IMDB_Keras.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "IMDB_Keras.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/RubenMcCarty/Machine-Learning-con-Python-001/blob/master/IMDB_Keras.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SZ1YMPZvFRpb"
      },
      "source": [
        "# **IMDB dataset** \n",
        "# Algoritmo para predecir las criticas positivas y negativas de películas de IMDB\n",
        "##  Creado por [M.Sc. Ruben Quispe](https://machinelearningmastery.com/tutorial-to-implement-k-nearest-neighbors-in-python-from-scratch/)\n",
        "### Curso [Machine Learning](https://eie.pe/curso-machine-learning/)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SyxCte0Lz-qg"
      },
      "source": [
        "La base de datos IMDB (Internet Movie Database), se compone de 50,000 críticas de películas, donde 25,000 críticas conforman el conjunto de entrenamiento y 25,000 constituyen el conjunto de prueba. En cada conjunto, el 50% de las críticas son positivas y el restante 50% son negativas. Sitio web de IMDB: https://www.imdb.com/<br>\n",
        "La base de datos IMDB viene precargada en Keras:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S5k_oKrxz4tm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d98a7ee7-7995-46e4-a172-fb275595ba80"
      },
      "source": [
        "from tensorflow.keras.datasets import imdb\n",
        "\n",
        "(train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words=10000)"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb.npz\n",
            "17465344/17464789 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Tw1ZfnBc0wsg"
      },
      "source": [
        "El argumento num_word=10000, indica que sólo se tomarán en cuenta las 10,000 palabras más usuales dentro de la base de datos, esto con la intención de tener instancias de un tamaño más manejable."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0eDKjxwA1S3v"
      },
      "source": [
        "train_data y test_data son listas de críticas, donde cada crítica se compone de valores enteros, los cuales son el índice de alguna palabra en particular. Por su parte, train_labels y test_labels, son listas de 0's y 1's, donde 0 representa que la crítica es negativa y 1 que es positiva."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UxKz5kF514ng",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0f5f2a57-5091-4c3b-c3e2-263ed5042dc5"
      },
      "source": [
        "train_data[20000]"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[1,\n",
              " 6,\n",
              " 185,\n",
              " 255,\n",
              " 2478,\n",
              " 39,\n",
              " 4,\n",
              " 2,\n",
              " 15,\n",
              " 59,\n",
              " 9,\n",
              " 6,\n",
              " 1985,\n",
              " 448,\n",
              " 725,\n",
              " 6,\n",
              " 223,\n",
              " 1795,\n",
              " 7,\n",
              " 35,\n",
              " 2,\n",
              " 3612,\n",
              " 7,\n",
              " 5,\n",
              " 556,\n",
              " 18,\n",
              " 9781,\n",
              " 112,\n",
              " 31,\n",
              " 688,\n",
              " 8,\n",
              " 41,\n",
              " 501,\n",
              " 2199,\n",
              " 34,\n",
              " 349,\n",
              " 59,\n",
              " 4136,\n",
              " 4,\n",
              " 4281,\n",
              " 2,\n",
              " 5,\n",
              " 880,\n",
              " 4,\n",
              " 349,\n",
              " 59,\n",
              " 892,\n",
              " 1454,\n",
              " 11,\n",
              " 119,\n",
              " 19,\n",
              " 6,\n",
              " 243,\n",
              " 132,\n",
              " 41,\n",
              " 113,\n",
              " 739,\n",
              " 8,\n",
              " 193,\n",
              " 6,\n",
              " 471,\n",
              " 18,\n",
              " 4,\n",
              " 128,\n",
              " 54,\n",
              " 59,\n",
              " 9,\n",
              " 3540,\n",
              " 5,\n",
              " 41,\n",
              " 1461,\n",
              " 9,\n",
              " 556,\n",
              " 34,\n",
              " 6,\n",
              " 1143,\n",
              " 7,\n",
              " 3959,\n",
              " 9266,\n",
              " 174,\n",
              " 34,\n",
              " 134,\n",
              " 2484,\n",
              " 687,\n",
              " 4,\n",
