: [Movie, BigData, PySpark, AnalyticsZoo, NCF]\n",
        "- author: \"<a href='https://nbviewer.jupyter.org/github/intel-analytics/analytics-zoo/blob/master/apps/recommendation-ncf/ncf-explicit-feedback.ipynb'>Analytics Zoo</a>\"\n",
        "- image:"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5tJLYN9yYJxO"
      },
      "source": [
        "## Introduction"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f4e-a8D8YSXJ"
      },
      "source": [
        "NCF Recommender with Explict Feedback\n",
        "\n",
        "In this notebook we demostrate how to build a neural network recommendation system, Neural Collaborative Filtering(NCF) with explict feedback. We use Recommender API in Analytics Zoo to build a model, and use optimizer of BigDL to train the model. \n",
        "\n",
        "The system ([Recommendation systems: Principles, methods and evaluation](http://www.sciencedirect.com/science/article/pii/S1110866515000341)) normally prompts the user through the system interface to provide ratings for items in order to construct and improve the model. The accuracy of recommendation depends on the quantity of ratings provided by the user.  \n",
        "\n",
        "NCF([He, 2015](https://www.comp.nus.edu.sg/~xiangnan/papers/ncf.pdf)) leverages a multi-layer perceptrons to learn the user–item interaction function, at the mean time, NCF can express and generalize matrix factorization under its framework. includeMF(Boolean) is provided for users to build a NCF with or without matrix factorization. \n",
        "\n",
        "Data: \n",
        "* The dataset we used is movielens-1M ([link](https://grouplens.org/datasets/movielens/1m/)), which contains 1 million ratings from 6000 users on 4000 movies.  There're 5 levels of rating. We will try classify each (user,movie) pair into 5 classes and evaluate the effect of algortithms using Mean Absolute Error.  \n",
        "  \n",
        "References: \n",
        "* A Keras implementation of Movie Recommendation([notebook](https://github.com/ririw/ririw.github.io/blob/master/assets/Recommending%20movies.ipynb)) from the [blog](http://blog.richardweiss.org/2016/09/25/movie-embeddings.html).\n",
        "* Nerual Collaborative filtering ([He, 2015](https://www.comp.nus.edu.sg/~xiangnan/papers/ncf.pdf))\n",
        "\n",
        "Python interface:\n",
        "\n",
        "```python\n",
        "ncf = NeuralCF(user_count, item_count, class_num, user_embed=20, item_embed=20, hidden_layers=(40, 20, 10), include_mf=True, mf_embed=20)\n",
        "```\n",
        "\n",
        "- `user_count`: The number of users. Positive int.\n",
        "- `item_count`: The number of classes. Positive int.\n",
        "- `class_num`: The number of classes. Positive int.\n",
        "- `user_embed`: Units of user embedding. Positive int. Default is 20.\n",
        "- `item_embed`: itemEmbed Units of item embedding. Positive int. Default is 20.\n",
        "- `hidden_layers`: Units of hidden layers for MLP. Tuple of positive int. Default is (40, 20, 10).\n",
        "- `include_mf`: Whether to include Matrix Factorization. Boolean. Default is True.\n",
        "- `mf_embed`: Units of matrix factorization embedding. Positive int. Default is 20."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IPqTsm9pSckt"
      },
      "source": [
        "## Installation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u5UkOC-3SuwP"
      },
      "source": [
        "### Install Java 8"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nY8_OcwpSyha"
      },
      "source": [
        "Run the command on the colaboratory file to install jdk 1.8"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8oH5g_58SScM",
        "outputId": "37ff9435-54d8-4846-f6b4-3ee62f6def22"
      },
      "source": [
        "# Install jdk8\n",
        "!apt-get install openjdk-8-jdk-headless -qq > /dev/null\n",
        "# Set jdk environment path which enables you to run Pyspark in your Colab environment.\n",
        "import os\n",
