ML-para-Data-Science/Notebooks/Clase1-extension.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "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": "3.6.8"
    },
    "colab": {
      "name": "MNIST.ipynb",
      "provenance": []
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mv90CIC47YfR"
      },
      "source": [
        "## Introducción a Sklearn 🧠 (Extensión Skorch)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LCI0Yx2z7Yfk"
      },
      "source": [
        "Este cuaderno explica cómo definir y entrenar una red neuronal simple con PyTorch y usarla a través de skorch con SciKit-Learn.\n",
        "\n",
        "\n",
        "**Nota**: Si está ejecutando esto en [un cuaderno Colab](https://colab.research.google.com/github/AcecomFCUNI/Topicos-IA/blob/master/ML-para-Data-Science/Notebooks/Clase1-extension.ipynb), le recomendamos habilite una GPU gratuita yendo a:\n",
        "\n",
        "> **Runtime**   →   **Cambiar el tipo de runtime**   →   **Acerlación por hardware: GPU**\n",
        "\n",
        "Si está ejecutando en colab, debe instalar las dependencias y descargar el conjunto de datos ejecutando la siguiente celda:\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V3dS6zy17Yfn"
      },
      "source": [
        "! [ ! -z \"$COLAB_GPU\" ] && pip install torch scikit-learn==0.20.* skorch"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "!pip install scikit-learn==0.20.* skorch"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "e1dDONjy7Yfp"
      },
      "source": [
        "from sklearn.datasets import fetch_openml\n",
        "from sklearn.model_selection import train_test_split\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wLQ1uMaA7Yfq"
      },
      "source": [
        "## Cargando datos\n",
        "Usando SciKit-Learns ```fetch_openml``` para cargar datos MNIST."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XhjQ9Er37Yfr"
      },
      "source": [
        "mnist = fetch_openml('mnist_784', cache=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nNV2TOhg7Yfs",
        "outputId": "b0663593-7be2-4ef3-b765-95ec96aef4cb"
      },
      "source": [
        "mnist.data.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(70000, 784)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_SV9WFXC7Yfu"
      },
      "source": [
        "## Pre-procesamiento de los datos\n",
        "\n",
        "Cada imagen del conjunto de datos MNIST está codificada en un vector de 784 dimensiones, que representa una imagen de 28 x 28 píxeles. Cada píxel tiene un valor entre 0 y 255, que corresponde al valor de gris de un píxel. <br />\n",
        "El método anterior ```featch_mldata``` para cargar MNIST devuelve ``` data``` y ```target``` como ``` uint8``` que convertimos a ```float32``` y ```int64``` respectivamente."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FCtyzPJz7Yfw"
      },
      "source": [
        "X = mnist.data.astype('float32')\n",
        "y = mnist.target.astype('int64')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CKW0pmV47Yfx"
      },
      "source": [
        "Para evitar grandes pesos que se ocupan de los valores de píxeles entre [0, 255], escalamos \"X\" hacia abajo. Un rango de uso común es [0, 1]."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0OGksD8F7Yfy"
      },
      "source": [
        "X /= 255.0"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gPRGE_Je7Yfz",
        "outputId": "afd9b4a0-cf81-4541-beda-5b21b47b0e3e"
      },
      "source": [
        "X.min(), X.max()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(0.0, 1.0)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vE8izbP37Yfz"
      },
      "source": [
        "Nota: los datos no están normalizados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jwotAkbO7Yf0"
      },
      "source": [
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=42)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FaSQhqJ77Yf1"
      },
      "source": [
        "assert(X_train.shape[0] + X_test.shape[0] == mnist.data.shape[0])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XCPn3gNk7Yf1",
        "outputId": "acc1a806-703d-4284-810d-c79eafecdd23"
