03_MLDeepChem/MNIST_with_TensorGraph_from_DeepChem.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "MNIST with TensorGraph from DeepChem.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyMclceB2TtREZNQ3b7hpRZU",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/dbetm/DeepLearningLifeSciences/blob/main/03_MLDeepChem/MNIST_with_TensorGraph_from_DeepChem.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E3sQF23y4kVp"
      },
      "source": [
        "DeepChem uses the *TensorGraph* class to construct nonstandard deep learning architectures.\n",
        "\n",
        "We'll construct a convolutional architecture shown below:\n",
        "\n",
        "![cnn.png](data:image/png;base64,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)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yAZ2VMER4T-w"
      },
      "source": [
        "# SETUP DEEPCHEM\n",
        "!pip install --pre deepchem\n",
        "import deepchem\n",
        "deepchem.__version__"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i9Lc5HSv5wn6"
      },
      "source": [
        "## Loading dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Lu-sA3or6Ral"
      },
      "source": [
        "import deepchem as dc\n",
        "import deepchem.models.layers as layers\n",
        "import numpy as np\n",
        "from keras.utils.np_utils import to_categorical\n",
        "import tensorflow as tf\n",
        "import tensorflow_datasets as tfds\n",
        "from tensorflow.keras.layers import Conv2D, Dense, Flatten, Reshape, Softmax"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qcgGODxyhm9M",
        "outputId": "4beb06ab-58ac-4bcc-aa32-d1ce0470a57e"
      },
      "source": [
        "mnist = tf.keras.datasets.mnist\n",
        "(images_train, labels_train), (images_test, labels_test) = mnist.load_data()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3_tdeRKiimTy"
      },
      "source": [
        "# cast dtype for images\n",
        "images_train = images_train.astype('float32')\n",
        "images_test = images_test.astype('float32')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iN7tf9lOVvmF"
      },
      "source": [
        "# Tranforms labels like one-hot vectors\n",
        "num_classes = 10\n",
        "labels_train = to_categorical(labels_train, num_classes=num_classes)\n",
        "labels_test = to_categorical(labels_test, num_classes=num_classes)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IQkGDBiMik51",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "84433998-09d4-402a-d174-fd87b9992e4d"
      },
      "source": [
        "train_dataset = dc.data.NumpyDataset(images_train, labels_train)\n",
        "test_dataset = dc.data.NumpyDataset(images_test, labels_test)\n",
        "print(train_dataset)\n",
        "print(test_dataset)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<NumpyDataset X.shape: (60000, 28, 28), y.shape: (60000, 10), w.shape: (60000, 1), task_names: [0 1 2 3 4 5 6 7 8 9]>\n",
            "<NumpyDataset X.shape: (10000, 28, 28), y.shape: (10000, 10), w.shape: (10000, 1), task_names: [0 1 2 3 4 5 6 7 8 9]>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VqFA61XllXvB",
        "outputId": "e25e77ff-efd4-40b9-fdb0-001d5b84acb3"
      },
      "source": [
        "print(train_dataset.X.shape[1],\"x\", train_dataset.X.shape[2], \"=\", train_dataset.X.shape[1] * train_dataset.X.shape[2])\n",
        "num_features = train_dataset.X.shape[1] * train_dataset.X.shape[2]\n",
        "num_tasks = 10"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "28 x 28 = 784\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Cm6Mp4wZkWOf"
      },
      "source": [
        "## Defining architecture of the Convolutional Neural Network"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kepIOiaxkwbP"
      },
      "source": [
        "# https://notebook.community/deepchem/deepchem/examples/tutorials/02_Learning_MNIST_Digit_Classifiers\n",
        "keras_model = tf.keras.Sequential(layers=[\n",
        "  Reshape((28, 28, 1)),\n",
        "  Conv2D(filters=32, kernel_size=5, activation=tf.nn.relu),\n",
        "  Conv2D(filters=64, kernel_size=5, activation=tf.nn.relu),\n",
        "  Flatten(),\n",
        "  Dense(1024, activation=tf.nn.relu),\n",
        "  Dense(10),\n",
        "  Softmax()\n",
        "])\n",
        "\n",
        "model = dc.models.KerasModel(model=keras_model, loss=dc.models.losses.CategoricalCrossEntropy())"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JwVgEFjJN1CN"
      },
      "source": [
        "## Training\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2fZe_B9XN3kh",
        "outputId": "3b485757-e7e0-4e13-946d-afe79a91f26d"
      },
      "source": [
        "model.fit(dataset=train_dataset, nb_epoch=10)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.016151312589645386"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Uh05P9wEOWyV"
      },
      "source": [
        "## Measuring the performance"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Msnh2eqJYf4P",
        "outputId": "8b6b0559-3051-4aa8-96b9-8996b6d754ec"
      },
      "source": [
        "metric = dc.metrics.Metric(dc.metrics.accuracy_score)\n",
        "# Accuracy over train and test datasets\n",
        "train_scores = model.evaluate(train_dataset, [metric])\n",
        "print(\"TRAIN SCORES\", train_scores)\n",
        "test_scores = model.evaluate(test_dataset, [metric])\n",
        "print(\"TEST SCORES\", test_scores)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "TRAIN SCORES {'accuracy_score': 0.9991583333333335}\n",
            "TEST SCORES {'accuracy_score': 0.9957800000000001}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gbHP53QmyKzf"
      },
      "source": [
        "## TESTING"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6_8PT_BPyMOv",
        "outputId": "37e667f1-400a-4b3d-8977-c2ccfde98b06"
      },
      "source": [
        "random_index = np.random.randint(0, len(images_test))\n",
        "print(tf.argmax(labels_test[random_index], axis=0))\n",
        "print(labels_test[random_index])\n",
        "\n",
        "prediction = model.predict_on_batch([test_dataset.X[random_index]])\n",
        "\n",
        "prediction = np.squeeze(prediction)\n",
        "\n",
        "categorical_prediction = tf.nn.softmax(prediction).numpy()\n",
        "print(categorical_prediction)\n",
        "\n",
        "categorical_prediction = np.squeeze(categorical_prediction)\n",
        "\n",
        "index = -1\n",
        "max_output = 0.0\n",
        "for i in range(len(categorical_prediction)):\n",
        "  if categorical_prediction[i] > max_output:\n",
        "    max_output = categorical_prediction[i]\n",
        "    index = i\n",
        "\n",
        "\n",
        "print(\"PREDICTION {} with prob:{} \".format(index, categorical_prediction[index]))\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tf.Tensor(8, shape=(), dtype=int64)\n",
            "[0. 0. 0. 0. 0. 0. 0. 0. 1. 0.]\n",
            "[0.08533674 0.08533674 0.08533674 0.08533674 0.08533674 0.08533674\n",
            " 0.08533674 0.08533674 0.2319693  0.08533674]\n",
            "PREDICTION 8 with prob:0.23196929693222046 \n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}