        "os.environ[\"JAVA_HOME\"] = \"/usr/lib/jvm/java-8-openjdk-amd64\"\n",
        "!update-alternatives --set java /usr/lib/jvm/java-8-openjdk-amd64/jre/bin/java"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "update-alternatives: using /usr/lib/jvm/java-8-openjdk-amd64/jre/bin/java to provide /usr/bin/java (java) in manual mode\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zPqef37WS2r-"
      },
      "source": [
        "### Install Analytics Zoo from pip"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Mo4najsbS5RX"
      },
      "source": [
        "You can add the following command on your colab file to install the analytics-zoo via pip easily:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RtHKGTcSS1cD",
        "outputId": "262ef946-24f1-419e-9fd8-b9276f29b3d9"
      },
      "source": [
        "# Install latest release version of analytics-zoo \n",
        "# Installing analytics-zoo from pip will automatically install pyspark, bigdl, and their dependencies.\n",
        "!pip install analytics-zoo"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting analytics-zoo\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/49/32/2fe969c682b8683fb3792f06107fe8ac56f165e307d327756c506aa76d31/analytics_zoo-0.10.0-py2.py3-none-manylinux1_x86_64.whl (158.9MB)\n",
            "\u001b[K     |████████████████████████████████| 158.9MB 44kB/s \n",
            "\u001b[?25hCollecting bigdl==0.12.2\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/1a/40/81fed203a633536dbb83454f1a102ebde86214f576572769290afb9427c9/BigDL-0.12.2-py2.py3-none-manylinux1_x86_64.whl (114.1MB)\n",
            "\u001b[K     |████████████████████████████████| 114.1MB 78kB/s \n",
            "\u001b[?25hCollecting conda-pack==0.3.1\n",
            "  Downloading https://files.pythonhosted.org/packages/e9/e7/d942780c4281a665f34dbfffc1cd1517c5843fb478c133a1e1fa0df30cd6/conda_pack-0.3.1-py2.py3-none-any.whl\n",
            "Collecting pyspark==2.4.3\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/37/98/244399c0daa7894cdf387e7007d5e8b3710a79b67f3fd991c0b0b644822d/pyspark-2.4.3.tar.gz (215.6MB)\n",
            "\u001b[K     |████████████████████████████████| 215.6MB 63kB/s \n",
            "\u001b[?25hRequirement already satisfied: numpy>=1.7 in /usr/local/lib/python3.7/dist-packages (from bigdl==0.12.2->analytics-zoo) (1.19.5)\n",
            "Requirement already satisfied: six>=1.10.0 in /usr/local/lib/python3.7/dist-packages (from bigdl==0.12.2->analytics-zoo) (1.15.0)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.7/dist-packages (from conda-pack==0.3.1->analytics-zoo) (57.0.0)\n",
            "Collecting py4j==0.10.7\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/e3/53/c737818eb9a7dc32a7cd4f1396e787bd94200c3997c72c1dbe028587bd76/py4j-0.10.7-py2.py3-none-any.whl (197kB)\n",
            "\u001b[K     |████████████████████████████████| 204kB 37.1MB/s \n",
            "\u001b[?25hBuilding wheels for collected packages: pyspark\n",
            "  Building wheel for pyspark (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for pyspark: filename=pyspark-2.4.3-py2.py3-none-any.whl size=215964968 sha256=f00edffc8d08d9fa648fa23ecd0f5e6877c59541ad212b7e8720efc22cf5c0f3\n",
            "  Stored in directory: /root/.cache/pip/wheels/8d/20/f0/b30e2024226dc112e256930dd2cd4f06d00ab053c86278dcf3\n",
            "Successfully built pyspark\n",
            "Installing collected packages: py4j, pyspark, bigdl, conda-pack, analytics-zoo\n",
            "Successfully installed analytics-zoo-0.10.0 bigdl-0.12.2 conda-pack-0.3.1 py4j-0.10.7 pyspark-2.4.3\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "M8KiG7WhTAPa"
      },
      "source": [
        "### Initialize context"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZWWxCdrkTCqY"
      },
      "source": [
        "Call init_nncontext() that will create a SparkContext with optimized performance configurations."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Wy6CnH-8S81r",
        "outputId": "b36c7b07-c603-4619-d32c-16c27724ed05"