              " 255,\n",
              " 1747,\n",
              " 8,\n",
              " 41,\n",
              " 1115,\n",
              " 771,\n",
              " 5,\n",
              " 4846,\n",
              " 1060,\n",
              " 23,\n",
              " 4,\n",
              " 3959,\n",
              " 1793,\n",
              " 8,\n",
              " 4,\n",
              " 288,\n",
              " 2,\n",
              " 3935,\n",
              " 10,\n",
              " 10,\n",
              " 2,\n",
              " 2,\n",
              " 2,\n",
              " 1080,\n",
              " 2,\n",
              " 2,\n",
              " 778,\n",
              " 19,\n",
              " 368,\n",
              " 6073,\n",
              " 2,\n",
              " 2,\n",
              " 837,\n",
              " 34,\n",
              " 2,\n",
              " 1976,\n",
              " 2,\n",
              " 2,\n",
              " 37,\n",
              " 17,\n",
              " 2,\n",
              " 2,\n",
              " 2,\n",
              " 2251,\n",
              " 59,\n",
              " 9,\n",
              " 6,\n",
              " 1985,\n",
              " 4,\n",
              " 2,\n",
              " 173,\n",
              " 9,\n",
              " 54,\n",
              " 4,\n",
              " 370,\n",
              " 4110,\n",
              " 978,\n",
              " 968,\n",
              " 200,\n",
              " 41,\n",
              " 2977,\n",
              " 4,\n",
              " 636,\n",
              " 2763,\n",
              " 2672,\n",
              " 41,\n",
              " 8557,\n",
              " 2,\n",
              " 16,\n",
              " 242,\n",
              " 6,\n",
              " 2,\n",
              " 1664,\n",
              " 1563,\n",
              " 2,\n",
              " 9,\n",
              " 472,\n",
              " 24,\n",
              " 6,\n",
              " 1985,\n",
              " 59,\n",
              " 9,\n",
              " 1533,\n",
              " 6,\n",
              " 55,\n",
              " 565,\n",
              " 7472,\n",
              " 10,\n",
              " 10,\n",
              " 17,\n",
              " 6,\n",
              " 22,\n",
              " 1985,\n",
              " 255,\n",
              " 11,\n",
              " 631,\n",
              " 62,\n",
              " 28,\n",
              " 77,\n",
              " 128,\n",
              " 48,\n",
              " 2,\n",
              " 109,\n",
              " 66,\n",
              " 16,\n",
              " 6,\n",
              " 668,\n",
              " 1985,\n",
              " 19,\n",
              " 41,\n",
              " 861,\n",
              " 2,\n",
              " 6,\n",
              " 87,\n",
              " 227,\n",
              " 7,\n",
              " 3588,\n",
              " 21,\n",
              " 17,\n",
              " 230,\n",
              " 17,\n",
              " 3469,\n",
              " 2391,\n",
              " 5,\n",
              " 541,\n",
              " 1554,\n",
              " 140,\n",
              " 14,\n",
              " 31,\n",
              " 9,\n",
              " 254,\n",
              " 8,\n",
              " 1559,\n",
              " 2,\n",
              " 2,\n",
              " 9,\n",
              " 345,\n",
              " 2516,\n",
              " 42,\n",
              " 2,\n",
              " 5758,\n",
              " 469,\n",
              " 4,\n",
              " 22,\n",
              " 63,\n",
              " 944,\n",
              " 6,\n",
              " 1257,\n",
              " 1166,\n",
              " 7,\n",
              " 2172,\n",
              " 599,\n",
              " 6203,\n",
              " 2,\n",
              " 2,\n",
              " 17,\n",
              " 2,\n",
              " 5,\n",
              " 8548,\n",
              " 2234,\n",
              " 5252,\n",
              " 17,\n",
              " 2,\n",
              " 26,\n",
              " 52,\n",
              " 696,\n",
              " 1850]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fsC8CcZt1-Ye",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "73408748-d8d6-4237-dcd5-0cba8d93840c"
      },
      "source": [
        "len(train_data[1786]) "
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "113"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "429IzTsz202o",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5b8eb66d-1a87-40b6-da87-f731be8e7372"
      },
      "source": [
        "train_labels[3] #La crítica 51 del conjunto de entrenamiento es negativa."
