aula-11/Aula_11.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Aula_11.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KlcibMKA_a-j",
        "colab_type": "text"
      },
      "source": [
        "# Aula 11 - Redes neurais"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TqWhkCoq_jyp",
        "colab_type": "text"
      },
      "source": [
        "Neste notebook iremos utilizar a biblioteca [Keras](https://www.keras.io) para construir redes neurais.\n",
        "\n",
        "Faremos dois exemplos: XOR lógico (para classificação) e preço de casas (para regressão)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Kc_OQIKowdob",
        "colab_type": "text"
      },
      "source": [
        "## Exemplo 1 - XOR lógico\n",
        "\n",
        "Para começar, precisamos importar algumas bibliotecas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "86xOUdmcI5pp",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.models import Sequential\n",
        "from keras.layers.core import Dense, Activation\n",
        "from keras.utils import to_categorical\n",
        "from keras.utils import np_utils"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V8442v5lno6D",
        "colab_type": "text"
      },
      "source": [
        "Iniciaremos com uma tarefa simples: aprender a função lógica XOR. Para isto, utilizaremos a tabela verdade desta operação como dataset para treinar a rede neural.\n",
        "\n",
        "A | B | XOR\n",
        "- | - | -\n",
        "0 | 0 | 0\n",
        "0 | 1 | 1\n",
        "1 | 0 | 1\n",
        "1 | 1 | 0\n",
        "\n",
        "Em particular, utilizaremos as colunas A e B como entradas da rede neural e a coluna XOR como saída. Especificamente, cada exemplo será composto por uma entrada X (par de dois valores representando A e B) e por uma saída y (valor individual representando XOR)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CjH5gofKI7pj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# entradas e saídas referentes à operação lógica XOR\n",
        "X = np.array([[0,0],[0,1],[1,0],[1,1]]).astype('float32')\n",
        "y = np.array([[0],[1],[1],[0]]).astype('float32')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PAvbA-I8pfSo",
        "colab_type": "text"
      },
      "source": [
        "Resolveremos a tarefa acima como um problema de classificação. Desta forma, é conveniente codificar a saída de forma categórica (mais detalhes sobre este tipo de procedimento [aqui](https://machinelearningmastery.com/why-one-hot-encode-data-in-machine-learning/))."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZuBzEWSNI9nG",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# codifica as saídas de forma categórica\n",
        "y = np_utils.to_categorical(y)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DjoSjdr6p-nM",
        "colab_type": "text"
      },
      "source": [
        "Na categorização acima, os valores 0 e 1 são representados por [1, 0] e [0, 1], respectivamente. Isto fica evidente ao imprimir a variável."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PnzYoE-K_qfP",
        "colab_type": "code",
        "outputId": "488a560a-0f73-4171-aa9b-5c221321d1ed",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 84
        }
      },
      "source": [
        "print(y)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1. 0.]\n",
            " [0. 1.]\n",
            " [0. 1.]\n",
            " [1. 0.]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P5uZ4cDmJIW5",
        "colab_type": "text"
      },
      "source": [
        "Nosso próximo passo consiste em construir a rede neural. Para isto, utilizaremos o modelo sequencial do Keras. Neste modelo, podemos empilhar as diferentes camadas da rede com bastante facilidade."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P0CDYpW8qc_3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# cria um modelo sequencial\n",
        "model = Sequential()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WXzPkXjyqfmL",
        "colab_type": "text"
      },
      "source": [
        "Após criar o modelo sequencial, podemos empilhar as camadas utilizando a função `add`. Primeiramente, criaremos uma camada de 64 neurônios, cada qual com uma função de ativação do tipo tangente hiperbólica. O parâmetro `input_dim` é necessário para definir a dimensão das entradas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fm5nAVvYIHdj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# 1a camada: 64 neurônios com a mesma entrada com tamanho compatível com a entrada\n",
        "model.add(Dense(64, activation='tanh', input_dim=X.shape[1]))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U9Y6qnuhsSIk",
        "colab_type": "text"
      },
      "source": [
        "Ao final, criaremos uma camada de saída. Considerando que nossas saídas são categóricas, cada qual de dimensão 2, nossa rede também precisa ter duas saídas. Desta forma, criaremos uma camada de saída com dois neurônios, cada qual com uma função de ativação do tipo sigmóide. Note que nesta camada não foi preciso definir a dimensão de entrada; isto acontece porque o Keras define automaticamente a dimensão de entrada dos neurônios de uma camada oculta com base nas saídas da camada anterior."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DFfhWDT8sRpr",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# 2a camada (saída): 2 neurônios, totalmente conectada\n",
        "model.add(Dense(2, activation='sigmoid'))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "teK8GeVEs0lh",
        "colab_type": "text"
      },
      "source": [
        "Definida a rede, agora podemos compilar a mesma para gerar o modelo final. Na compilação, devemos definir o algoritmo (`optimizer`) que será utilizado para treinar a rede bem como a função de erro (`loss`) e as métricas (`metrics`). O Keras possui diversos algoritmos de otimização ([lista](https://keras.io/api/optimizers/)), funções de erro ([lista](https://keras.io/api/losses/)) e métricas ([lista](https://keras.io/api/metrics/)). No nosso caso, estamos utilizando métricas e função de erro próprias para a tarefa de classificação. Para otimização, estamos utilizando o `adam`, que é um excelente algoritmo para redes neurais."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NpL404GRJNsN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# compila o modelo, utilizando a entropia cruzada como função de erro, \n",
        "# com o algoritmo Adam, e avaliando o modelo através da métrica acurácia\n",
        "model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics = [\"binary_crossentropy\", \"accuracy\"])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kSJ10DVvtq-Q",
        "colab_type": "text"
      },
      "source": [
        "Para ter uma visão geral do modelo resultante, podemos imprimir um resumo do mesmo. Note que o modelo criado possui 322 parâmetros. Isto significa que, durante o treinamento, o algoritmo de otimização terá que ajustar o valor de 322 variáveis de modo a reduzir o erro."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LG5pVD34trMg",
        "colab_type": "code",
        "outputId": "2f134914-edaf-47a0-cea1-ac336f886531",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 218
        }
      },
      "source": [
        "# mostra um resumo do modelo compilado\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_2\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense_3 (Dense)              (None, 64)                192       \n",
            "_________________________________________________________________\n",
            "dense_4 (Dense)              (None, 2)                 130       \n",
            "=================================================================\n",
            "Total params: 322\n",
            "Trainable params: 322\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dp02cfV5JS2n",