      },
      "source": [
        "from zoo.common.nncontext import*\n",
        "\n",
        "sc = init_nncontext()"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Prepending /usr/local/lib/python3.7/dist-packages/bigdl/share/conf/spark-bigdl.conf to sys.path\n",
            "Adding /usr/local/lib/python3.7/dist-packages/zoo/share/lib/analytics-zoo-bigdl_0.12.2-spark_2.4.3-0.10.0-jar-with-dependencies.jar to BIGDL_JARS\n",
            "Prepending /usr/local/lib/python3.7/dist-packages/zoo/share/conf/spark-analytics-zoo.conf to sys.path\n",
            "pyspark_submit_args is:  --driver-class-path /usr/local/lib/python3.7/dist-packages/zoo/share/lib/analytics-zoo-bigdl_0.12.2-spark_2.4.3-0.10.0-jar-with-dependencies.jar:/usr/local/lib/python3.7/dist-packages/bigdl/share/lib/bigdl-0.12.2-jar-with-dependencies.jar pyspark-shell \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nLQymd-_TbKx"
      },
      "source": [
        "Analytics Zoo provides three Recommenders, including Wide and Deep (WND) model, Neural network-based Collaborative Filtering (NCF) model and Session Recommender model. Easy-to-use Keras-Style defined models which provides compile and fit methods for training. Alternatively, they could be fed into NNFrames or BigDL Optimizer.\n",
        "\n",
        "WND and NCF recommenders can handle either explict or implicit feedback, given corresponding features."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_KTBMy52T5he"
      },
      "source": [
        "## Imports"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nsib3Wg_T7E5",
        "outputId": "0e1481ac-014e-45a1-ec48-1156aae11038"
      },
      "source": [
        "from zoo.pipeline.api.keras.layers import *\n",
        "from zoo.models.recommendation import UserItemFeature\n",
        "from zoo.models.recommendation import NeuralCF\n",
        "from zoo.common.nncontext import init_nncontext\n",
        "import matplotlib\n",
        "from sklearn import metrics\n",
        "from operator import itemgetter\n",
        "from bigdl.util.common import *\n",
        "\n",
        "import os\n",
        "import numpy as np\n",
        "\n",
        "matplotlib.use('agg')\n",
        "import matplotlib.pyplot as plt\n",
        "%pylab inline"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Populating the interactive namespace from numpy and matplotlib\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Cl6H99sBTgOt"
      },
      "source": [
        "## Download movielens dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wkKAeEORTPZ_",
        "outputId": "37526f44-10ec-4969-f223-9da3f3236fa6"
      },
      "source": [
        "!wget http://files.grouplens.org/datasets/movielens/ml-1m.zip\n",
        "!unzip ml-1m.zip"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-06-27 11:51:34--  http://files.grouplens.org/datasets/movielens/ml-1m.zip\n",
            "Resolving files.grouplens.org (files.grouplens.org)... 128.101.65.152\n",
            "Connecting to files.grouplens.org (files.grouplens.org)|128.101.65.152|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 5917549 (5.6M) [application/zip]\n",
            "Saving to: ‘ml-1m.zip’\n",
            "\n",
            "ml-1m.zip           100%[===================>]   5.64M  29.8MB/s    in 0.2s    \n",
            "\n",
            "2021-06-27 11:51:35 (29.8 MB/s) - ‘ml-1m.zip’ saved [5917549/5917549]\n",
            "\n",
            "Archive:  ml-1m.zip\n",
            "   creating: ml-1m/\n",
            "  inflating: ml-1m/movies.dat        \n",
            "  inflating: ml-1m/ratings.dat       \n",
            "  inflating: ml-1m/README            \n",
            "  inflating: ml-1m/users.dat         \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cizJHBZhVV15"
      },
      "source": [
        "## Read the dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vT8aZM0BT2oi"
      },
      "source": [
        "def read_data_sets(data_dir):\n",
        "  rating_files = os.path.join(data_dir,\"ratings.dat\")\n",