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Cv3Q6aDZ23ye"
      },
      "source": [
        "Dado que establecimos la restricción de utilizar solamente las 10,000 palabras más usuales, ninguno de los índices que constituyen cada crítica, puede tener un valor mayor a 10,000:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YqFEgGb13RX4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1b11326a-17a1-458e-c859-c61aeb404e45"
      },
      "source": [
        "max([max(sequence) for sequence in train_data])"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "9999"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XUEKRBde3XBF"
      },
      "source": [
        "El código siguiente permite decodificar las críticas:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FwUNXsRG3bJ_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "470a4ad1-db94-4737-dd77-7502ada03cc2"
      },
      "source": [
        "#word_index es un diccionario que permite mapear cada palabra a su respectivo índice\n",
        "word_index = imdb.get_word_index()\n",
        "# Aplicamos la operación inversa, es decir, mapeamos de índices a palabras.\n",
        "reverse_word_index = dict([(value, key) for (key, value) in word_index.items()]) #dict significa dictionary, y te devuelve una lista de tuplas pares, donde\n",
        "#el primer valor de la lista, representa la clave o ID, y el segundo valor denota el valor de dicha clave.\n",
        "# Procedemos a decodificar la crítica. La razón por la que a cada índice le restamos 3, es\n",
        "#porque 0, 1 y 2 son índices reservados para los valores de \"padding\", \"comienzo de la secuencia\" y \"desconocido\".\n",
        "# El argumento \"?\" se regresa en caso de que el índice en cuestión no tenga asociado ninguna palabra.\n",
        "decoded_review = ' '.join([reverse_word_index.get(i - 3, '?') for i in train_data[15678]])"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb_word_index.json\n",
            "1646592/1641221 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jPJAR1XV6Aax",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 137
        },
        "outputId": "e45b9e67-3313-4d36-c5b5-4a65223d9cdf"
      },
      "source": [
        "decoded_review"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "\"? as much as i have enjoyed the hanzo the razor movies three is definitely enough ? got the gold ' the final adventure for the japanese ? with the impressive package is a fairly enjoyable piece of ? cinema but offers little new in terms of ideas whilst taking a big step backwards as far as ? is concerned br br the film opens with the appearance of a female ghost and looks as though it is going to explore supernatural territory something which might have taken the series in an interesting new direction unfortunately after the ? turns out to be nothing but a scooby doo style ? ? up by a corrupt ? official keen to keep people away from the lake where he is hiding stolen gold director ? ? is content to ? familiar elements from the first two films the result being a rather stale affair br br once again hanzo heads an investigation that requires him to ? women through the use of his mighty penis slice up his enemies and abuse his superiors on the way we get wild ? good natured rape hanzo forces himself on women who wind up ? his ? and bloody sword fights br br if you've already seen and appreciated the first two films you might as well watch this ? to complete the set but be warned this is probably the least satisfying one of them all br br 6 5 out of 10 rounded up to 7 for imdb\""
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Al09obtH6SZg"
      },
      "source": [
        "# **Preprocesamiento de los datos**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UQHydxeT6Vru"
      },
      "source": [
        "Aplicamos el enfoque de one-hot encoding a los datos de entrada. Por ejemplo, si una de nuestras instancias se representa por la lista [2, 7, 15], al aplicar one-hot encoding, dicha lista se convertirá en un vector de 10,000 elementos, donde a los índices 2, 7 y 15, se les asignará el valor de 1. En el problema que estamos abordando, todas las listas de críticas se transforman en vectores de 10,000 elementos, debido a que el número máximo de palabras que se están considerando, son 10,000. <br>\n",
        "La razón por la que se aplica esta operación en los datos de entrada, es porque las redes neuronales convolucionales no reciben como entrada listas de valores, sino tensores.<br>\n",
        "Ejemplo:\n",
        "\n",
        "**Crítica -> [2, 7, 15]** <br>\n",