        "colab_type": "text"
      },
      "source": [
        "Definidos os dados e a rede, agora podemos iniciar o treinamento. Para isto, devemos utilizar a função `fit`, especificando os dados de entrada e saída (no nosso caso, `X` e `y`) e o número de épocas (`epochs`)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M5D7OMCuIHnJ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# treina o modelo\n",
        "history = model.fit(X, y, epochs=1000, verbose=0) "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w0JEjML_u1a2",
        "colab_type": "text"
      },
      "source": [
        "No código acima, armazenamos o retorno do treinamento na variável `history` para poder visualizar a evolução do processo de treinamento. Para isto, podemos plotar o erro ao longo do tempo como segue."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zW6VKearKMZZ",
        "colab_type": "code",
        "outputId": "2733854d-0041-4575-ca88-6d8c7c6e856b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "SjtDxiDsvL6B",
        "colab_type": "text"
      },
      "source": [
        "Conforme pode ser observado, o erro final é bastante baixo. No entanto, ele só chega a níveis adequados no final do treinamento. \n",
        "\n",
        "Da mesma forma que fizemos com o erro, podemos plotar também a acurácia do modelo."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YyZ8DPqzR0n8",
        "colab_type": "code",
        "outputId": "61557647-c754-4cf4-9fd0-53e4cf557398",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['accuracy'])\n",
        "plt.title('model accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEWCAYAAAB8LwAVAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAcYUlEQVR4nO3dfZwdVZ3n8c+XBBKQQAKJDOQZDUjcVYI9AQZnQBGMICKMMwKCgIwZFRRY2BkYXWAz6+q8XoyPw/CgRh5EAkZgskwkE8LDLAtCGhMRIpGAYDqAREMCIgGS/PaPOp1UmhtSnXT1TVV936/XfXHrVNW9p7pCf/ucU1VHEYGZmVlP27W7AmZmtm1yQJiZWUsOCDMza8kBYWZmLTkgzMysJQeEmZm15IAwAyRdLel/Fdz2KUkfKLtOZu3mgDAzs5YcEGY1Imlgu+tg9eGAsMpIXTv/XdLDkl6W9D1Je0j6iaSXJN0haVhu+49IelTSSkl3S9ovt26SpJ+l/W4EBvf4rg9LWpj2vU/SuwrW8WhJCyS9KGmppEt6rH9v+ryVaf1pqXxHSf8s6WlJqyTdm8oOk9TV4ufwgfT+EkkzJf1A0ovAaZImS7o/fcezkv5F0g65/d8paa6kFZJ+K+kfJP2JpD9K2j233QGSlkvavsixW/04IKxq/hI4AtgHOAb4CfAPwAiyf89fAJC0D3ADcE5aNxv4P5J2SL8sbwWuA3YDfpQ+l7TvJGA68LfA7sCVwCxJgwrU72Xgk8BQ4Gjgs5I+mj53bKrvt1Od9gcWpv0uBd4D/Fmq098B6wr+TI4FZqbvvB5YC5wLDAcOBg4HPpfqMAS4A7gd2At4OzAvIp4D7gb+Ove5pwAzIuL1gvWwmnFAWNV8OyJ+GxHLgP8LPBARCyJiNXALMClt93Hg3yNibvoFdymwI9kv4IOA7YFvRMTrETETmJ/7jqnAlRHxQESsjYhrgFfTfm8qIu6OiF9ExLqIeJgspA5Nq08C7oiIG9L3/j4iFkraDvgUcHZELEvfeV9EvFrwZ3J/RNyavvOViHgoIn4aEWsi4imygOuuw4eB5yLinyNidUS8FBEPpHXXACcDSBoAnEgWotZQDgirmt/m3r/SYnnn9H4v4OnuFRGxDlgKjEzrlsXGT6p8Ovd+LHBe6qJZKWklMDrt96YkHSjprtQ1swr4DNlf8qTPeKLFbsPJurharStiaY867CPpNknPpW6n/12gDgD/BkyUNJ6slbYqIh7cwjpZDTggrK6eIftFD4Akkf1yXAY8C4xMZd3G5N4vBb4cEUNzr50i4oYC3/tDYBYwOiJ2Ba4Aur9nKfC2Fvv8Dli9iXUvAzvljmMAWfdUXs9HMl8OPAZMiIhdyLrg8nXYu1XFUyvsJrJWxCm49dB4Dgirq5uAoyUdngZZzyPrJroPuB9YA3xB0vaSjgcm5/b9DvCZ1BqQpLekwechBb53CLAiIlZLmkzWrdTteuADkv5a0kBJu0vaP7VupgNfk7SXpAGSDk5jHr8CBqfv3x74ErC5sZAhwIvAHyS9A/hsbt1twJ6SzpE0SNIQSQfm1l8LnAZ8BAdE4zkgrJYiYjHZX8LfJvsL/RjgmIh4LSJeA44n+0W4gmy84ubcvp3Ap4F/AV4AlqRti/gcME3SS8BFZEHV/bm/AY4iC6sVZAPU706rzwd+QTYWsgL4J2C7iFiVPvO7ZK2fl4GNrmpq4XyyYHqJLOxuzNXhJbLuo2OA54DHgffl1v8/ssHxn0VEvtvNGkieMMjM8iTdCfwwIr7b7rpYezkgzGw9SX8KzCUbQ3mp3fWx9nIXk5kBIOkasnskznE4GLgFYWZmm+AWhJmZtVSbB3sNHz48xo0b1+5qmJlVykMPPfS7iOh5bw1Qo4AYN24cnZ2d7a6GmVmlSNrk5czuYjIzs5YcEGZm1pIDwszMWnJAmJlZSw4IMzNrqbSAkDRd0vOSHtnEekn6lqQlyqaQPCC37lRJj6fXqWXV0czMNq3MFsTVwJQ3Wf8hYEJ6TSV7hj2SdgMuBg4kewTzxcrNM2xmZv2jtPsgIuI/JY17k02OBa5Ns3r9VNJQSXsChwFzI2IFgKS5ZEFTZLKWbcIjy1bxH48+1+5qmFlD/MmuO3LSgWM2v2EvtfNGuZFsPFViVyrbVPkbSJpK1vpgzJi+/+FsqcvveYJ/f/hZNpqvzMysJPuPHlq7gNhqEXEVcBVAR0fHNvPUwbVrg333GMKcc/+i3VUxM9ti7byKaRnZHMHdRqWyTZVXRrxhimAzs+ppZ0DMAj6ZrmY6CFgVEc8Cc4AjJQ1Lg9NHprJKcfeSmVVdaV1Mkm4gG3AeLqmL7Mqk7QEi4gpgNtn8vEuAPwKnp3UrJP0j2dy8ANO6B6yrwlNsmFkdlHkV04mbWR/AmZtYNx2YXka9+kMAchPCzCrOd1KXwC0IM6sDB0QpArcfzKzqHBAlcQ+TmVWdA6IEEQ4IM6s+B0QJApA7mcys4hwQJQiPUptZDTggSpBd5truWpiZbR0HREmcD2ZWdQ6IEoSbEGZWAw6IEngEwszqwAFRggjfKGdm1eeAKIl7mMys6hwQJXE+mFnVOSBKkN1J7Ygws2pzQJTAM8qZWR04IEoQ4S4mM6s+B0QJ/LA+M6sDB0RJ/LA+M6s6B0QJAvcxmVn1OSBK4Ie5mlkdOCBKkM0HYWZWbQ6IMniQ2sxqwAFREg9Sm1nVOSBK4BvlzKwOHBAl8H0QZlYHDogSeL4gM6sDB0QJsvkgnBBmVm0OiJK4BWFmVeeAKIGHqM2sDhwQJfCd1GZWBw6IEmSD1O5jMrNqKzUgJE2RtFjSEkkXtFg/VtI8SQ9LulvSqNy6tZIWptesMuvZ5yI8RG1mlTewrA+WNAC4DDgC6ALmS5oVEYtym10KXBsR10h6P/AV4JS07pWI2L+s+pXNDQgzq7oyWxCTgSUR8WREvAbMAI7tsc1E4M70/q4W6yvJQxBmVgdlBsRIYGluuSuV5f0cOD69Pw4YImn3tDxYUqekn0r6aKsvkDQ1bdO5fPnyvqz7VvGUo2ZWB+0epD4fOFTSAuBQYBmwNq0bGxEdwEnANyS9refOEXFVRHRERMeIESP6rdKbE4QHqc2s8kobgyD7ZT86tzwqla0XEc+QWhCSdgb+MiJWpnXL0n+flHQ3MAl4osT69hm3IMysDspsQcwHJkgaL2kH4ARgo6uRJA2X1F2HC4HpqXyYpEHd2wCHAPnBbTMzK1lpARERa4CzgDnAL4GbIuJRSdMkfSRtdhiwWNKvgD2AL6fy/YBOST8nG7z+ao+rn7ZpfpqrmdVBmV1MRMRsYHaPsoty72cCM1vsdx/wX8usW5myq5icEGZWbe0epK6liHALwswqzwFREueDmVWdA8LMzFpyQJTAg9RmVgcOiBIEnlHOzKrPAVECtyDMrA4cECVxQJhZ1TkgSuCnuZpZHTggShDhMQgzqz4HRAkCfCOEmVWeA6IMfpqrmdWAA8LMzFpyQJQgwBMGmVnlOSBKkA1Sm5lVmwOiBFkLot21MDPbOg6IEnjKUTOrAweEmZm15IAoQRAepDazynNAlMBdTGZWBw6IEkTghDCzynNAlMTPYjKzqnNAmJlZSw6IEkSE74Mws8pzQJTAQxBmVgcOiBJ4ylEzqwMHRAnCc8qZWQ04IEriq5jMrOocECVwF5OZ1YEDogR+mquZ1YEDogThSanNrAZKDQhJUyQtlrRE0gUt1o+VNE/Sw5LuljQqt+5USY+n16ll1rPveZDazKqvtICQNAC4DPgQMBE4UdLEHptdClwbEe8CpgFfSfvuBlwMHAhMBi6WNKysupbBXUxmVnVltiAmA0si4smIeA2YARzbY5uJwJ3p/V259R8E5kbEioh4AZgLTCmxrn3KT3M1szooFBCSbpZ0tKTeBMpIYGluuSuV5f0cOD69Pw4YImn3gvsiaaqkTkmdy5cv70XVyuVBajOrg6K/8P8VOAl4XNJXJe3bR99/PnCopAXAocAyYG3RnSPiqojoiIiOESNG9FGVtl5E+D4IM6u8QgEREXdExCeAA4CngDsk3SfpdEnbb2K3ZcDo3PKoVJb/3Gci4viImAR8MZWtLLLvtsxD1GZWB4W7jFLXz2nA3wALgG+SBcbcTewyH5ggabykHYATgFk9PnN4rtvqQmB6ej8HOFLSsDQ4fWQqqwx3MZlZ1Q0sspGkW4B9geuAYyLi2bTqRkmdrfaJiDWSziL7xT4AmB4Rj0qaBnRGxCzgMOArkgL4T+DMtO8KSf9IFjIA0yJixRYdYRt4kNrM6qBQQADfioi7Wq2IiI5N7RQRs4HZPcouyr2fCczcxL7T2dCiqJRsPghHhJlVW9EupomShnYvpK6fz5VUp8rzGISZ1UHRgPh0GjwGIN2b8OlyqmRmZtuCogExQLk+k3SX9A7lVKkG/DRXM6uBomMQt5MNSF+Zlv82lVkL2ZSjTggzq7aiAfH3ZKHw2bQ8F/huKTWqgWyQut21MDPbOoUCIiLWAZenl22GB6nNrA6K3gcxgexJqxOBwd3lEbF3SfWqPDcgzKzqig5Sf5+s9bAGeB9wLfCDsipVdZ5y1MzqoGhA7BgR8wBFxNMRcQlwdHnVqrbAN8qZWfUVHaR+NT0z6fH0+IxlwM7lVava/KgNM6uDoi2Is4GdgC8A7wFOBio2DWj/8SC1mdXBZlsQ6aa4j0fE+cAfgNNLr1UduAlhZhW32RZERKwF3tsPdamP8I1yZlZ9RccgFkiaBfwIeLm7MCJuLqVW26DnX1zNV37yGKtf3/yEd6+vW+ermMys8ooGxGDg98D7c2UBNCYgHnxqBbcsWMbY3Xdi0MA3b3jtu8cQJo/frZ9qZmZWjqJ3Ujd+3CHSyPP3Tu3g7W8d0t7KmJn1g6J3Un+fFhfnRMSn+rxG2yhfmWRmTVO0i+m23PvBwHHAM31fnW1XdDchPPhsZg1RtIvpx/llSTcA95ZSo22cB5/NrCmK3ijX0wTgrX1ZkapwPphZUxQdg3iJjbvhnyObI6IxunuY/IwlM2uKol1Mjb9sJzxMbWYNU6iLSdJxknbNLQ+V9NHyqrXtWd+CaG81zMz6TdExiIsjYlX3QkSsBC4up0rbNvcwmVlTFA2IVtsVvUS2Fja0IJwQZtYMRQOiU9LXJL0tvb4GPFRmxbY1HoEws6YpGhCfB14DbgRmAKuBM8uq1Lao+0Y5dzGZWVMUvYrpZeCCkuuyTXMLwsyapuhVTHMlDc0tD5M0p7xqbbvcgjCzpijaxTQ8XbkEQES8QNPupPaNcmbWMEUDYp2kMd0LksbRsF4X3yhnZk1TNCC+CNwr6TpJPwDuAS7c3E6SpkhaLGmJpDeMYUgaI+kuSQskPSzpqFQ+TtIrkham1xW9Oagy+EY5M2uaooPUt0vqAKYCC4BbgVfebB9JA4DLgCOALmC+pFkRsSi32ZeAmyLickkTgdnAuLTuiYjYvzcHU6b1D/t2QphZQxR9WN/fAGcDo4CFwEHA/Ww8BWlPk4ElEfFk+owZwLFAPiAC2CW935UKzDHhG+XMrCmKdjGdDfwp8HREvA+YBKx8810YCSzNLXelsrxLgJMldZG1Hj6fWzc+dT3dI+nPW32BpKmSOiV1Ll++vOChbJkNT3Mt9WvMzLYZRQNidUSsBpA0KCIeA/btg+8/Ebg6IkYBRwHXSdoOeBYYExGTgP8G/FDSLj13joirIqIjIjpGjBjRB9XZNA9Sm1nTFH2eUle6D+JWYK6kF4CnN7PPMmB0bnlUKss7A5gCEBH3SxpMdknt88CrqfwhSU8A+wCdBevb5zxIbWZNU3SQ+rj09hJJd5GNF9y+md3mAxMkjScLhhOAk3ps8xvgcOBqSfuRzXe9XNIIYEVErJW0N9kMdk8WqWtZ1rcfnBBm1hC9fiJrRNxTcLs1ks4C5gADgOkR8aikaUBnRMwCzgO+I+lcst/Bp0VESPoLYJqk14F1wGciYkVv61oGD1KbWVOU+sjuiJhNNvicL7so934RcEiL/X4M/LjMuvVaeAzCzJql6CB14/k+CDNrGgdEQR6kNrOmcUD0kh/WZ2ZN4YAoaP2EQW2uh5lZf3FAFOQhajNrGgdEQX7Uhpk1jQOioPVXMbmTycwawgHRW84HM2sIB0RB6wepHRBm1hAOCDMza8kBUZBvlDOzpnFAFNQ9H4RvlDOzpnBA9JLjwcyawgFRkB/mamZN44AoyE9zNbOmcUAUtGGQ2glhZs3ggChowyB1mytiZtZPHBBmZtaSA6IgD1KbWdM4IHrJXUxm1hQOiII2TBjkhDCzZnBAFOT5IMysaRwQveR8MLOmcEAU5DFqM2saB0RBG7qY3IYws2ZwQBS0/ka5NtfDzKy/OCAK8iC1mTWNA8LMzFpyQBS04WmubkKYWTM4IIryszbMrGFKDQhJUyQtlrRE0gUt1o+RdJekBZIelnRUbt2Fab/Fkj5YZj2LCDz+YGbNMrCsD5Y0ALgMOALoAuZLmhURi3KbfQm4KSIulzQRmA2MS+9PAN4J7AXcIWmfiFhbVn03J8JXMJlZs5TZgpgMLImIJyPiNWAGcGyPbQLYJb3fFXgmvT8WmBERr0bEr4El6fPMzKyflBkQI4GlueWuVJZ3CXCypC6y1sPne7EvkqZK6pTUuXz58r6qd0tBeIDazBql3YPUJwJXR8Qo4CjgOkmF6xQRV0VER0R0jBgxorRKZt/lLiYza5bSxiCAZcDo3PKoVJZ3BjAFICLulzQYGF5w337lQWoza5oyWxDzgQmSxkvagWzQeVaPbX4DHA4gaT9gMLA8bXeCpEGSxgMTgAdLrOtmZS0IJ4SZNUdpLYiIWCPpLGAOMACYHhGPSpoGdEbELOA84DuSziX7I/20yGbmeVTSTcAiYA1wZjuvYDIza6Iyu5iIiNlkg8/5soty7xcBh2xi3y8DXy6zfr0ReBDCzJql3YPU1eF8MLOGcUAU5EFqM2saB0RB4WcxmVnDOCB6wVcxmVmTOCAKinAXk5k1iwOioMCD1GbWLA6IgrIWhCPCzJrDAWFmZi05IAoKwl1MZtYoDoiCwoMQZtYwDohecD6YWZM4IAqK8IRBZtYsDggzM2vJAVGQn8VkZk3jgCjIU46aWdM4IAoKPAZhZs3igCjID3M1s6ZxQPSC2w9m1iQOiII8SG1mTeOAKCjrYnJCmFlzOCAKC7cgzKxRHBAFeZDazJrGAdELbkCYWZM4IArylKNm1jQOiIKy+SCcEGbWHA6IgjwGYWZN44AoyPdBmFnTOCB6wflgZk3igCgoG6R2RJhZczggCgo8CGFmzeKAKMr5YGYNU2pASJoiabGkJZIuaLH+65IWptevJK3MrVubWzerzHoW4UFqM2uagWV9sKQBwGXAEUAXMF/SrIhY1L1NRJyb2/7zwKTcR7wSEfuXVb8t4YAwsyYpLSCAycCSiHgSQNIM4Fhg0Sa2PxG4uMT6tLTyj6/xV1fcv9ntnlu1mmFv2aEfamRmtm0oMyBGAktzy13Aga02lDQWGA/cmSseLKkTWAN8NSJubbHfVGAqwJgxY7aoktttJybssfNmt5uwx84cvPfuW/QdZmZVVGZA9MYJwMyIWJsrGxsRyyTtDdwp6RcR8UR+p4i4CrgKoKOjY4uGkXcZvD3/+on3bGm9zcxqq8xB6mXA6NzyqFTWygnADfmCiFiW/vskcDcbj0+YmVnJygyI+cAESeMl7UAWAm+4GknSO4BhwP25smGSBqX3w4FD2PTYhZmZlaC0LqaIWCPpLGAOMACYHhGPSpoGdEZEd1icAMyI2OhxePsBV0paRxZiX81f/WRmZuVT1OQxpR0dHdHZ2dnuapiZVYqkhyKio9U630ltZmYtOSDMzKwlB4SZmbXkgDAzs5ZqM0gtaTnw9FZ8xHDgd31UnarwMddf044XfMy9NTYiRrRaUZuA2FqSOjc1kl9XPub6a9rxgo+5L7mLyczMWnJAmJlZSw6IDa5qdwXawMdcf007XvAx9xmPQZiZWUtuQZiZWUsOCDMza6nxASFpiqTFkpZIuqDd9ekrkkZLukvSIkmPSjo7le8maa6kx9N/h6VySfpW+jk8LOmA9h7BlpM0QNICSbel5fGSHkjHdmN6/DySBqXlJWn9uHbWe0tJGipppqTHJP1S0sF1P8+Szk3/rh+RdIOkwXU7z5KmS3pe0iO5sl6fV0mnpu0fl3Rqb+rQ6ICQNAC4DPgQMBE4UdLE9taqz6wBzouIicBBwJnp2C4A5kXEBGBeWobsZzAhvaYCl/d/lfvM2cAvc8v/BHw9It4OvACckcrPAF5I5V9P21XRN4HbI+IdwLvJjr2251nSSOALQEdE/Bey6QROoH7n+WpgSo+yXp1XSbsBF5NN9zwZuLg7VAqJiMa+gIOBObnlC4EL212vko7134AjgMXAnqlsT2Bxen8lcGJu+/XbVelFNnPhPOD9wG2AyO4wHdjznJPNVXJwej8wbad2H0Mvj3dX4Nc9613n88yG+e53S+ftNuCDdTzPwDjgkS09r8CJwJW58o2229yr0S0INvxD69aVymolNaknAQ8Ae0TEs2nVc8Ae6X1dfhbfAP4OWJeWdwdWRsSatJw/rvXHnNavSttXyXhgOfD91K32XUlvocbnObLpiC8FfgM8S3beHqLe57lbb8/rVp3vpgdE7UnaGfgxcE5EvJhfF9mfFLW5zlnSh4HnI+KhdtelHw0EDgAuj4hJwMts6HYAanmehwHHkoXjXsBbeGNXTO31x3ltekAsA0bnlkelslqQtD1ZOFwfETen4t9K2jOt3xN4PpXX4WdxCPARSU8BM8i6mb4JDJXUPb1u/rjWH3Navyvw+/6scB/oAroi4oG0PJMsMOp8nj8A/DoilkfE68DNZOe+zue5W2/P61ad76YHxHxgQrr6YQeyga5Zm9mnEiQJ+B7wy4j4Wm7VLKD7SoZTycYmuss/ma6GOAhYlWvKVkJEXBgRoyJiHNm5vDMiPgHcBXwsbdbzmLt/Fh9L21fqL+2IeA5YKmnfVHQ4sIgan2eyrqWDJO2U/p13H3Ntz3NOb8/rHOBIScNSy+vIVFZMuwdh2v0CjgJ+BTwBfLHd9enD43ovWfPzYWBheh1F1vc6D3gcuAPYLW0vsiu6ngB+QXaFSNuPYyuO/zDgtvR+b+BBYAnwI2BQKh+clpek9Xu3u95beKz7A53pXN8KDKv7eQb+J/AY8AhwHTCobucZuIFsjOV1spbiGVtyXoFPpWNfApzemzr4URtmZtZS07uYzMxsExwQZmbWkgPCzMxackCYmVlLDggzM2vJAWG2DZB0WPfTZ822FQ4IMzNryQFh1guSTpb0oKSFkq5Mc0/8QdLX0/wE8ySNSNvuL+mn6fn8t+Se3f92SXdI+rmkn0l6W/r4nXPzOlyf7hI2axsHhFlBkvYDPg4cEhH7A2uBT5A9LK4zIt4J3EP2/H2Aa4G/j4h3kd3d2l1+PXBZRLwb+DOyu2Uhe+LuOWRzk+xN9nwhs7YZuPlNzCw5HHgPMD/9cb8j2cPS1gE3pm1+ANwsaVdgaETck8qvAX4kaQgwMiJuAYiI1QDp8x6MiK60vJBsLoB7yz8ss9YcEGbFCbgmIi7cqFD6Hz2229Ln17yae78W//9pbeYuJrPi5gEfk/RWWD8/8Fiy/4+6nyJ6EnBvRKwCXpD056n8FOCeiHgJ6JL00fQZgyTt1K9HYVaQ/0IxKygiFkn6EvAfkrYje8rmmWST9ExO654nG6eA7HHMV6QAeBI4PZWfAlwpaVr6jL/qx8MwK8xPczXbSpL+EBE7t7seZn3NXUxmZtaSWxBmZtaSWxBmZtaSA8LMzFpyQJiZWUsOCDMza8kBYWZmLf1/beYQF7cYqNoAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2i9J-h3YvZ5l",
        "colab_type": "text"
      },
      "source": [
        "Bem como anteriormente, note que a acurácia aumenta com o treinamento do modelo."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_WAnRBduJbyg",
        "colab_type": "text"
      },
      "source": [
        "Finalmente, agora podemos avaliar nosso modelo. Para isto, utilizaremos a função `evaluate`, que recebe como argumentos dados de entrada (`X`) e saída (`y`) e retorna as métricas anteriormente definidas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9fmulfWKJb6X",
        "colab_type": "code",
        "outputId": "64b9511a-f0e3-43ba-eb78-29497e85f3f7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 168
        }
      },
      "source": [
        "# avalia o modelo\n",
        "score = model.evaluate(X, y)\n",
        "print(\"\\nAcurácia: \", score[-1])\n",
        "\n",
        "# verifica as predições\n",
        "print(\"\\nPredições:\")\n",
        "pred = model.predict_proba(X)\n",
        "print(pred)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r4/4 [==============================] - 0s 357us/step\n",
            "\n",
            "Acurácia:  1.0\n",
            "\n",
            "Predições:\n",
            "[[0.9956118  0.00387996]\n",
            " [0.00456738 0.99577177]\n",
            " [0.00527772 0.9953675 ]\n",
            " [0.9951901  0.00414902]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4A7wyHTVv3fX",
        "colab_type": "text"
      },
      "source": [
        "Note que nosso modelo obteve uma acurácia de 100%. Nenhuma surpresa, dado que o problema XOR é bastante simples.\n",
        "\n",
        "Para melhorar o entendimento, abaixo podemos comparar as saídas da rede com as saídas esperadas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8dxblcXoIHrQ",
        "colab_type": "code",
        "outputId": "8702c09e-a44d-4fe5-ebc4-cf57e889789a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 101
        }
      },
      "source": [
        "# imprime as predições de forma mais clara\n",
        "print(\"Entrada\\t\\tSaída esperada\\t\\tSaída prevista\")\n",
        "for i in range(len(X)):\n",
        "    print(f\"{X[i]}\\t\\t{y[i]}\\t\\t\\t{[round(v) for v in pred[i]]}\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Entrada\t\tSaída esperada\t\tSaída prevista\n",
            "[0. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 1.]\t\t[0. 1.]\t\t\t[0.0, 1.0]\n",
            "[1. 0.]\t\t[0. 1.]\t\t\t[0.0, 1.0]\n",
            "[1. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jWvKJovtMkPe",
        "colab_type": "text"
      },
      "source": [
        "## Exemplo 2 - Boston Housing\n",
        "\n",
        "Para começar, precisamos importar algumas bibliotecas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gKXTL5LoxBvM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.models import Sequential\n",
        "from keras.layers.core import Dense, Activation\n",
        "from keras.utils import to_categorical\n",
        "from keras.utils import np_utils\n",
        "from keras.datasets import boston_housing"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MDXImo-qxB95",
        "colab_type": "text"
      },
      "source": [
        "Neste exemplo, resolveremos a tarefa de regressão. Para isto, utilizaremos o dataset Boston Housing, que contém informações sobre imóveis e seus preços na cidade de Boston (EUA). Este dataset pode ser facilmente importado através da função `boston_housing.load_data`. Esta função retorna duas tuplas, as quais representam os dados de treino (entrada e saída) e de teste (entrada e saída)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iPvnZUj7xVot",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "(train_data, train_targets), (test_data, test_targets) = boston_housing.load_data()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VuKLnG7TxVLR",
        "colab_type": "text"
      },
      "source": [
        "Para entender melhor estes dados, podemos imprimir uma pequena amostra como segue."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pdJIGNLHxs15",
        "colab_type": "code",
        "outputId": "b5223a23-1b79-4772-b835-53b51dc6adee",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 121
        }
      },
      "source": [
        "print('Shape dados de treino:   %s'% (train_data.shape,))\n",