      },
      "source": [
        "X_train.shape, y_train.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((52500, 784), (52500,))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JJp2Z5nP7Yf2"
      },
      "source": [
        "### Imprimiendo una selección de imágenes de entrenamiento y sus etiquetas"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bToSd7M97Yf3"
      },
      "source": [
        "def plot_example(X, y):\n",
        "    \"\"\"Ploteamos las primeras 5 imágenes y sus etiquetas en una fila..\"\"\"\n",
        "    for i, (img, y) in enumerate(zip(X[:5].reshape(5, 28, 28), y[:5])):\n",
        "        plt.subplot(151 + i)\n",
        "        plt.imshow(img)\n",
        "        plt.xticks([])\n",
        "        plt.yticks([])\n",
        "        plt.title(y)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K0tvyuzt7Yf4",
        "outputId": "20f6dd84-2bad-42b9-82fa-0604679fd1a8"
      },
      "source": [
        "plot_example(X_train, y_train)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7ff569a4a320>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Zbfe_SrN7Yf4"
      },
      "source": [
        "## Construyendo una red neuronal con PyTorch\n",
        "Es una red neuronal simple y completamente conectada con una capa oculta. La capa de entrada tiene 784 dimensiones (28x28), la capa oculta tiene 98 (= 784/8) unidades y la capa de salida 10 neuronas, que representan los dígitos 0 - 9."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8KW5p6Ou7Yf5"
      },
      "source": [
        "import torch\n",
        "from torch import nn\n",
        "import torch.nn.functional as F"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FBq_mO3O7Yf6"
      },
      "source": [
        "device = 'cuda' if torch.cuda.is_available() else 'cpu'"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S7v6a2v_7Yf6"
      },
      "source": [
        "mnist_dim = X.shape[1]\n",
        "hidden_dim = int(mnist_dim/8)\n",
        "output_dim = len(np.unique(mnist.target))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Off4-84N7Yf7",
        "outputId": "fb2369a2-da89-4eae-c124-2de86c6af49c"
      },
      "source": [
        "mnist_dim, hidden_dim, output_dim"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(784, 98, 10)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FazMk76H7Yf7"
      },
      "source": [
        "Una red neuronal en el marco de PyTorch."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AJtcxc_y7Yf8"
      },
      "source": [
        "class ClassifierModule(nn.Module):\n",
        "    def __init__(\n",
        "            self,\n",
        "            input_dim=mnist_dim,\n",
        "            hidden_dim=hidden_dim,\n",
        "            output_dim=output_dim,\n",
        "            dropout=0.5,\n",
        "    ):\n",
        "        super(ClassifierModule, self).__init__()\n",
        "        self.dropout = nn.Dropout(dropout)\n",
        "\n",
        "        self.hidden = nn.Linear(input_dim, hidden_dim)\n",
        "        self.output = nn.Linear(hidden_dim, output_dim)\n",
        "\n",
        "    def forward(self, X, **kwargs):\n",
        "        X = F.relu(self.hidden(X))\n",
        "        X = self.dropout(X)\n",
        "        X = F.softmax(self.output(X), dim=-1)\n",
        "        return X"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IZVXL4TI7Yf9"
      },
      "source": [
        "*Skorch* permite usar las redes de PyTorch en la configuración de SciKit-Learn:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bg9MP9587Yf9"
      },
      "source": [
        "from skorch import NeuralNetClassifier"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1X3VVUB47Yf-"
      },
      "source": [
        "torch.manual_seed(0)\n",
        "\n",
        "net = NeuralNetClassifier(\n",
        "    ClassifierModule,\n",
        "    max_epochs=20,\n",
        "    lr=0.1,\n",
        "    device=device,\n",
        ")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "26cTdzLF7Yf-",
        "outputId": "57508dea-bf80-49e5-ab0b-86879eafc4ff"
      },
      "source": [