        "**One-hot encoding de la crítica -> [0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, ... , 0]** <br>\n",
        "**Crítica -> [9, 7, 1, 2, 7]** <br>\n",
        "**One-hot encoding de la crítica -> [0, 1, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, ... , 0]**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "24fugxvF6Us9"
      },
      "source": [
        "import numpy as np\n",
        "\n",
        "def vectorize_sequences(sequences, dimension=10000): #sequences se refiere al conjunto de críticas\n",
        "    # Creamos matrices de ceros con dimensiones (len(sequences), dimension)\n",
        "    results = np.zeros((len(sequences), dimension)) #-> 25,000 X 10,000\n",
        "    for i, sequence in enumerate(sequences):\n",
        "        results[i, sequence] = 1.  # le asignamos el valor de uno al índice correspondiente\n",
        "    return results\n",
        "\n",
        "# Conjunto de entrenamiento vectorizado/one-hote encoding\n",
        "x_train = vectorize_sequences(train_data)\n",
        "# Conjunto de prueba vectorizado/one-hote encoding\n",
        "x_test = vectorize_sequences(test_data)"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TqF6gZrIUGBF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1dc4c880-eceb-46b0-b713-7a01f4b8275b"
      },
      "source": [
        "x_train[23976]"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([0., 1., 1., ..., 0., 0., 0.])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2g672_NBT82z",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ca47f734-5433-4b8b-81c5-ee7342b0c6c5"
      },
      "source": [
        "len(x_train[12])"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "10000"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lbNkuupnT8Uc",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f0549b0d-fc9e-48cd-95c6-60b389bd7ca6"
      },
      "source": [
        "x_train[1].shape"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(10000,)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YHugKBJ4UAd5",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ec376567-bdd4-49dc-8e42-da12f088785a"
      },
      "source": [
        "x_train.shape"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(25000, 10000)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VfJ9c6pd-GZn",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "fa4719c8-8e83-41e0-c4b4-7957a2bc12e4"
      },
      "source": [
        "x_train[15879]"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([0., 1., 1., ..., 0., 0., 0.])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eN8-njVs-ykK"
      },
      "source": [
        "También convertimos las etiquetas en tensores:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CpP5gJyI-1k9"
      },
      "source": [
        "# Etiquetas vectorizadas\n",
        "y_train = np.asarray(train_labels).astype('float32')\n",
        "y_test = np.asarray(test_labels).astype('float32')"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vFw005TuVIIB",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6b1a90e2-6dc8-4dd1-e9f7-440f2b8f885d"
      },
      "source": [
        "y_train[9]"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tyigcFvc_vI3"
      },
      "source": [
        "# **Construcción de la red neuronal**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z8Yn31s-_xpN"
      },
      "source": [
        "from tensorflow.keras import models\n",
        "from tensorflow.keras import layers\n",
        "\n",
        "model = models.Sequential()\n",
        "model.add(layers.Dense(16, activation='relu', input_shape=(10000,))) #Capa oculta 1 - input_shape puede verse como una representación implícita de la capa de entrada\n",
        "model.add(layers.Dense(16, activation='relu'))# Capa oculta 2\n",
        "model.add(layers.Dense(1, activation='sigmoid'))#Capa de salida"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "g_BA0lc-iuPY",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a9fb8474-ef4c-4d3e-e312-1282b0766ae2"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense (Dense)                (None, 16)                160016    \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 16)                272       \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 17        \n",
            "=================================================================\n",
            "Total params: 160,305\n",
            "Trainable params: 160,305\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xY04xdUYAFeE"
      },
      "source": [