        "print('Shape dados de teste:    %s'% (test_data.shape,))\n",
        "print('Exemplo de entrada:      %s'% (train_data[0]))\n",
        "print('Exemplo de saída:        %f'% (train_targets[0]))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shape dados de treino:   (404, 13)\n",
            "Shape dados de teste:    (102, 13)\n",
            "Exemplo de entrada:      [  1.23247   0.        8.14      0.        0.538     6.142    91.7\n",
            "   3.9769    4.      307.       21.      396.9      18.72   ]\n",
            "Exemplo de saída:        15.200000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vE5s5BXsxCT7",
        "colab_type": "text"
      },
      "source": [
        "Para que o modelo aprenda melhor, precisamos realizar um pré-processamento dos dados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zymgM_5Ayt6t",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "mean = train_data.mean(axis=0)\n",
        "train_data -= mean\n",
        "std = train_data.std(axis=0)\n",
        "train_data /= std\n",
        "\n",
        "test_data -= mean\n",
        "test_data /= std"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VhTkWCJSxCd6",
        "colab_type": "text"
      },
      "source": [
        "Definido o dataset, agora podemos criar a rede neural.\n",
        "\n",
        "Neste exemplo, estamos empilhando três camadas, sendo duas com 64 neurônios cada e uma (de saída) com apenas um neurônio. Note que a camada de saída possui apenas um neurônio porque estamos lidando com a tarefa de regressão."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-8sy3zPyy45B",
        "colab_type": "code",
        "outputId": "2260c38d-ba5e-4b2e-ca1b-deacb7e903d1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 252
        }
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Dense(128, activation='relu', input_shape=(train_data.shape[1],)))\n",
        "model.add(Dense(128, activation='relu'))\n",
        "model.add(Dense(1))\n",
        "\n",
        "model.compile(optimizer='rmsprop', loss='mse', metrics=['mse'])\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_3\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense_7 (Dense)              (None, 128)               1792      \n",
            "_________________________________________________________________\n",
            "dense_8 (Dense)              (None, 128)               16512     \n",
            "_________________________________________________________________\n",
            "dense_9 (Dense)              (None, 1)                 129       \n",
            "=================================================================\n",
            "Total params: 18,433\n",
            "Trainable params: 18,433\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sOkCIztKxCNO",
        "colab_type": "text"
      },
      "source": [
        "Agora podemos treinar o modelo utilizando nosso dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "74fYM4vIM6Si",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "history = model.fit(train_data, train_targets, epochs=2000, verbose=0)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "waxQtW0C34iZ",
        "colab_type": "text"
      },
      "source": [
        "Treinado o modelo, podemos analisar seu desempenho através do gráfico abaixo."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ok3_ho_4M6Yt",
        "colab_type": "code",
        "outputId": "9adc6147-e231-4cac-9b6a-7ed1ce40f528",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "plt.plot(history.history['loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "Y-Jvd0DT311e",
        "colab_type": "text"
      },
      "source": [
        "Para avaliar o modelo, podemos obter o valor do erro como segue."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fWBMgM5Q32FT",
        "colab_type": "code",
        "outputId": "47f5aa15-9849-421c-e214-7228c37ca3b9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 50
        }
      },
      "source": [
        "test_mse_score = model.evaluate(test_data, test_targets)\n",
        "test_mse_score"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "102/102 [==============================] - 0s 182us/step\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[15.152721554625268, 15.152722358703613]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kjL97zHB4J35",
        "colab_type": "text"
      },
      "source": [
        "Note que o erro acima está alto. Mas ainda assim, é uma aproximação razoável.\n",
        "\n",
        "Podemos ter uma ideia melhor deste erro através do gráfico abaixo. Quanto mais distante da linha, maior o erro."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G1D-t_DS2AHL",
        "colab_type": "code",
        "outputId": "69c78616-6c41-47e2-f46b-b86a4d645831",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        }
      },
      "source": [
        "test_predictions = model.predict(test_data).flatten()\n",
        "plt.scatter(test_targets, test_predictions)\n",
        "plt.xlabel('True Values [1000$]')\n",
        "plt.ylabel('Predictions [1000$]')\n",
        "plt.axis('equal')\n",
        "plt.xlim(plt.xlim())\n",
        "plt.ylim(plt.ylim())\n",
        "_ = plt.plot([-100, 100],[-100,100])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "a8wK-o4gMBUs",
        "colab_type": "text"
      },
      "source": [
        "## Exercícios"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GARK4PZu_tve",
        "colab_type": "text"
      },
      "source": [
        "### Exercício 1\n",
        "\n",
        "Repita o exemplo 1 para a expressão lógica abaixo.\n",
        "\n",
        "![alt text](https://www.electronics-tutorials.ws/wp-content/uploads/2018/05/combination-comb25.gif)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ay8JlpvaX3va",
        "colab_type": "text"
      },
      "source": [
        "#### Modelagem usando os mesmos parâmetros usados no exemplo 1"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vCvkVPfHAGks",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "X = np.array([\n",
        "              [0, 0, 0],\n",
        "              [0, 0, 1],\n",
        "              [0, 1, 0],\n",
        "              [0, 1, 1],\n",
        "              [1, 0, 0],\n",
        "              [1, 0, 1],\n",
        "              [1, 1, 0],\n",
        "              [1, 1, 1]\n",
        "]).astype('float32')\n",
        "\n",
        "y = np.array([[0], [0], [0], [0], [1], [0], [0], [0]]).astype('float32')\n",
        "\n",
        "y = np_utils.to_categorical(y)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YMhmvHXISUch",
        "colab_type": "code",
        "outputId": "66e65af4-138b-4041-e3a1-a5a6a00d67d8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 151
        }
      },
      "source": [
        "print(y)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1. 0.]\n",
            " [1. 0.]\n",
            " [1. 0.]\n",
            " [1. 0.]\n",
            " [0. 1.]\n",
            " [1. 0.]\n",
            " [1. 0.]\n",
            " [1. 0.]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Bl4l3muhSoni",
        "colab_type": "code",
        "outputId": "f95fec81-69b5-492f-acf2-1513da51c008",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 218
        }
      },
      "source": [
        "model = Sequential()\n",
        "\n",
        "model.add(Dense(64, activation='tanh', input_dim=X.shape[1]))\n",
        "model.add(Dense(2, activation='sigmoid'))\n",
        "model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics = [\"binary_crossentropy\", \"accuracy\"])\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_18\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense_38 (Dense)             (None, 64)                256       \n",
            "_________________________________________________________________\n",