        "net.fit(X_train, y_train);"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  epoch    train_loss    valid_acc    valid_loss     dur\n",
            "-------  ------------  -----------  ------------  ------\n",
            "      1        \u001b[36m0.8299\u001b[0m       \u001b[32m0.8893\u001b[0m        \u001b[35m0.4037\u001b[0m  0.8951\n",
            "      2        \u001b[36m0.4331\u001b[0m       \u001b[32m0.9113\u001b[0m        \u001b[35m0.3075\u001b[0m  0.7862\n",
            "      3        \u001b[36m0.3619\u001b[0m       \u001b[32m0.9240\u001b[0m        \u001b[35m0.2614\u001b[0m  0.8314\n",
            "      4        \u001b[36m0.3237\u001b[0m       \u001b[32m0.9305\u001b[0m        \u001b[35m0.2379\u001b[0m  0.7739\n",
            "      5        \u001b[36m0.2914\u001b[0m       \u001b[32m0.9371\u001b[0m        \u001b[35m0.2173\u001b[0m  0.7721\n",
            "      6        \u001b[36m0.2739\u001b[0m       \u001b[32m0.9413\u001b[0m        \u001b[35m0.1979\u001b[0m  0.7949\n",
            "      7        \u001b[36m0.2569\u001b[0m       \u001b[32m0.9449\u001b[0m        \u001b[35m0.1859\u001b[0m  0.7691\n",
            "      8        \u001b[36m0.2420\u001b[0m       \u001b[32m0.9461\u001b[0m        \u001b[35m0.1813\u001b[0m  0.7871\n",
            "      9        \u001b[36m0.2337\u001b[0m       \u001b[32m0.9496\u001b[0m        \u001b[35m0.1708\u001b[0m  0.7730\n",
            "     10        \u001b[36m0.2195\u001b[0m       \u001b[32m0.9532\u001b[0m        \u001b[35m0.1604\u001b[0m  0.7992\n",
            "     11        \u001b[36m0.2151\u001b[0m       \u001b[32m0.9547\u001b[0m        \u001b[35m0.1514\u001b[0m  0.7955\n",
            "     12        \u001b[36m0.2065\u001b[0m       \u001b[32m0.9560\u001b[0m        \u001b[35m0.1476\u001b[0m  0.7734\n",
            "     13        \u001b[36m0.2015\u001b[0m       \u001b[32m0.9563\u001b[0m        \u001b[35m0.1455\u001b[0m  0.8316\n",
            "     14        \u001b[36m0.1943\u001b[0m       \u001b[32m0.9587\u001b[0m        \u001b[35m0.1389\u001b[0m  0.7913\n",
            "     15        \u001b[36m0.1883\u001b[0m       0.9578        \u001b[35m0.1381\u001b[0m  0.8092\n",
            "     16        \u001b[36m0.1848\u001b[0m       \u001b[32m0.9596\u001b[0m        \u001b[35m0.1323\u001b[0m  0.7907\n",
            "     17        \u001b[36m0.1838\u001b[0m       \u001b[32m0.9606\u001b[0m        \u001b[35m0.1312\u001b[0m  0.7806\n",
            "     18        \u001b[36m0.1776\u001b[0m       \u001b[32m0.9623\u001b[0m        \u001b[35m0.1261\u001b[0m  0.7657\n",
            "     19        \u001b[36m0.1738\u001b[0m       \u001b[32m0.9625\u001b[0m        \u001b[35m0.1250\u001b[0m  0.7690\n",
            "     20        \u001b[36m0.1704\u001b[0m       \u001b[32m0.9627\u001b[0m        \u001b[35m0.1238\u001b[0m  0.7627\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z0xjiLJz7Yf_"
      },
      "source": [
        "## Predicción"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k894bnew7YgA"
      },
      "source": [
        "from sklearn.metrics import accuracy_score"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SkOhZY8d7YgA"
      },
      "source": [
        "y_pred = net.predict(X_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "emlRJwZ77YgB",
        "outputId": "eb84c771-bb9e-489d-d121-b00d57d4ae77"
      },
      "source": [
        "accuracy_score(y_test, y_pred)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9627428571428571"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Uk5gufEY7YgC"
      },
      "source": [
        "Un accuracy de aproximadamente el 96% para una red con una sola capa oculta no es tan mala.\n",
        "\n",
        "Echemos un vistazo a algunas predicciones que salieron mal:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3G5xhp1Z7YgC"
      },
      "source": [
        "error_mask = y_pred != y_test"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S8VelOjp7YgD",
        "outputId": "a29fbaeb-869e-475d-d118-a9ed9f10668a"
      },
      "source": [
        "plot_example(X_test[error_mask], y_pred[error_mask])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7ff5140a62b0>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}