        "# **Compilación**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8_T3TRHTAEIV"
      },
      "source": [
        "from tensorflow.keras import optimizers\n",
        "model.compile(optimizer=optimizers.RMSprop(lr=0.001), #Es una variación del descenso de gradiente\n",
        "              loss='binary_crossentropy', #Se relaciona a la función de error \n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k35EAUeHAtc_"
      },
      "source": [
        "# **Validación**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9NBURPE3Axt2"
      },
      "source": [
        "Buscando evaluar el rendimiento del modelo durante la fase de entrenamiento, construimos el conjunto de validación a partir del conjunto de entrenamiento."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hEWPTE0eAwIF"
      },
      "source": [
        "#Instancias\n",
        "x_val = x_train[:10000] #Instancias de validación\n",
        "#x_val = x_train[0:10000]\n",
        "partial_x_train = x_train[10000:] #Instancias de entrenamiento\n",
        "#partial_x_train = x_train[10000:25000] #Instancias de entrenamiento\n",
        "\n",
        "#Etiquetas\n",
        "y_val = y_train[:10000] #Etiquetas de validación  \n",
        "#y_val = y_train[0:10000] #Etiquetas de validación  \n",
        "partial_y_train = y_train[10000:] #Etiquetas de entrenamiento\n",
        "#partial_y_train = y_train[10000:25000] #Etiquetas de entrenamiento"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8jdBLQB9Yejh",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5f653c50-bea6-4377-f578-80ead725a8c4"
      },
      "source": [
        "#Conjunto de entrenamiento\n",
        "len(partial_x_train)"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "15000"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x1SoiGZgYjxN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "61782fff-66c6-4114-db33-80c1d04244e8"
      },
      "source": [
        "#Conjunto de validación\n",
        "len(x_val)"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "10000"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gj4P5iKfBN78"
      },
      "source": [
        "# **Entrenamiento**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b8TUQ6oSBQP9",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8c55198b-89ec-473c-b0be-299b6ae20b80"
      },
      "source": [
        "epochs_num = 20\n",
        "history = model.fit(partial_x_train,\n",
        "                    partial_y_train,\n",
        "                    epochs=epochs_num,\n",
        "                    batch_size=512,\n",
        "                    validation_data=(x_val, y_val))"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "30/30 [==============================] - 1s 45ms/step - loss: 0.5109 - accuracy: 0.7839 - val_loss: 0.4084 - val_accuracy: 0.8431\n",
            "Epoch 2/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.2980 - accuracy: 0.9015 - val_loss: 0.3122 - val_accuracy: 0.8800\n",
            "Epoch 3/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.2160 - accuracy: 0.9289 - val_loss: 0.2754 - val_accuracy: 0.8928\n",
            "Epoch 4/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.1709 - accuracy: 0.9445 - val_loss: 0.2776 - val_accuracy: 0.8880\n",
            "Epoch 5/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.1398 - accuracy: 0.9545 - val_loss: 0.2940 - val_accuracy: 0.8841\n",
            "Epoch 6/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.1145 - accuracy: 0.9636 - val_loss: 0.3155 - val_accuracy: 0.8795\n",
            "Epoch 7/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0977 - accuracy: 0.9705 - val_loss: 0.3123 - val_accuracy: 0.8838\n",
            "Epoch 8/20\n",
            "30/30 [==============================] - 1s 33ms/step - loss: 0.0807 - accuracy: 0.9769 - val_loss: 0.3603 - val_accuracy: 0.8745\n",
            "Epoch 9/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0690 - accuracy: 0.9801 - val_loss: 0.3588 - val_accuracy: 0.8816\n",
            "Epoch 10/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0539 - accuracy: 0.9865 - val_loss: 0.3836 - val_accuracy: 0.8777\n",
            "Epoch 11/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0472 - accuracy: 0.9875 - val_loss: 0.4206 - val_accuracy: 0.8761\n",
            "Epoch 12/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0380 - accuracy: 0.9907 - val_loss: 0.4393 - val_accuracy: 0.8750\n",
            "Epoch 13/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0277 - accuracy: 0.9951 - val_loss: 0.5384 - val_accuracy: 0.8565\n",
