mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow import keras\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. Buscando o Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist = keras.datasets.mnist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "(train_images, train_labels), (test_images, test_labels) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Normalizando os dados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_images = train_images / 255.0\n",
    "test_images = test_images / 255.0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Corrigindo canais de cores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)\n",
    "test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Modelo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelo = keras.Sequential([\n",
    "    keras.layers.Flatten(input_shape = (28, 28)), # Decompondo as Imagens\n",
    "    keras.layers.Dense(128, activation = \"relu\"),\n",
    "    keras.layers.Dense(64, activation = \"relu\"),\n",
    "    keras.layers.Dense(32, activation = \"relu\"),\n",
    "    keras.layers.Dense(10, activation = \"softmax\") # Camada de saída\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss', patience=10, restore_best_weights=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelo.compile(optimizer='nadam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 2.1 Treinamento"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "np.random.seed(30912)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1500/1500 [==============================] - 5s 3ms/step - loss: 0.2768 - accuracy: 0.9187 - val_loss: 0.1388 - val_accuracy: 0.9566\n",
      "Epoch 2/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.1152 - accuracy: 0.9654 - val_loss: 0.1095 - val_accuracy: 0.9685\n",
      "Epoch 3/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0793 - accuracy: 0.9753 - val_loss: 0.1003 - val_accuracy: 0.9697\n",
      "Epoch 4/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0603 - accuracy: 0.9814 - val_loss: 0.0996 - val_accuracy: 0.9712\n",
      "Epoch 5/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0467 - accuracy: 0.9848 - val_loss: 0.1138 - val_accuracy: 0.9693\n",
      "Epoch 6/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0403 - accuracy: 0.9864 - val_loss: 0.1023 - val_accuracy: 0.9732\n",
      "Epoch 7/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0345 - accuracy: 0.9881 - val_loss: 0.1252 - val_accuracy: 0.9694\n",
      "Epoch 8/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0287 - accuracy: 0.9903 - val_loss: 0.1101 - val_accuracy: 0.9741\n",
      "Epoch 9/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0254 - accuracy: 0.9915 - val_loss: 0.1330 - val_accuracy: 0.9702\n",
      "Epoch 10/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0252 - accuracy: 0.9918 - val_loss: 0.1112 - val_accuracy: 0.9745\n",
      "Epoch 11/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0196 - accuracy: 0.9933 - val_loss: 0.1479 - val_accuracy: 0.9707\n",
      "Epoch 12/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0199 - accuracy: 0.9932 - val_loss: 0.1534 - val_accuracy: 0.9687\n",
      "Epoch 13/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0204 - accuracy: 0.9932 - val_loss: 0.1267 - val_accuracy: 0.9745\n",
      "Epoch 14/100\n",
      "1500/1500 [==============================] - 4s 3ms/step - loss: 0.0157 - accuracy: 0.9945 - val_loss: 0.1258 - val_accuracy: 0.9745\n"
     ]
    }
   ],
   "source": [
    "treinamento = modelo.fit(train_images, train_labels, epochs=100, validation_split=0.2, callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 576x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "pd.DataFrame(treinamento.history).plot(figsize = (8, 5))\n",
    "plt.grid(True)\n",
    "plt.gca().set_ylim(0, 1) # Restringindo o Eixo y\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Teste Inicial"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 0s - loss: 0.0953 - accuracy: 0.9741 - 456ms/epoch - 1ms/step\n",