        "  rating_list = [i.strip().split(\"::\") for i in open(rating_files,\"r\").readlines()]    \n",
        "  movielens_data = np.array(rating_list).astype(int)\n",
        "  return movielens_data \n",
        "\n",
        "def get_id_pairs(data_dir):\n",
        "\tmovielens_data = read_data_sets(data_dir)\n",
        "\treturn movielens_data[:, 0:2]\n",
        "\n",
        "def get_id_ratings(data_dir):\n",
        "\tmovielens_data = read_data_sets(data_dir)\n",
        "\treturn movielens_data[:, 0:3]"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7n6vQz-3VR4I"
      },
      "source": [
        "movielens_data = get_id_ratings(\"/content/ml-1m\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GCRIqd0hV2-5"
      },
      "source": [
        "## Understand the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nY3Bbk1xV8wL",
        "outputId": "45879150-5873-4857-e06c-313fcdebd89b"
      },
      "source": [
        "min_user_id = np.min(movielens_data[:,0])\n",
        "max_user_id = np.max(movielens_data[:,0])\n",
        "min_movie_id = np.min(movielens_data[:,1])\n",
        "max_movie_id = np.max(movielens_data[:,1])\n",
        "rating_labels= np.unique(movielens_data[:,2])\n",
        "\n",
        "print(movielens_data.shape)\n",
        "print(min_user_id, max_user_id, min_movie_id, max_movie_id, rating_labels)"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(1000209, 3)\n",
            "1 6040 1 3952 [1 2 3 4 5]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vrs_dyQcVz8z"
      },
      "source": [
        "Each record is in format of (userid, movieid, rating_score). UserIDs range between 1 and 6040. MovieIDs range between 1 and 3952. Ratings are made on a 5-star scale (whole-star ratings only). Counts of users and movies are recorded for later use."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "O-Ks_RL2WMT7"
      },
      "source": [
        "## Transformation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cnUhH-ksWOxh"
      },
      "source": [
        "Transform original data into RDD of sample. We use optimizer of BigDL directly to train the model, it requires data to be provided in format of RDD(Sample). A Sample is a BigDL data structure which can be constructed using 2 numpy arrays, feature and label respectively. The API interface is Sample.from_ndarray(feature, label) Here, labels are tranformed into zero-based since original labels start from 1."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gmD0lS0eVnK7"
      },
      "source": [
        "def build_sample(user_id, item_id, rating):\n",
        "    sample = Sample.from_ndarray(np.array([user_id, item_id]), np.array([rating]))\n",
        "    return UserItemFeature(user_id, item_id, sample)"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BcwsB2SsWVBZ",
        "outputId": "e7a3f1f0-174a-4392-816d-6d5f9e4d6d17"
      },
      "source": [
        "pairFeatureRdds = sc.parallelize(movielens_data).map(lambda x: build_sample(x[0], x[1], x[2]-1))\n",
        "pairFeatureRdds.take(3)"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<zoo.models.recommendation.recommender.UserItemFeature at 0x7fde70f09910>,\n",
              " <zoo.models.recommendation.recommender.UserItemFeature at 0x7fde70d792d0>,\n",
              " <zoo.models.recommendation.recommender.UserItemFeature at 0x7fde55409e10>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wfqh_7FsWkrw"
      },
      "source": [
        "## Split"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qw37Flz4Wm0F"
      },
      "source": [
        "Randomly split the data into train (80%) and validation (20%)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "plRhKQ0BWdD5",
        "outputId": "1629568b-f74c-4a67-cb88-774a265963a6"
      },
      "source": [
        "trainPairFeatureRdds, valPairFeatureRdds = pairFeatureRdds.randomSplit([0.8, 0.2], seed= 1)\n",
        "valPairFeatureRdds.cache()\n",
        "\n",