            "dense_39 (Dense)             (None, 2)                 130       \n",
            "=================================================================\n",
            "Total params: 386\n",
            "Trainable params: 386\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "20jVfxQcTMw1",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "history = model.fit(X, y, epochs=1000, verbose=0) "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uvENDr9jTRn-",
        "colab_type": "code",
        "outputId": "cf753671-63bf-43c7-8f8d-537a66d6264f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "code",
      "metadata": {
        "id": "XK7cyR-2TWUh",
        "colab_type": "code",
        "outputId": "56791ee0-ad70-4f62-d29b-ad74d111c88c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['accuracy'])\n",
        "plt.title('model accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "code",
      "metadata": {
        "id": "200Pc6QjTb9c",
        "colab_type": "code",
        "outputId": "650c97ed-0323-49fc-ef47-d238a8bb8e1e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        }
      },
      "source": [
        "# avalia o modelo\n",
        "score = model.evaluate(X, y)\n",
        "print(\"\\nAcurácia: \", score[-1])\n",
        "\n",
        "# verifica as predições\n",
        "print(\"\\nPredições:\")\n",
        "pred = model.predict_proba(X)\n",
        "print(pred)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r8/8 [==============================] - 0s 2ms/step\n",
            "\n",
            "Acurácia:  1.0\n",
            "\n",
            "Predições:\n",
            "[[9.9447173e-01 5.8782995e-03]\n",
            " [9.9999797e-01 2.4255924e-06]\n",
            " [9.9999803e-01 2.2446695e-06]\n",
            " [1.0000000e+00 1.2198366e-08]\n",
            " [1.2593597e-02 9.8602122e-01]\n",
            " [9.9500823e-01 5.7833791e-03]\n",
            " [9.9456584e-01 5.1988363e-03]\n",
            " [9.9999809e-01 2.2693991e-06]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zRcvTq-hTckT",
        "colab_type": "code",
        "outputId": "ce21ce4c-ab9a-4567-db5a-d79e14dd9a9c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 168
        }
      },
      "source": [
        "# imprime as predições de forma mais clara\n",
        "print(\"Entrada\\t\\tSaída esperada\\t\\tSaída prevista\")\n",
        "for i in range(len(X)):\n",
        "    print(f\"{X[i]}\\t\\t{y[i]}\\t\\t\\t{[round(v) for v in pred[i]]}\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Entrada\t\tSaída esperada\t\tSaída prevista\n",
            "[0. 0. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 0. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 1. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 1. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 0. 0.]\t\t[0. 1.]\t\t\t[0.0, 1.0]\n",
            "[1. 0. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 1. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 1. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yrdYXd_qX_vR",
        "colab_type": "text"
      },
      "source": [
        "#### Modelagem usando parâmetros diferentes\n",
        "\n",
        "Mudei para conter mais uma camada, e as duas ainda possuirem 200 neurônios cada. \n",
        "\n",
        "Para a modelagem foi usadado a função de perda `poisson` e para otimização a `adamax`, que se encaixaram melhor que as primeiras usadas.\n",
        "\n",
        "Por último, foi alterado a quantidade de rodadas para garantir a acurácia em 1 logo início das iterações."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "42sfibUPXNuS",
        "colab_type": "code",
        "outputId": "92ca74d6-e6da-44cd-fd54-6dfa9069900d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 252
        }
      },
      "source": [
        "model = Sequential()\n",
        "\n",
        "model.add(Dense(200, activation='tanh', input_dim=X.shape[1]))\n",
        "model.add(Dense(200, activation='tanh'))\n",
        "model.add(Dense(2, activation='sigmoid'))\n",
        "model.compile(loss=\"poisson\", optimizer=\"adamax\", metrics = [\"binary_crossentropy\", \"accuracy\"])\n",
        "history = model.fit(X, y, epochs=2000, verbose=0) \n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_25\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "dense_57 (Dense)             (None, 200)               800       \n",
            "_________________________________________________________________\n",
            "dense_58 (Dense)             (None, 200)               40200     \n",
            "_________________________________________________________________\n",
            "dense_59 (Dense)             (None, 2)                 402       \n",
            "=================================================================\n",
            "Total params: 41,402\n",
            "Trainable params: 41,402\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7NPoDRglXQnS",
        "colab_type": "code",
        "outputId": "db713a98-60f7-4e50-81eb-52adffe9780a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "code",
      "metadata": {
        "id": "LVwFpHT_XYeQ",
        "colab_type": "code",
        "outputId": "23f3c95b-f45e-4a01-e8c0-2aeaa1f76b96",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "plt.plot(history.history['accuracy'])\n",
        "plt.title('model accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "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": "code",
      "metadata": {
        "id": "4IUfzip1Xa1X",
        "colab_type": "code",
        "outputId": "f93a0b06-ef00-45a1-b4f4-07159095200a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        }
      },
      "source": [
        "# avalia o modelo\n",
        "score = model.evaluate(X, y)\n",
        "print(\"\\nAcurácia: \", score[-1])\n",
        "\n",
        "# verifica as predições\n",
        "print(\"\\nPredições:\")\n",
        "pred = model.predict_proba(X)\n",
        "print(pred)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r8/8 [==============================] - 0s 3ms/step\n",
            "\n",
            "Acurácia:  1.0\n",
            "\n",
            "Predições:\n",
            "[[9.9971521e-01 5.5489309e-05]\n",
            " [1.0000000e+00 3.5565890e-09]\n",
            " [1.0000000e+00 3.3997463e-09]\n",
            " [1.0000000e+00 1.4996764e-10]\n",
            " [5.6229670e-05 9.9568844e-01]\n",
            " [9.9671674e-01 1.4502803e-05]\n",
            " [9.9684745e-01 1.4425763e-05]\n",
            " [1.0000000e+00 1.7092356e-09]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M7U3LDlxXdCq",
        "colab_type": "code",
        "outputId": "c7220055-1372-4618-cfcd-a5b7ba277cb4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 168
        }
      },
      "source": [
        "# imprime as predições de forma mais clara\n",
        "print(\"Entrada\\t\\tSaída esperada\\t\\tSaída prevista\")\n",
        "for i in range(len(X)):\n",
        "    print(f\"{X[i]}\\t\\t{y[i]}\\t\\t\\t{[round(v) for v in pred[i]]}\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Entrada\t\tSaída esperada\t\tSaída prevista\n",
            "[0. 0. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 0. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 1. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[0. 1. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 0. 0.]\t\t[0. 1.]\t\t\t[0.0, 1.0]\n",
            "[1. 0. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 1. 0.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n",
            "[1. 1. 1.]\t\t[1. 0.]\t\t\t[1.0, 0.0]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lXPiNQjg_ty1",
        "colab_type": "text"
      },
      "source": [
        "### Exercício 2\n",
        "\n",
        "Construa um modelo capaz de fazer o reconhecimento de dígitos do dataset MNIST ([link](https://keras.io/api/datasets/mnist/))."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZbEVXjx5zCIf",