      "\n",
      "Resultado do Teste: 0.9741\n"
     ]
    }
   ],
   "source": [
    "test_loss, test_acc = modelo.evaluate(test_images, test_labels, verbose=2)\n",
    "print(f'\\nResultado do Teste: {test_acc:.4f}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "O objetivo é 98%. Vamos modificar a rede."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 2.2 Otimizando a Rede"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras_tuner.tuners import RandomSearch\n",
    "\n",
    "def construir_modelo(hp):\n",
    "    modelo = keras.Sequential()\n",
    "    modelo.add(keras.layers.Flatten(input_shape = (28, 28)))\n",
    "    \n",
    "    unidades = hp.Int('Tamanho da Camada', min_value=32, max_value=512, step=32)\n",
    "    ativacao = hp.Choice('Função de Ativação', values=[\"relu\", \"elu\"])\n",
    "\n",
    "    modelo.add(keras.layers.Dense(units=unidades, activation=ativacao))\n",
    "    modelo.add(keras.layers.Dense(units=unidades, activation=ativacao))\n",
    "    modelo.add(keras.layers.Dense(units=unidades, activation=ativacao))\n",
    "    modelo.add(keras.layers.Dense(10, activation='softmax'))\n",
    "\n",
    "    # Tune a taxa de aprendizado\n",
    "    \n",
    "    modelo.compile(optimizer=\"nadam\",\n",
    "                   loss='sparse_categorical_crossentropy',\n",
    "                   metrics=['accuracy'])\n",
    "    \n",
    "    return modelo"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Reduzindo a paciência"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial 10 Complete [00h 01m 09s]\n",
      "val_accuracy: 0.9746666550636292\n",
      "\n",
      "Best val_accuracy So Far: 0.9782500267028809\n",
      "Total elapsed time: 00h 12m 23s\n"
     ]
    }
   ],
   "source": [
    "tuner = RandomSearch(\n",
    "    construir_modelo,\n",
    "    objective='val_accuracy',\n",
    "    max_trials=10\n",
    ")\n",
    "\n",
    "tuner.search(train_images, train_labels, epochs=100, validation_split=0.2, callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "512\n",
      "relu\n"
     ]
    }
   ],
   "source": [
    "parametros = tuner.get_best_hyperparameters()[0]\n",
    "print(parametros.get('Tamanho da Camada'))\n",
    "print(parametros.get('Função de Ativação'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelo = keras.Sequential()\n",
    "modelo.add(keras.layers.Flatten(input_shape = (28, 28)))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(10, activation='softmax'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelo.compile(optimizer='nadam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1500/1500 [==============================] - 13s 8ms/step - loss: 0.2072 - accuracy: 0.9367 - val_loss: 0.1133 - val_accuracy: 0.9662\n",
      "Epoch 2/100\n",
      "1500/1500 [==============================] - 11s 8ms/step - loss: 0.0961 - accuracy: 0.9708 - val_loss: 0.1057 - val_accuracy: 0.9700\n",
      "Epoch 3/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0682 - accuracy: 0.9794 - val_loss: 0.1055 - val_accuracy: 0.9698\n",
      "Epoch 4/100\n",
      "1500/1500 [==============================] - 13s 8ms/step - loss: 0.0535 - accuracy: 0.9837 - val_loss: 0.1088 - val_accuracy: 0.9722\n",
      "Epoch 5/100\n",
      "1500/1500 [==============================] - 13s 8ms/step - loss: 0.0457 - accuracy: 0.9855 - val_loss: 0.0932 - val_accuracy: 0.9745\n",
      "Epoch 6/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0375 - accuracy: 0.9889 - val_loss: 0.0893 - val_accuracy: 0.9768\n",
      "Epoch 7/100\n",
      "1500/1500 [==============================] - 13s 8ms/step - loss: 0.0336 - accuracy: 0.9903 - val_loss: 0.1209 - val_accuracy: 0.9721\n",
      "Epoch 8/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0315 - accuracy: 0.9902 - val_loss: 0.0994 - val_accuracy: 0.9771\n",