        "train_rdd= trainPairFeatureRdds.map(lambda pair_feature: pair_feature.sample)\n",
        "val_rdd= valPairFeatureRdds.map(lambda pair_feature: pair_feature.sample)\n",
        "val_rdd.persist()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "PythonRDD[3] at RDD at PythonRDD.scala:53"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PCWetBlaWuyb",
        "outputId": "bcaeaa9c-ef6e-4169-8eb1-6ea542a88e0f"
      },
      "source": [
        "train_rdd.count()"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "800491"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "czo63lHuW2DB",
        "outputId": "655ac868-afc6-4dc1-bf94-92bb3feda2d4"
      },
      "source": [
        "train_rdd.take(3)"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[Sample: features: [JTensor: storage: [1.000e+00 1.193e+03], shape: [2], float], labels: [JTensor: storage: [4.], shape: [1], float],\n",
              " Sample: features: [JTensor: storage: [  1. 914.], shape: [2], float], labels: [JTensor: storage: [2.], shape: [1], float],\n",
              " Sample: features: [JTensor: storage: [1.000e+00 3.408e+03], shape: [2], float], labels: [JTensor: storage: [3.], shape: [1], float]]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KnONxSMsXARF"
      },
      "source": [
        "## Build model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IhpOgCvFXG2e"
      },
      "source": [
        "In Analytics Zoo, it is simple to build NCF model by calling NeuralCF API. You need specify the user count, item count and class number according to your data, then add hidden layers as needed, you can also choose to include matrix factorization in the network. The model could be fed into an Optimizer of BigDL or NNClassifier of analytics-zoo. Please refer to the document for more details. In this example, we demostrate how to use optimizer of BigDL."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4xMI7FZqW3zv",
        "outputId": "29be18e7-4be2-4ef4-8cdf-8dd642716e33"
      },
      "source": [
        "ncf = NeuralCF(user_count=max_user_id, \n",
        "               item_count=max_movie_id, \n",
        "               class_num=5, \n",
        "               hidden_layers=[20, 10], \n",
        "               include_mf = False)"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "creating: createZooKerasInput\n",
            "creating: createZooKerasFlatten\n",
            "creating: createZooKerasSelect\n",
            "creating: createZooKerasFlatten\n",
            "creating: createZooKerasSelect\n",
            "creating: createZooKerasEmbedding\n",
            "creating: createZooKerasEmbedding\n",
            "creating: createZooKerasFlatten\n",
            "creating: createZooKerasFlatten\n",
            "creating: createZooKerasMerge\n",
            "creating: createZooKerasDense\n",
            "creating: createZooKerasDense\n",
            "creating: createZooKerasDense\n",
            "creating: createZooKerasModel\n",
            "creating: createZooNeuralCF\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BVfHXNF2XMLp"
      },
      "source": [
        "## Compile model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "R_Xgwv_MXUUI"
      },
      "source": [
        "Compile model given specific optimizers, loss, as well as metrics for evaluation. Optimizer tries to minimize the loss of the neural net with respect to its weights/biases, over the training set. To create an Optimizer in BigDL, you want to at least specify arguments: model(a neural network model), criterion(the loss function), traing_rdd(training dataset) and batch size. Please refer to [ProgrammingGuide](https://bigdl-project.github.io/master/#ProgrammingGuide/optimization/) and [Optimizer](https://bigdl-project.github.io/master/#APIGuide/Optimizers/Optimizer/) for more details to create efficient optimizers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "EpiU9RG_XKKM",
        "outputId": "7fbae0d6-4c77-437e-bfde-a2597a0d0655"
      },
      "source": [
        "ncf.compile(optimizer= \"adam\",\n",