        "colab_type": "text"
      },
      "source": [
        "#### Imports"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xs07YUPiAJKp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "2797b58c-5aba-4cf0-9329-502cbafb4090"
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from numpy import mean\n",
        "from numpy import std\n",
        "from keras.models import Sequential\n",
        "from keras.layers.core import Dense, Activation\n",
        "from keras.utils import to_categorical\n",
        "from keras.utils import np_utils\n",
        "from keras.datasets import mnist\n",
        "from sklearn.model_selection import KFold\n",
        "from keras.layers import Conv2D\n",
        "from keras.layers import MaxPooling2D\n",
        "from keras.layers import Dense\n",
        "from keras.layers import Flatten\n",
        "from keras.optimizers import SGD\n",
        "from keras.models import load_model\n",
        "from keras.preprocessing.image import load_img\n",
        "from keras.preprocessing.image import img_to_array"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "npgTj9pv44Wt",
        "colab_type": "code",
        "outputId": "6050937c-dde0-4dce-b01e-db4acaf6fe4d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "(train_data, train_targets), (test_data, test_targets) = mnist.load_data()"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 1s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k7xb3B8X6ak5",
        "colab_type": "code",
        "outputId": "7a4bdb66-9ffb-4d90-b67a-61a68e23a589",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "print('Shape dados de treino:   in: %s      out: %s'% (train_data.shape, train_targets.shape))\n",
        "print('Shape dados de teste:    in: %s      out: %s'% (test_data.shape, test_targets.shape))\n",
        "print('Exemplo de entrada:      %s'% (train_data[0]))\n",
        "print('Exemplo de saída:        %f'% (train_targets[0]))"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Shape dados de treino:   in: (60000, 28, 28)      out: (60000,)\n",
            "Shape dados de teste:    in: (10000, 28, 28)      out: (10000,)\n",
            "Exemplo de entrada:      [[  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   3  18  18  18 126 136\n",
            "  175  26 166 255 247 127   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0  30  36  94 154 170 253 253 253 253 253\n",
            "  225 172 253 242 195  64   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0  49 238 253 253 253 253 253 253 253 253 251\n",
            "   93  82  82  56  39   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0  18 219 253 253 253 253 253 198 182 247 241\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0  80 156 107 253 253 205  11   0  43 154\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0  14   1 154 253  90   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0 139 253 190   2   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0  11 190 253  70   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0  35 241 225 160 108   1\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0  81 240 253 253 119\n",
            "   25   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0  45 186 253 253\n",
            "  150  27   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0  16  93 252\n",
            "  253 187   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 249\n",
            "  253 249  64   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0  46 130 183 253\n",
            "  253 207   2   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0  39 148 229 253 253 253\n",
            "  250 182   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0  24 114 221 253 253 253 253 201\n",
            "   78   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0  23  66 213 253 253 253 253 198  81   2\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0  18 171 219 253 253 253 253 195  80   9   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0  55 172 226 253 253 253 253 244 133  11   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0 136 253 253 253 212 135 132  16   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]\n",
            " [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0\n",
            "    0   0   0   0   0   0   0   0   0   0]]\n",
            "Exemplo de saída:        5.000000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BthQYLq3E3j_",
        "colab_type": "code",
        "outputId": "b7821b75-f20f-4606-d940-6e911279990b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 248
        }
      },
      "source": [
        "for i in range(2):\n",
        "\tplt.axis(\"off\")\n",
        "\tplt.imshow(train_data2[i], cmap=plt.get_cmap('gray'))\n",
        "\tplt.savefig('sample_' + str(i) + '.png')\n",
        "plt.show()"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAOcAAADnCAYAAADl9EEgAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAGc0lEQVR4nO3dOWhVfx7G4bmjWChqSKMgiGihqEgaFUQQkSCCFlGbgJViZcAqjZ1FRHApRItUgo1YujRaxKUQBHFpAvZKOo1L3Ii50w0M5H7zN8vkvcnzlHk5nlP44YA/Tmw0m81/AXn+Pd8PAExOnBBKnBBKnBBKnBBqaTU2Gg3/lAtzrNlsNib7uTcnhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhBInhFo63w/A/1qyZEm5r169ek7v39fX13Jbvnx5ee3mzZvL/cyZM+V++fLllltvb2957c+fP8v94sWL5X7+/Plynw/enBBKnBBKnBBKnBBKnBBKnBBKnBDKOeck1q9fX+7Lli0r9z179pT73r17W24dHR3ltceOHSv3+fT+/ftyv3btWrn39PS03L5+/Vpe+/bt23J/+vRpuSfy5oRQ4oRQ4oRQ4oRQ4oRQ4oRQjWaz2XpsNFqPbayrq6vch4aGyn2uP9tKNTExUe4nT54s92/fvk373iMjI+X+6dOncn/37t207z3Xms1mY7Kfe3NCKHFCKHFCKHFCKHFCKHFCKHFCqEV5ztnZ2VnuL168KPeNGzfO5uPMqqmefXR0tNz379/fcvv9+3d57WI9/50p55zQZsQJocQJocQJocQJocQJocQJoRblr8b8+PFjuff395f74cOHy/3169flPtWviKy8efOm3Lu7u8t9bGys3Ldt29ZyO3v2bHkts8ubE0KJE0KJE0KJE0KJE0KJE0KJE0Ityu85Z2rVqlXlPtV/Vzc4ONhyO3XqVHntiRMnyv327dvlTh7fc0KbESeEEieEEieEEieEEieEEieEWpTfc87Uly9fZnT958+fp33t6dOny/3OnTvlPtX/sUkOb04IJU4IJU4IJU4IJU4IJU4I5ZOxebBixYqW2/3798tr9+3bV+6HDh0q90ePHpU7/38+GYM2I04IJU4IJU4IJU4IJU4IJU4I5ZwzzKZNm8r91atX5T46Olrujx8/LveXL1+23G7cuFFeW/1dojXnnNBmxAmhxAmhxAmhxAmhxAmhxAmhnHO2mZ6ennK/efNmua9cuXLa9z537ly537p1q9xHRkamfe+FzDkntBlxQihxQihxQihxQihxQihxQijnnAvM9u3by/3q1avlfuDAgWnfe3BwsNwHBgbK/cOHD9O+dztzzgltRpwQSpwQSpwQSpwQSpwQSpwQyjnnItPR0VHuR44cablN9a1oozHpcd1/DQ0NlXt3d3e5L1TOOaHNiBNCiRNCiRNCiRNCiRNCOUrhH/v161e5L126tNzHx8fL/eDBgy23J0+elNe2M0cp0GbECaHECaHECaHECaHECaHECaHqgynazo4dO8r9+PHj5b5z586W21TnmFMZHh4u92fPns3oz19ovDkhlDghlDghlDghlDghlDghlDghlHPOMJs3by73vr6+cj969Gi5r1279q+f6Z/68+dPuY+MjJT7xMTEbD5O2/PmhFDihFDihFDihFDihFDihFDihFDOOefAVGeJvb29LbepzjE3bNgwnUeaFS9fviz3gYGBcr93795sPs6C580JocQJocQJocQJocQJocQJoRylTGLNmjXlvnXr1nK/fv16uW/ZsuWvn2m2vHjxotwvXbrUcrt79255rU++Zpc3J4QSJ4QSJ4QSJ4QSJ4QSJ4QSJ4RasOecnZ2dLbfBwcHy2q6urnLfuHHjtJ5pNjx//rzcr1y5Uu4PHz4s9x8/fvz1MzE3vDkhlDghlDghlDghlDghlDghlDghVOw55+7du8u9v7+/3Hft2tVyW7du3bSeabZ8//695Xbt2rXy2gsXLpT72NjYtJ6JPN6cEEqcEEqcEEqcEEqcEEqcEEqcECr2nLOnp2dG+0wMDw+X+4MHD8p9fHy83KtvLkdHR8trWTy8OSGUOCGUOCGUOCGUOCGUOCGUOCFUo9lsth4bjdYjMCuazWZjsp97c0IocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUIocUKo8ldjAvPHmxNCiRNCiRNCiRNCiRNCiRNC/QfM6zUP81ILVgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QsGw3Lwq7Zxo",
        "colab_type": "text"
      },
      "source": [
        "Usando `reshape` para ter um único canal de cores."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UP7MRKHsu0VB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_data = train_data.reshape((train_data.shape[0], 28, 28, 1))\n",
        "test_data = test_data.reshape((test_data.shape[0], 28, 28, 1))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "35N2Yc-z7uAa",
        "colab_type": "text"
      },
      "source": [
        "Visto que possuímos 10 classes, separadas em números inteiros, podemos então aplicar a técnica de *hot encoding* e transformar em um vetor binário, onde 1 sempre vai representar uma classe."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h0gxjjtzvPL2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_targets = to_categorical(train_targets)\n",