      "Epoch 9/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0267 - accuracy: 0.9918 - val_loss: 0.1306 - val_accuracy: 0.9751\n",
      "Epoch 10/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0266 - accuracy: 0.9925 - val_loss: 0.1525 - val_accuracy: 0.9707\n",
      "Epoch 11/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0245 - accuracy: 0.9931 - val_loss: 0.1050 - val_accuracy: 0.9781\n",
      "Epoch 12/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0211 - accuracy: 0.9944 - val_loss: 0.1576 - val_accuracy: 0.9757\n",
      "Epoch 13/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0224 - accuracy: 0.9940 - val_loss: 0.1454 - val_accuracy: 0.9745\n",
      "Epoch 14/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0184 - accuracy: 0.9950 - val_loss: 0.1487 - val_accuracy: 0.9768\n",
      "Epoch 15/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0205 - accuracy: 0.9942 - val_loss: 0.1688 - val_accuracy: 0.9758\n",
      "Epoch 16/100\n",
      "1500/1500 [==============================] - 12s 8ms/step - loss: 0.0180 - accuracy: 0.9955 - val_loss: 0.1166 - val_accuracy: 0.9787\n"
     ]
    }
   ],
   "source": [
    "treinamento = modelo.fit(train_images, train_labels, epochs=100, validation_split=0.2, callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 1s - loss: 0.0846 - accuracy: 0.9788 - 906ms/epoch - 3ms/step\n",
      "\n",
      "Resultado do Teste: 0.9788\n"
     ]
    }
   ],
   "source": [
    "test_loss, test_acc = modelo.evaluate(test_images, test_labels, verbose=2)\n",
    "print(f'\\nResultado do Teste: {test_acc:.4f}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Testando outro otimizador"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1500/1500 [==============================] - 10s 6ms/step - loss: 0.2255 - accuracy: 0.9329 - val_loss: 0.1916 - val_accuracy: 0.9454\n",
      "Epoch 2/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.1005 - accuracy: 0.9728 - val_loss: 0.1386 - val_accuracy: 0.9602\n",
      "Epoch 3/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0770 - accuracy: 0.9790 - val_loss: 0.1295 - val_accuracy: 0.9722\n",
      "Epoch 4/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0662 - accuracy: 0.9833 - val_loss: 0.1251 - val_accuracy: 0.9748\n",
      "Epoch 5/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0499 - accuracy: 0.9869 - val_loss: 0.1205 - val_accuracy: 0.9778\n",
      "Epoch 6/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0440 - accuracy: 0.9887 - val_loss: 0.1377 - val_accuracy: 0.9761\n",
      "Epoch 7/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0376 - accuracy: 0.9908 - val_loss: 0.1447 - val_accuracy: 0.9772\n",
      "Epoch 8/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0308 - accuracy: 0.9922 - val_loss: 0.1718 - val_accuracy: 0.9787\n",
      "Epoch 9/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0272 - accuracy: 0.9937 - val_loss: 0.1508 - val_accuracy: 0.9772\n",
      "Epoch 10/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0215 - accuracy: 0.9947 - val_loss: 0.1744 - val_accuracy: 0.9785\n",
      "Epoch 11/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0181 - accuracy: 0.9956 - val_loss: 0.2075 - val_accuracy: 0.9765\n",
      "Epoch 12/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0179 - accuracy: 0.9959 - val_loss: 0.2016 - val_accuracy: 0.9778\n",
      "Epoch 13/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0161 - accuracy: 0.9964 - val_loss: 0.2239 - val_accuracy: 0.9767\n",
      "Epoch 14/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0121 - accuracy: 0.9972 - val_loss: 0.1970 - val_accuracy: 0.9793\n",
      "Epoch 15/100\n",
      "1500/1500 [==============================] - 9s 6ms/step - loss: 0.0146 - accuracy: 0.9968 - val_loss: 0.2390 - val_accuracy: 0.9768\n"
     ]
    }
   ],
   "source": [
    "modelo = keras.Sequential()\n",
    "modelo.add(keras.layers.Flatten(input_shape = (28, 28)))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(10, activation='softmax'))\n",