        "            loss= \"sparse_categorical_crossentropy\",\n",
        "            metrics=['accuracy'])"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "creating: createAdam\n",
            "creating: createZooKerasSparseCategoricalCrossEntropy\n",
            "creating: createZooKerasSparseCategoricalAccuracy\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pOvfFZP3Xk0r"
      },
      "source": [
        "## Collect logs"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ult1tu6QXnaX"
      },
      "source": [
        "You can leverage tensorboard to see the summaries."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E44IwOlzXjYW"
      },
      "source": [
        "tmp_log_dir = create_tmp_path()\n",
        "ncf.set_tensorboard(tmp_log_dir, \"training_ncf\")"
      ],
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IKipPFhsXqxS"
      },
      "source": [
        "## Train the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4IHzRgUyXpu8"
      },
      "source": [
        "ncf.fit(train_rdd, \n",
        "        nb_epoch= 10, \n",
        "        batch_size= 8000, \n",
        "        validation_data=val_rdd)"
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YLj6L9JPXxri"
      },
      "source": [
        "## Prediction"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uzDas0oIX11i"
      },
      "source": [
        "Zoo models make inferences based on the given data using model.predict(val_rdd) API. A result of RDD is returned. predict_class returns the predicted label."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gUkaK6TUXuDJ",
        "outputId": "d2698771-6f5f-411c-f784-5f832806f570"
      },
      "source": [
        "results = ncf.predict(val_rdd)\n",
        "results.take(5)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[array([0.00806034, 0.02006759, 0.18866627, 0.63047034, 0.15273541],\n",
              "       dtype=float32),\n",
              " array([0.00254512, 0.00567304, 0.06865695, 0.46720853, 0.4559163 ],\n",
              "       dtype=float32),\n",
              " array([0.00535633, 0.02252354, 0.12654975, 0.4168996 , 0.42867082],\n",
              "       dtype=float32),\n",
              " array([0.00396674, 0.01450644, 0.12073953, 0.60501534, 0.255772  ],\n",
              "       dtype=float32),\n",
              " array([0.00394357, 0.01327773, 0.14935595, 0.6658978 , 0.16752498],\n",
              "       dtype=float32)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "aLhHUm7sZm1Y",
        "outputId": "f85dcae9-e6eb-46b7-e748-27b031bff564"
      },
      "source": [
        "results_class = ncf.predict_class(val_rdd)\n",
        "results_class.take(5)"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[4, 4, 5, 4, 4]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WqOfNL7gZrG6"
      },
      "source": [
        "In Analytics Zoo, Recommender has provied 3 unique APIs to predict user-item pairs and make recommendations for users or items given candidates."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9ffsw5z7aBlO"
      },
      "source": [
        "### Predict for user item pairs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NXHMNrSsZnOD",
        "outputId": "6115e527-45ab-4a7b-f145-83f37839cda4"
      },
      "source": [
        "userItemPairPrediction = ncf.predict_user_item_pair(valPairFeatureRdds)\n",
        "for result in userItemPairPrediction.take(5): print(result)"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "UserItemPrediction [user_id: 1, item_id: 661, prediction: 4, probability: 0.630470335483551]\n",
            "UserItemPrediction [user_id: 1, item_id: 2804, prediction: 4, probability: 0.46720853447914124]\n",
            "UserItemPrediction [user_id: 1, item_id: 1035, prediction: 5, probability: 0.42867082357406616]\n",
            "UserItemPrediction [user_id: 1, item_id: 2687, prediction: 4, probability: 0.6050153374671936]\n",
            "UserItemPrediction [user_id: 1, item_id: 2321, prediction: 4, probability: 0.665897786617279]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SCgDNVQaZ-mJ"