        "test_targets = to_categorical(test_targets)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lNlaq4lh8Skq",
        "colab_type": "text"
      },
      "source": [
        "O dataset possui inteiros entre 0 e 255, contudo, não é uma boa prática usar esta escala para modelagem, então é necessário mudar a escala. \n",
        "Para isso normalizamos os pixels para escala de cinza, sendo usado o range [0,1].\n",
        "Nesse processo devemos então mudar os dados de inteiros para decimais, e dividir pelo maior valor, que no caso é 255."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "po6mTJ_1ymuo",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# convert from integers to floats\n",
        "train_data = train_data.astype('float32')\n",
        "test_data = test_data.astype('float32')\n",
        "# normalize to range 0-1\n",
        "train_data = train_data / 255.0\n",
        "test_data = test_data / 255.0"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ymwmosh5946_",
        "colab_type": "text"
      },
      "source": [
        "#### Modelo\n",
        "\n",
        "Para definir o modelo, segundo a literatura, é importante se preocupar com dois aspectos: extração de recursos que é composta por camadas convolucionais e de pool (agrupamento), e o backend para fazer a predição.\n",
        "\n",
        "Na modelagem, é usado uma camada convolucional com um filtro de 3x3, seguido por um pool de 2x2. Uma camada densa de 100 neurônios. E uma saída com 10 neurônios, já possuímos 10 classes de saída.\n",
        "\n",
        "Para copilar o modelo, foi usada uma configuração para otimizar a descida de gradiente, com taxa de aprendizado de 0,01 e peso de 0,9."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TSc4Omy1zwPZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# define cnn model\n",
        "def define_model():\n",
        "\tmodel = Sequential()\n",
        "\tmodel.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', input_shape=(28, 28, 1)))\n",
        "\tmodel.add(MaxPooling2D((2, 2)))\n",
        "\tmodel.add(Flatten())\n",
        "\tmodel.add(Dense(100, activation='relu', kernel_initializer='he_uniform'))\n",
        "\tmodel.add(Dense(10, activation='softmax'))\n",
        "\t# compile model\n",
        "\topt = SGD(lr=0.01, momentum=0.9)\n",
        "\tmodel.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "\treturn model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FFybtBa2HzOO",
        "colab_type": "text"
      },
      "source": [
        "#### Avaliação\n",
        "\n",
        "Para realizar a avaliação, é usado um método chamado [five-fold cross-validation](https://https://machinelearningmastery.com/k-fold-cross-validation/). Foi usado então o valor k=5 como baseline para repetir a avaliação e não gerar uma execução tão longa.\n",
        "Cada teste é setado para 20% do dataset de treino (nesse caso fica em 12.000 amostras). \n",
        "\n",
        "Para realizar a divisão dos dados de treino, usando k, é embaralhado de forma a garantir que os dados não estarão repetidos em cada dataset (treino e teste) de cada *k fold*.\n",
        "\n",
        "É treinado então para 10 épocas com um batch de 32 exemplos. O set de teste então será utilizado para avaliar o modelo em cada época (permitindo criar uma curva de aprendizado) e no final da avaliação (possibilita um histórico de cada execução, assim como a acurácia de cada k)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XTWfQPJHz6En",
        "colab_type": "code",
        "outputId": "44f8aef4-008e-4001-86a0-a962a986a27f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 104
        }
      },
      "source": [
        "n_folds=5\n",
        "\n",
        "scores, histories = list(), list()\n",
        "# prepare cross validation\n",
        "kfold = KFold(n_folds, shuffle=True, random_state=1)\n",
        "# enumerate splits\n",
        "for train_ix, test_ix in kfold.split(train_data):\n",
        "\t# define model\n",
        "\tmodel = define_model()\n",
        "\t# select rows for train and test\n",
        "\ttrainX, trainY, testX, testY = train_data[train_ix], train_targets[train_ix], train_data[test_ix], train_targets[test_ix]\n",
        "\t# fit model\n",
        "\thistory = model.fit(trainX, trainY, epochs=10, batch_size=32, validation_data=(testX, testY), verbose=0)\n",
        "\t# evaluate model\n",
        "\t_, acc = model.evaluate(testX, testY, verbose=0)\n",
        "\tprint('> %.3f' % (acc * 100.0))\n",
        "\t# stores scores\n",
        "\tscores.append(acc)\n",
        "\thistories.append(history)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "> 98.483\n",
            "> 98.442\n",
            "> 98.600\n",
            "> 98.725\n",
            "> 98.733\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "baAddDg6Ma-y",
        "colab_type": "text"
      },
      "source": [
        "Abaixo foi extraído o comportamento do modelo durante o treinamento e uma estimativa de performance do mesmo.\n",
        "\n",
        "Os gráficos abaixo mostram a perda (Entropy Loss) e acurácia (Classification Accuracy).\n",
        "\n",
        "Azul indica avaliação realizada sobre o dataset de treino, e laranja sobre o dataset de treino."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n5wEvkfl0yTB",
        "colab_type": "code",
        "outputId": "466bf907-309e-4be6-8129-0bab8c6f3a76",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 368
        }
      },
      "source": [
        "for i in range(len(histories)):\n",
        "\t# plot loss\n",
        "\tplt.subplot(2, 1, 1)\n",
        "\tplt.title('Cross Entropy Loss')\n",
        "\tplt.plot(histories[i].history['loss'], color='blue', label='train')\n",
        "\tplt.plot(histories[i].history['val_loss'], color='orange', label='test')\n",
        "\t# plot accuracy\n",
        "\tplt.subplot(2, 1, 2)\n",
        "\tplt.title('Classification Accuracy')\n",
        "\tplt.plot(histories[i].history['accuracy'], color='blue', label='train')\n",
        "\tplt.plot(histories[i].history['val_accuracy'], color='orange', label='test')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:3: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
            "  This is separate from the ipykernel package so we can avoid doing imports until\n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:8: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
            "  \n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ce8jEOMEc-vS",
        "colab_type": "text"
      },
      "source": [
        "Para a acurácia, ainda podemos classificar a média e desvio padrão. Isso fornece uma estimativa da performance média esperada do modelo treinado. \n",
        "\n",
        "Por fim, plotamos um gráfico whisker plot."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8buT8zdkbT1U",
        "colab_type": "code",
        "outputId": "3e01cd02-f633-432a-ad99-2695bf858074",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        }
      },
      "source": [
        "# print summary\n",
        "print('Accuracy: mean=%.3f std=%.3f, n=%d' % (mean(scores)*100, std(scores)*100, len(scores)))\n",
        "# box and whisker plots of results\n",
        "plt.boxplot(scores)\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy: mean=98.717 std=0.109, n=5\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"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k7cdvn8Cg1Tg",
        "colab_type": "text"
      },
      "source": [
        "Como podemos ver nas avaliações realizadas, o modelo avaliou com algo por volta de 98 (ou 0.98 para 1). Podemos então fazer algumas mudanças na modelagem para ter resultados melhores.\n",
        "\n",