            "Epoch 14/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0263 - accuracy: 0.9947 - val_loss: 0.5183 - val_accuracy: 0.8661\n",
            "Epoch 15/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0180 - accuracy: 0.9976 - val_loss: 0.5352 - val_accuracy: 0.8696\n",
            "Epoch 16/20\n",
            "30/30 [==============================] - 1s 33ms/step - loss: 0.0171 - accuracy: 0.9973 - val_loss: 0.5732 - val_accuracy: 0.8683\n",
            "Epoch 17/20\n",
            "30/30 [==============================] - 1s 31ms/step - loss: 0.0115 - accuracy: 0.9988 - val_loss: 0.6100 - val_accuracy: 0.8653\n",
            "Epoch 18/20\n",
            "30/30 [==============================] - 1s 30ms/step - loss: 0.0111 - accuracy: 0.9983 - val_loss: 0.6417 - val_accuracy: 0.8671\n",
            "Epoch 19/20\n",
            "30/30 [==============================] - 1s 32ms/step - loss: 0.0055 - accuracy: 0.9998 - val_loss: 0.6722 - val_accuracy: 0.8656\n",
            "Epoch 20/20\n",
            "30/30 [==============================] - 1s 30ms/step - loss: 0.0077 - accuracy: 0.9984 - val_loss: 0.7009 - val_accuracy: 0.8645\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "z7XCBSPuBfP0"
      },
      "source": [
        "Podemos ver que la llamada al método fit(), genera un objeto History, el cual tiene un miembro history, que a su vez es un diccionario que contiene información acerca de todo el proceso de entrenamiento."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZDkkkrIJB1kG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "20ab21c5-2772-4e66-cb2d-dafd46e127b1"
      },
      "source": [
        "history_dict = history.history\n",
        "history_dict.keys()"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U6kAejSpCBFN"
      },
      "source": [
        "Graficamos el desempeño de la red neuronal durante la fase de entrenamiento. A continuación se muestra la comparación entre la pérdida que ocurre en el conjunto de entrenamiento y la pérdida relacionada al conjunto de validación. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FjM7QAztB5it",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 579
        },
        "outputId": "e8a51c0b-c264-4f0e-870a-f446a971614a"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "max_val_acc = max(val_acc)\n",
        "max_val_acc_epoch= val_acc.index(max(val_acc)) + 1\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "min_val_loss = min(val_loss)\n",
        "min_val_loss_epoch= val_loss.index(min(val_loss)) + 1\n",
        "\n",
        "epochs = range(len(acc))\n",
        "\n",
        "#Precisión\n",
        "plt.plot(epochs, acc, 'go', label='Precisión en el entrenamiento')\n",
        "plt.plot(epochs, val_acc, 'r', label='Precisión en la validación')\n",
        "plt.plot(max_val_acc_epoch, max_val_acc, 'bo', label='Maximum accuracy')\n",
        "plt.title('Precisión durante el entrenamiento y la validación')\n",
        "plt.legend()\n",
        "print('Best accuracy epoch ', max_val_acc_epoch)\n",
        "plt.figure()\n",
        "\n",
        "#Pérdida\n",
        "plt.plot(epochs, loss, 'go', label='Pérdida en el entrenamiento')\n",
        "plt.plot(epochs, val_loss, 'r', label='Pérdida en la validación')\n",
        "plt.plot(min_val_loss_epoch, min_val_loss, 'bo', label='Minimum loss')\n",
        "plt.title('Pérdida durante el entrenamiento y la validación')\n",
        "plt.legend()\n",
        "print('Best loss epoch ', min_val_loss_epoch)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Best accuracy epoch  3\n",
            "Best loss epoch  3\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3WnpimJkQoo6"
      },
      "source": [
        "Con base a las gráficas previas, podemos observar que, aproximadamente a partir de la época 4, el modelo deja de mejorar y comienza a presentarse una disminución en su rendimiento."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "T4SLTEP6eoGt"
      },
      "source": [
        "# **Overfitting - Sobreajuste:**<br>\n",
        "Esto ocurre cuando el desempeño del modelo en el conjunto de entrenamiento es superior al rendimiento del mismo modelo en el conjunto de validación.<br>\n",
        "Para mitigar el overfitting, debemos buscar que el rendimiento del modelo tanto en el conjunto de entrenamiento como en el conjunto de validación, sea similar y con valores apropiados."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pQJfOSgVESRE"
      },
      "source": [
        "# **Re-entrenamiento y evaluación del modelo final** "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rFnzQ2F3v6iC"