    "\n",
    "modelo.compile(optimizer='rmsprop', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "treinamento = modelo.fit(train_images, train_labels, epochs=100, validation_split=0.2, callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 1s - loss: 0.1130 - accuracy: 0.9772 - 854ms/epoch - 3ms/step\n",
      "\n",
      "Resultado do Teste: 0.9772\n"
     ]
    }
   ],
   "source": [
    "test_loss, test_acc = modelo.evaluate(test_images, test_labels, verbose=2)\n",
    "print(f'\\nResultado do Teste: {test_acc:.4f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "1500/1500 [==============================] - 8s 5ms/step - loss: 0.2892 - accuracy: 0.9122 - val_loss: 0.1160 - val_accuracy: 0.9641\n",
      "Epoch 2/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.1061 - accuracy: 0.9668 - val_loss: 0.1171 - val_accuracy: 0.9648\n",
      "Epoch 3/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0701 - accuracy: 0.9778 - val_loss: 0.0966 - val_accuracy: 0.9741\n",
      "Epoch 4/100\n",
      "1500/1500 [==============================] - 8s 5ms/step - loss: 0.0505 - accuracy: 0.9841 - val_loss: 0.0920 - val_accuracy: 0.9742\n",
      "Epoch 5/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0388 - accuracy: 0.9880 - val_loss: 0.0928 - val_accuracy: 0.9758\n",
      "Epoch 6/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0312 - accuracy: 0.9900 - val_loss: 0.0871 - val_accuracy: 0.9778\n",
      "Epoch 7/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0233 - accuracy: 0.9928 - val_loss: 0.0847 - val_accuracy: 0.9784\n",
      "Epoch 8/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0222 - accuracy: 0.9934 - val_loss: 0.0944 - val_accuracy: 0.9757\n",
      "Epoch 9/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0134 - accuracy: 0.9955 - val_loss: 0.1364 - val_accuracy: 0.9733\n",
      "Epoch 10/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0178 - accuracy: 0.9946 - val_loss: 0.0981 - val_accuracy: 0.9782\n",
      "Epoch 11/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0133 - accuracy: 0.9960 - val_loss: 0.0871 - val_accuracy: 0.9805\n",
      "Epoch 12/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0070 - accuracy: 0.9978 - val_loss: 0.1065 - val_accuracy: 0.9792\n",
      "Epoch 13/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0109 - accuracy: 0.9966 - val_loss: 0.0984 - val_accuracy: 0.9804\n",
      "Epoch 14/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0111 - accuracy: 0.9967 - val_loss: 0.1013 - val_accuracy: 0.9805\n",
      "Epoch 15/100\n",
      "1500/1500 [==============================] - 7s 5ms/step - loss: 0.0056 - accuracy: 0.9982 - val_loss: 0.1149 - val_accuracy: 0.9787\n",
      "Epoch 16/100\n",
      "1500/1500 [==============================] - 8s 5ms/step - loss: 0.0083 - accuracy: 0.9977 - val_loss: 0.1110 - val_accuracy: 0.9785\n",
      "Epoch 17/100\n",
      "1500/1500 [==============================] - 8s 5ms/step - loss: 0.0055 - accuracy: 0.9981 - val_loss: 0.1012 - val_accuracy: 0.9813\n"
     ]
    }
   ],
   "source": [
    "modelo = keras.Sequential()\n",
    "modelo.add(keras.layers.Flatten(input_shape = (28, 28)))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(512, activation=\"relu\"))\n",
    "modelo.add(keras.layers.Dense(10, activation='softmax'))\n",
    "\n",
    "modelo.compile(optimizer=keras.optimizers.SGD(learning_rate = 0.3), loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "treinamento = modelo.fit(train_images, train_labels, epochs=100, validation_split=0.2, callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "313/313 - 1s - loss: 0.0782 - accuracy: 0.9800 - 878ms/epoch - 3ms/step\n",
      "\n",
      "Resultado do Teste: 0.9800\n"
     ]
    }
   ],
   "source": [
    "test_loss, test_acc = modelo.evaluate(test_images, test_labels, verbose=2)\n",
    "print(f'\\nResultado do Teste: {test_acc:.4f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}