        "Um deles é aumentar a profundidade da cnn adicionando mais convolucional e pool layers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yc_dSy5KjVLm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# define cnn model\n",
        "def define_model():\n",
        "\tmodel = Sequential()\n",
        "\tmodel.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', input_shape=(28, 28, 1)))\n",
        "\tmodel.add(MaxPooling2D((2, 2)))\n",
        "\tmodel.add(Conv2D(64, (3, 3), activation='relu', kernel_initializer='he_uniform'))\n",
        "\tmodel.add(Conv2D(64, (3, 3), activation='relu', kernel_initializer='he_uniform'))\n",
        "\tmodel.add(MaxPooling2D((2, 2)))\n",
        "\tmodel.add(Conv2D(128, (3, 3), activation='relu', kernel_initializer='he_uniform'))\n",
        "\tmodel.add(MaxPooling2D((2, 2)))\n",
        "\tmodel.add(Flatten())\n",
        "\tmodel.add(Dense(100, activation='relu', kernel_initializer='he_uniform'))\n",
        "\tmodel.add(Dense(10, activation='softmax'))\n",
        "\t# compile model\n",
        "\topt = SGD(lr=0.01, momentum=0.9)\n",
        "\tmodel.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "\treturn model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TPqwh01Rjqc6",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 104
        },
        "outputId": "ff5466c1-f3fd-4e54-ecfc-c21cca284475"
      },
      "source": [
        "n_folds=5\n",
        "\n",
        "scores, histories = list(), list()\n",
        "# prepare cross validation\n",
        "kfold = KFold(n_folds, shuffle=True, random_state=1)\n",
        "# enumerate splits\n",
        "for train_ix, test_ix in kfold.split(train_data):\n",
        "\t# define model\n",
        "\tmodel = define_model()\n",
        "\t# select rows for train and test\n",
        "\ttrainX, trainY, testX, testY = train_data[train_ix], train_targets[train_ix], train_data[test_ix], train_targets[test_ix]\n",
        "\t# fit model\n",
        "\thistory = model.fit(trainX, trainY, epochs=10, batch_size=32, validation_data=(testX, testY), verbose=0)\n",
        "\t# evaluate model\n",
        "\t_, acc = model.evaluate(testX, testY, verbose=0)\n",
        "\tprint('> %.3f' % (acc * 100.0))\n",
        "\t# stores scores\n",
        "\tscores.append(acc)\n",
        "\thistories.append(history)"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "> 99.067\n",
            "> 99.033\n",
            "> 98.850\n",
            "> 99.075\n",
            "> 98.983\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MoTZoqPLn9bQ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 335
        },
        "outputId": "0d4944db-43a3-4969-b070-05bc9e5dc7ef"
      },
      "source": [
        "for i in range(len(histories)):\n",
        "\t# plot loss\n",
        "\tplt.subplot(1, 1, 1)\n",
        "\tplt.title('Cross Entropy Loss')\n",
        "\tplt.plot(histories[i].history['loss'], color='blue', label='train')\n",
        "\tplt.plot(histories[i].history['val_loss'], color='orange', label='test')\n",
        "plt.show()"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:3: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
            "  This is separate from the ipykernel package so we can avoid doing imports until\n"
          ],
          "name": "stderr"
        },
        {
          "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": "code",
      "metadata": {
        "id": "hkXWwy1HrmgR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 335
        },
        "outputId": "a03c7087-ec27-4d85-c319-54ae6a54857d"
      },
      "source": [
        "for i in range(len(histories)):\n",
        "\t# plot accuracy\n",
        "\tplt.subplot(1, 1, 1)\n",
        "\tplt.title('Classification Accuracy')\n",
        "\tplt.plot(histories[i].history['accuracy'], color='blue', label='train')\n",
        "\tplt.plot(histories[i].history['val_accuracy'], color='orange', label='test')\n",
        "plt.show()"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:3: MatplotlibDeprecationWarning: Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
            "  This is separate from the ipykernel package so we can avoid doing imports until\n"
          ],
          "name": "stderr"
        },
        {
          "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": "code",
      "metadata": {
        "id": "whv10p_nsXAK",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "34a2df79-4547-428c-eade-8d04055abe6d"
      },
      "source": [
        "# print summary\n",
        "print('Accuracy: mean=%.3f std=%.3f, n=%d' % (mean(scores)*100, std(scores)*100, len(scores)))\n",
        "# box and whisker plots of results\n",
        "plt.boxplot(scores)\n",
        "plt.show()"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy: mean=99.002 std=0.082, n=5\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"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "913S4DTjsvbp",
        "colab_type": "text"
      },
      "source": [
        "Logo, podemos notar que com o aumento de camadas na cnn, conseguimos um aumento significativo na acurácia da rede."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y77cv2KWuXot",
        "colab_type": "text"
      },
      "source": [
        "#### Finalizando o modelo\n",
        "\n",
        "Já com o modelo finalizado, usaremos ele para conjunto de treinamento."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HR65hxRlssyc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model.fit(train_data, train_targets, epochs=10, batch_size=32, verbose=0)\n",
        "model.save('final_model.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DB_UKlq0vhwg",
        "colab_type": "text"
      },
      "source": [
        "Agora podemos avaliar o modelo usando o dataset de teste."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bbqSch3mu4Bq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "7ddc35c5-058b-4de0-c6b6-40461beae69f"
      },
      "source": [
        "model_loaded = load_model('final_model.h5')\n",
        "# evaluate model on test dataset\n",
        "_, acc = model_loaded.evaluate(test_data, test_targets, verbose=0)\n",
        "print('> %.3f' % (acc * 100.0))"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "> 99.140\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Cw4MefSwwRD2",
        "colab_type": "text"
      },
      "source": [
        "#### Fazendo predição usando imagens\n",
        "\n",
        "Para o modelo criado, prevê que a imagem seja em escala de cinza, alinhada contendo uma escrita do digito, e que a imagem tenha tamanho de 28x28 pixels.\n",
        "\n",
        "O código abaixo irá mostrar isso, carregando uma imagem e setando ela para o tamanho necessário, mudando para ter apenas um canal.\n",
        "É importante ser feito isso porque o modelo foi treinado com essas características, logo as imagens que são usadas para predizer, também devem estar no mesmo formato."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "85yKNAM1wrbi",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def load_image(filename):\n",
        "\timg = load_img(filename, color_mode=\"grayscale\", target_size=(28, 28))\n",
        "\timg = img_to_array(img)\n",
        "\timg = img.reshape(1, 28, 28, 1)\n",
        "\t# normalizing data\n",
        "\timg = img.astype('float32')\n",
        "\timg = img / 255.0\n",
        "\treturn img"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Qid045zaySir",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "361f6d44-59ef-418e-dc34-451c4990bc0d"
      },
      "source": [
        "# predict the class\n",
        "img = load_img(\"digit_7.png\")\n",
        "digit = model_loaded.predict_classes(img)\n",
        "print(digit[0])"
      ],
      "execution_count": 71,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "7\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}