digit-recognizer.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import zipfile\n",
    "from pathlib import Path\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "\n",
    "from neural_network import DenseLayer, NeuralNetwork\n",
    "\n",
    "data_path = Path(\"data\")\n",
    "with zipfile.ZipFile(data_path / \"digit-recognizer.zip\", \"r\") as zip_ref:\n",
    "    zip_ref.extractall(\"data\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Carregando e processando os dados\n",
    "train = pd.read_csv(data_path / \"train.csv\")\n",
    "X_train = train.iloc[:, 1:].values / 255.0  # Normalização simples\n",
    "Y_train = train.iloc[:, 0].values.reshape(-1, 1)  # type: ignore\n",
    "\n",
    "# One-hot encoding das labels\n",
    "Y_train = np.eye(10)[\n",
    "    Y_train.flatten()\n",
    "].T  # One-hot encode e transpose para (n_classes, n_samples)\n",
    "\n",
    "# Normalização dos dados de entrada\n",
    "scaler = StandardScaler()\n",
    "X_train = scaler.fit_transform(X_train).T  # Transpor para (n_features, n_samples)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Inicializa a rede neural\n",
    "nn = NeuralNetwork()\n",
    "\n",
    "# Definir as camadas da rede neural\n",
    "layers = [\n",
    "    DenseLayer(input_size=784, output_size=128, activation=\"relu\"),\n",
    "    DenseLayer(input_size=128, output_size=64, activation=\"relu\"),\n",
    "    DenseLayer(input_size=64, output_size=10, activation=\"softmax\"),\n",
    "]\n",
    "\n",
    "for layer in layers:\n",
    "    nn.add_layer(layer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0, Cost: 3.191427373424325, Accuracy: 25.69%\n",
      "Epoch 5, Cost: 1.9767107844276965, Accuracy: 61.61%\n",
      "Epoch 10, Cost: 0.9342281811949328, Accuracy: 84.91%\n",
      "Epoch 15, Cost: 0.6863550159211198, Accuracy: 88.71%\n",
      "Epoch 20, Cost: 0.5941095627791009, Accuracy: 90.27%\n",
      "Epoch 25, Cost: 0.5348368748518596, Accuracy: 91.38%\n",
      "Epoch 30, Cost: 0.49596785660116854, Accuracy: 92.10%\n"
     ]
    }
   ],
   "source": [
    "# Treinamento da rede neural\n",
    "nn.train(\n",
    "    X_train,\n",
    "    Y_train,\n",
    "    epochs=35,\n",
    "    learning_rate=0.31,\n",
    "    cost_function=\"cross_entropy\",\n",
    "    batch_size=64,\n",
    "    verbose_freq=5,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 2000x200 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Fazer previsões\n",
    "predictions = nn.predict(X_train)\n",
    "\n",
    "# Visualizar algumas imagens e suas previsões\n",
    "def plot_images(X, y_true, y_pred, num_images=10):\n",
    "    fig, axes = plt.subplots(1, num_images, figsize=(20, 2))\n",
    "    for i in range(num_images):\n",
    "        ax = axes[i] # type: ignore\n",
    "        ax.imshow(X[:, i].reshape(28, 28), cmap='gray')\n",
    "        ax.set_title(f'True: {np.argmax(y_true[:, i])}\\nPred: {y_pred[i]}')\n",
    "        ax.axis('off')\n",
    "    plt.show()\n",
    "\n",
    "# Selecionar algumas imagens para exibir\n",
    "num_images = 10\n",
    "indices = np.random.choice(X_train.shape[1], num_images, replace=False)\n",
    "X_sample = X_train[:, indices]\n",
    "Y_sample = Y_train[:, indices]\n",
    "pred_sample = predictions[indices]\n",
    "\n",
    "# Mostrar as imagens e as previsões\n",
    "plot_images(X_sample, Y_sample, pred_sample, num_images=num_images)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "neural-network-H6dbbr34-py3.12",
   "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": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}