      },
      "source": [
        "### Recommend 3 items for each user given candidates in the feature RDDs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "W2uMAeiOZ5RJ",
        "outputId": "7a558bf0-9d57-4922-a487-d1fe67c4f3a0"
      },
      "source": [
        "userRecs = ncf.recommend_for_user(valPairFeatureRdds, 3)\n",
        "for result in userRecs.take(5): print(result)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "UserItemPrediction [user_id: 4904, item_id: 750, prediction: 5, probability: 0.8882462382316589]\n",
            "UserItemPrediction [user_id: 4904, item_id: 1198, prediction: 5, probability: 0.8594017028808594]\n",
            "UserItemPrediction [user_id: 4904, item_id: 912, prediction: 5, probability: 0.8534067869186401]\n",
            "UserItemPrediction [user_id: 1084, item_id: 50, prediction: 5, probability: 0.7139193415641785]\n",
            "UserItemPrediction [user_id: 1084, item_id: 2324, prediction: 5, probability: 0.6326656937599182]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kOHBbcDAaH8K"
      },
      "source": [
        "### Recommend 3 users for each item given candidates in the feature RDDs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CLp29k3kaHB-",
        "outputId": "ba1442fc-6485-4871-952a-e4c6195f1855"
      },
      "source": [
        "itemRecs = ncf.recommend_for_item(valPairFeatureRdds, 3)\n",
        "for result in itemRecs.take(5): print(result)"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "UserItemPrediction [user_id: 2909, item_id: 384, prediction: 4, probability: 0.40346845984458923]\n",
            "UserItemPrediction [user_id: 1409, item_id: 384, prediction: 3, probability: 0.4329181909561157]\n",
            "UserItemPrediction [user_id: 4789, item_id: 384, prediction: 3, probability: 0.41791829466819763]\n",
            "UserItemPrediction [user_id: 1940, item_id: 1084, prediction: 5, probability: 0.8245269060134888]\n",
            "UserItemPrediction [user_id: 2497, item_id: 1084, prediction: 5, probability: 0.7757750749588013]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FRJoSBJGaO43"
      },
      "source": [
        "## Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "t8AOXDp9aSwD"
      },
      "source": [
        "Plot the train and validation loss curves"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 407
        },
        "id": "MbS3TGZbaKva",
        "outputId": "344d1ab3-415e-4c82-a2a5-3b17988ee2f5"
      },
      "source": [
        "#retrieve train and validation summary object and read the loss data into ndarray's. \n",
        "train_loss = np.array(ncf.get_train_summary(\"Loss\"))\n",
        "val_loss = np.array(ncf.get_validation_summary(\"Loss\"))\n",
        "#plot the train and validation curves\n",
        "# each event data is a tuple in form of (iteration_count, value, timestamp)\n",
        "plt.figure(figsize = (12,6))\n",
        "plt.plot(train_loss[:,0],train_loss[:,1],label='train loss')\n",
        "plt.plot(val_loss[:,0],val_loss[:,1],label='val loss',color='green')\n",
        "plt.scatter(val_loss[:,0],val_loss[:,1],color='green')\n",
        "plt.legend();\n",
        "plt.xlim(0,train_loss.shape[0]+10)\n",
        "plt.grid(True)\n",
        "plt.title(\"loss\")"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 1.0, 'loss')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-Yh804RoaW8M"
      },
      "source": [
        "Plot accuracy"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 390
        },
        "id": "F0JtfjEMaVLc",
        "outputId": "de3e140b-b46c-4e32-ca39-680be11f689c"
      },
      "source": [
        "plt.figure(figsize = (12,6))\n",
        "top1 = np.array(ncf.get_validation_summary(\"Top1Accuracy\"))\n",
        "plt.plot(top1[:,0],top1[:,1],label='top1')\n",
        "plt.title(\"top1 accuracy\")\n",
        "plt.grid(True)\n",
        "plt.legend();"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}