      },
      "source": [
        "Una vez que se seleccionó el modelo tomando como base el rendimiento de este en el conjunto de validación, una práctica recomendada es volver a entrar la arquitectura con los mismos parámetros con los que se generó el modelo seleccionado y con el conjunto de entrenamiento original, es decir, integrado el conjunto de validación al conjunto de entrenamiento \"parcial\". "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1PQdKOxeEbFM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bc0cba53-69d7-446f-af8c-50c5ef53f86a"
      },
      "source": [
        "model = models.Sequential()\n",
        "model.add(layers.Dense(16, activation='relu', input_shape=(10000,)))\n",
        "model.add(layers.Dense(16, activation='relu'))\n",
        "model.add(layers.Dense(1, activation='sigmoid'))\n",
        "\n",
        "model.compile(optimizer=optimizers.RMSprop(lr=0.001),\n",
        "              loss='binary_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "# Entrenamos el modelo en el conjunto de entrenamiento original\n",
        "model.fit(x_train, y_train, epochs=5, batch_size=512)"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/5\n",
            "49/49 [==============================] - 1s 25ms/step - loss: 0.4687 - accuracy: 0.8194\n",
            "Epoch 2/5\n",
            "49/49 [==============================] - 1s 24ms/step - loss: 0.2720 - accuracy: 0.9064\n",
            "Epoch 3/5\n",
            "49/49 [==============================] - 1s 24ms/step - loss: 0.2060 - accuracy: 0.9270\n",
            "Epoch 4/5\n",
            "49/49 [==============================] - 1s 25ms/step - loss: 0.1742 - accuracy: 0.9380\n",
            "Epoch 5/5\n",
            "49/49 [==============================] - 1s 24ms/step - loss: 0.1485 - accuracy: 0.9487\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f9c42dbef28>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bsgLrSXXBuUI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "58d81c93-72f7-4bec-e730-04113052fb4c"
      },
      "source": [
        "#Evaluamos el modelo final\n",
        "loss_test, acc_test = model.evaluate(x_test, y_test)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "782/782 [==============================] - 1s 2ms/step - loss: 0.3155 - accuracy: 0.8773\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qBCyFJo4Ejss",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f44af17e-4758-4e67-8168-9d5f78f23a8a"
      },
      "source": [
        "print(loss_test, acc_test)"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.31554871797561646 0.8773199915885925\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4v58mbPDEqC1"
      },
      "source": [
        "# **Predicciones del modelo final**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Vyah_Qp1EvVV"
      },
      "source": [
        "Luego de generar un modelo apropiado para clasificar críticas, podemos usarlo para predecir la probabilidad de que cada nueva crítica que sea leída por el modelo, se clasifique como positiva o negativa. Este proceso predictivo es muy importante, ya que así es como se desplegaría esta red neuronal en \"la vida real\", es decir, en algún servicio de consumo."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yWED4A2ZFJCc",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3f6192b3-47fc-42ba-a7a6-51f92bf1355b"
      },
      "source": [
        "model.predict(x_test)"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[0.24245262],\n",
              "       [0.999631  ],\n",
              "       [0.9650076 ],\n",
              "       ...,\n",
              "       [0.1443555 ],\n",
              "       [0.10413879],\n",
              "       [0.772051  ]], dtype=float32)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AD5AC53WFNaz",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "83410982-7937-4c1c-b113-7dd725edbd3c"
      },
      "source": [
        "y_test"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([0., 1., 1., ..., 0., 0., 0.], dtype=float32)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 29
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FekKrabcxc-h"
      },
      "source": [
        "**NOTA:** La información presentada en este notebook, fue tomada de https://github.com/fchollet/deep-learning-with-python-notebooks/blob/master/3.5-classifying-movie-reviews.ipynb. La única variación sutil del presente notebook con respecto a la fuente original, tiene que ver con la descripción de ciertas partes del código."
      ]
    }
  ]
}