{ "cells": [ { "cell_type": "markdown", "metadata": { "id": "view-in-github", "colab_type": "text" }, "source": [ "\"Open" ] }, { "cell_type": "markdown", "metadata": { "id": "XUZA3OvtRzK-" }, "source": [ "\n", "

\n", " \n", "

Cifar 10 資料庫
\n", "
圖片來源: https://www.kaggle.com/
\n", "

" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "tS5xFeQwe9Xu" }, "outputs": [], "source": [ "import matplotlib.pyplot as plt\n", "import numpy as np\n", "import pandas as pd" ] }, { "cell_type": "code", "source": [ "from tensorflow.keras.models import Sequential\n", "from tensorflow.keras.layers import Dense, Flatten\n", "from tensorflow.keras.layers import Conv2D, MaxPool2D\n", "from tensorflow.keras.optimizers import SGD, Adam\n", "from tensorflow.keras.datasets import cifar10\n", "from tensorflow.keras.utils import to_categorical\n", "from tensorflow.keras.preprocessing.image import load_img" ], "metadata": { "id": "kEEDoYWJqT8w" }, "execution_count": null, "outputs": [] }, { "cell_type": "markdown", "metadata": { "id": "e8zYubaPfy-S" }, "source": [ "### 1. 讀入CIFAR--10資料集並進行資料預處理" ] }, { "cell_type": "markdown", "metadata": { "id": "YleM4DH4NoOA" }, "source": [ "CIFAR 10 是包含 10 種類的彩色小圖資料集，每張圖的尺寸為 $32\\times32$\n", "\n", "10 個類別分別是：飛機、交通工具、鳥、貓、鹿、狗、青蛙、馬、船、卡車" ] }, { "cell_type": "markdown", "metadata": { "id": "AOAGjiPogF0w" }, "source": [ "讀取 CIFAR 10 資料集" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "4c5EYPsuNThg" }, "outputs": [], "source": [ "class_name = ['飛機', '汽車', '鳥', '貓', '鹿', '狗', '青蛙', '馬', '船', '卡車']" ] }, { "cell_type": "code", "source": [ "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n", "x_train = x_train/255\n", "x_test = x_test/255\n", "y_train = to_categorical(y_train, 10)\n", "y_test = to_categorical(y_test, 10)" ], "metadata": { "id": "BykCOsFexlgE" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "plt.imshow(x_test[5])\n", "print(f\"這是{class_name[y_test[5].argmax()]}\");" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 283 }, "id": "pg7EMYqItzQA", "outputId": "fe10660a-b949-4e96-dbae-43f972871aaf" }, "execution_count": null, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "這是青蛙\n" ] }, { "output_type": "display_data", "data": { "text/plain": [ "
" ], "image/png": "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\n" }, "metadata": { "needs_background": "light" } } ] }, { "cell_type": "code", "source": [ "y_test[5].flatten()" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "_Es9mEkSuqLL", "outputId": "bc250e9a-7f6d-4993-db94-c646af04ee9c" }, "execution_count": null, "outputs": [ { "output_type": "execute_result", "data": { "text/plain": [ "(10000, 1)" ] }, "metadata": {}, "execution_count": 15 } ] }, { "cell_type": "code", "source": [ "x_train = x_train/255\n", "x_test = x_test/255\n", "y_train = to_categorical(y_train, 10)\n", "y_test = to_categorical(y_test, 10)" ], "metadata": { "id": "NJiJS-j7uK_j" }, "execution_count": null, "outputs": [] }, { "cell_type": "markdown", "metadata": { "id": "swp6zTOST8gD" }, "source": [ "### 2. 第一代：過去標準逐層打造神經網路的方法" ] }, { "cell_type": "markdown", "source": [ "在這個部分，我們將逐步帶領大家建立經典的 CNN 模型 LeNet-5 的變形。\n", "\n", "LeNet-5 分成兩個部分，分別為卷積層與全連接層，兩部份之間是透過扁平層 (Flatten) ，將卷積層最後輸出的 2 維向量壓扁成 1 維向量。" ], "metadata": { "id": "JBrNY0f-2oGg" } }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "jGEAVln9TRwa" }, "outputs": [], "source": [ "model = Sequential()\n", "model.add(Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same', activation='relu')) \n", "model.add(MaxPool2D())\n", "model.add(Conv2D(128, (3, 3), padding='same', activation='relu')) \n", "model.add(MaxPool2D())\n", "model.add(Conv2D(512, (3, 3), padding='same', activation='relu')) \n", "model.add(MaxPool2D())\n", "model.add(Flatten())\n", "model.add(Dense(256, activation='relu'))\n", "model.add(Dense(10, activation='softmax'))" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "floU0of05vJj", "outputId": "f08c8a60-c6ce-4ff5-92e7-c8c2f0bc936f" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Model: \"sequential\"\n", "___________________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "===========================================================================\n", " conv2d (Conv2D) (None, 32, 32, 32) 896 \n", " \n", " max_pooling2d (MaxPooling2D) (None, 16, 16, 32) 0 \n", " \n", " conv2d_1 (Conv2D) (None, 16, 16, 128) 36992 \n", " \n", " max_pooling2d_1 (MaxPooling2D) (None, 8, 8, 128) 0 \n", " \n", " conv2d_2 (Conv2D) (None, 8, 8, 512) 590336 \n", " \n", " max_pooling2d_2 (MaxPooling2D) (None, 4, 4, 512) 0 \n", " \n", " flatten (Flatten) (None, 8192) 0 \n", " \n", " dense (Dense) (None, 256) 2097408 \n", " \n", " dense_1 (Dense) (None, 10) 2570 \n", " \n", "===========================================================================\n", "Total params: 2,728,202\n", "Trainable params: 2,728,202\n", "Non-trainable params: 0\n", "___________________________________________________________________________\n" ] } ], "source": [ "model.summary()" ] }, { "cell_type": "markdown", "metadata": { "id": "GWdFSj9OcYWl" }, "source": [ "#### 編譯模型: 設定模型訓練時的設定\n", "\n", "- Optimizer: Stochastic Gradient Descent (SGD)\n", "- Loss: categorical cross-entropy " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "86fvl2DHcKwj" }, "outputs": [], "source": [ "model.compile(loss='categorical_crossentropy', \n", " optimizer=Adam(),\n", " metrics=['accuracy'])" ] }, { "cell_type": "markdown", "metadata": { "id": "-Zz5yHOzdUgv" }, "source": [ "#### 訓練模型: 透過訓練來學習分類資料的函數" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "LVdaaVOJcKrk", "colab": { "base_uri": "https://localhost:8080/" }, "outputId": "590dad31-ff24-4259-c010-c8a6a8bf48ab" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Epoch 1/3\n", "391/391 [==============================] - 26s 38ms/step - loss: 1.4611 - accuracy: 0.4717 - val_loss: 1.1093 - val_accuracy: 0.6097\n", "Epoch 2/3\n", "391/391 [==============================] - 14s 36ms/step - loss: 0.9935 - accuracy: 0.6487 - val_loss: 1.0079 - val_accuracy: 0.6478\n", "Epoch 3/3\n", "391/391 [==============================] - 14s 36ms/step - loss: 0.8006 - accuracy: 0.7194 - val_loss: 0.7835 - val_accuracy: 0.7292\n" ] }, { "output_type": "execute_result", "data": { "text/plain": [ "" ] }, "metadata": {}, "execution_count": 34 } ], "source": [ "model.fit(x_train, y_train, \n", " batch_size=128, \n", " epochs=3,\n", " validation_data=(x_test, y_test)\n", " )" ] }, { "cell_type": "markdown", "metadata": { "id": "5IJhvV_BSo9x" }, "source": [ "#### 模型預測: 預測資料集的準確率" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "btoBDRueSokB", "colab": { "base_uri": "https://localhost:8080/" }, "outputId": "37c4cb03-545c-4c02-f5b8-2242d9c94608" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "1563/1563 [==============================] - 11s 7ms/step - loss: 0.6452 - accuracy: 0.7798\n", "313/313 [==============================] - 2s 7ms/step - loss: 0.7835 - accuracy: 0.7292\n" ] } ], "source": [ "score_train = model.evaluate(x_train, y_train)\n", "score_test = model.evaluate(x_test, y_test)" ] }, { "cell_type": "code", "source": [ "print(f'Train Accuracy: {score_train[1]*100}')\n", "print(f'Test Accuracy: {score_test[1]*100}')" ], "metadata": { "id": "_aZb0x5t2zcZ", "colab": { "base_uri": "https://localhost:8080/" }, "outputId": "df9cc553-b434-41c0-e646-f0e225649387" }, "execution_count": null, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Train Accuracy: 77.97799706459045\n", "Test Accuracy: 72.9200005531311\n" ] } ] }, { "cell_type": "markdown", "metadata": { "id": "0G7lUOLHQEjr" }, "source": [ "## 3. 第二代：用串列將神經網路層一次放進模型中" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "14YYzYYwcKpE" }, "outputs": [], "source": [ "many_layers = [Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same', activation='relu'), \n", " MaxPool2D(), \n", " Conv2D(128, (3, 3), padding='same', activation='relu'), \n", " MaxPool2D(), \n", " Conv2D(512, (3, 3), padding='same', activation='relu'), \n", " MaxPool2D(),\n", " Flatten(), \n", " Dense(256, activation='relu'), \n", " Dense(10, activation='softmax'), \n", " ]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "xn7EzCCgqkDs" }, "outputs": [], "source": [ "second_model = Sequential(many_layers)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "hvmVUw4krEJq", "outputId": "ab95abc1-f908-4f3b-e3bf-f612d77519cb" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Model: \"sequential_1\"\n", "___________________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "===========================================================================\n", " conv2d_3 (Conv2D) (None, 32, 32, 32) 896 \n", " \n", " max_pooling2d_3 (MaxPooling2D) (None, 16, 16, 32) 0 \n", " \n", " conv2d_4 (Conv2D) (None, 16, 16, 128) 36992 \n", " \n", " max_pooling2d_4 (MaxPooling2D) (None, 8, 8, 128) 0 \n", " \n", " conv2d_5 (Conv2D) (None, 8, 8, 512) 590336 \n", " \n", " max_pooling2d_5 (MaxPooling2D) (None, 4, 4, 512) 0 \n", " \n", " flatten_1 (Flatten) (None, 8192) 0 \n", " \n", " dense_2 (Dense) (None, 256) 2097408 \n", " \n", " dense_3 (Dense) (None, 10) 2570 \n", " \n", "===========================================================================\n", "Total params: 2,728,202\n", "Trainable params: 2,728,202\n", "Non-trainable params: 0\n", "___________________________________________________________________________\n" ] } ], "source": [ "second_model.summary()" ] }, { "cell_type": "markdown", "source": [ "## 4. 第三代：用兩個串列定義神經網路，再一次放進模型中" ], "metadata": { "id": "8831zmPI27gx" } }, { "cell_type": "code", "source": [ "CNN_layers = [Conv2D(32, (3, 3), input_shape=(32, 32, 3), padding='same',activation='relu'), \n", " MaxPool2D(), \n", " Conv2D(128, (3, 3), padding='same', activation='relu'), \n", " MaxPool2D(), \n", " Conv2D(512, (3, 3), padding='same', activation='relu'), \n", " MaxPool2D(), \n", " Flatten(), \n", " ]" ], "metadata": { "id": "BIF4WqQH27pw" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "FC_layers = [Dense(units=256, activation='relu'), \n", " Dense(units=10, activation='softmax'), \n", " ]" ], "metadata": { "id": "qMSAZe3U3IQk" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "third_model = Sequential(CNN_layers+FC_layers)" ], "metadata": { "id": "EOAbbhUR3TDE" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "third_model.summary()" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "nO1DDeNv3TFB", "outputId": "598a52f5-3e52-4772-cae2-b7d484be467d" }, "execution_count": null, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Model: \"sequential_2\"\n", "___________________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "===========================================================================\n", " conv2d_6 (Conv2D) (None, 32, 32, 32) 896 \n", " \n", " max_pooling2d_6 (MaxPooling2D) (None, 16, 16, 32) 0 \n", " \n", " conv2d_7 (Conv2D) (None, 16, 16, 128) 36992 \n", " \n", " max_pooling2d_7 (MaxPooling2D) (None, 8, 8, 128) 0 \n", " \n", " conv2d_8 (Conv2D) (None, 8, 8, 512) 590336 \n", " \n", " max_pooling2d_8 (MaxPooling2D) (None, 4, 4, 512) 0 \n", " \n", " flatten_2 (Flatten) (None, 8192) 0 \n", " \n", " dense_4 (Dense) (None, 256) 2097408 \n", " \n", " dense_5 (Dense) (None, 10) 2570 \n", " \n", "===========================================================================\n", "Total params: 2,728,202\n", "Trainable params: 2,728,202\n", "Non-trainable params: 0\n", "___________________________________________________________________________\n" ] } ] }, { "cell_type": "markdown", "source": [ "### 5. 第四代：可以方便修改部份神經網路，不會動到別人！" ], "metadata": { "id": "BiBo4aPm3h0A" } }, { "cell_type": "code", "source": [ "new_FC_layers = [Dense(1024, activation='relu'), \n", " Dense(128, activation='relu'), \n", " Dense(10, activation='softmax'), \n", " ]\n" ], "metadata": { "id": "eXA_dDWr3THH" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "fourth_model = Sequential(CNN_layers+new_FC_layers)" ], "metadata": { "id": "nsN6AAC43TJN" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "fourth_model.summary()" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "_Pi2Xomo3TLP", "outputId": "f3291410-8a95-462e-88bf-75f28a7ff845" }, "execution_count": null, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Model: \"sequential_3\"\n", "___________________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "===========================================================================\n", " conv2d_6 (Conv2D) (None, 32, 32, 32) 896 \n", " \n", " max_pooling2d_6 (MaxPooling2D) (None, 16, 16, 32) 0 \n", " \n", " conv2d_7 (Conv2D) (None, 16, 16, 128) 36992 \n", " \n", " max_pooling2d_7 (MaxPooling2D) (None, 8, 8, 128) 0 \n", " \n", " conv2d_8 (Conv2D) (None, 8, 8, 512) 590336 \n", " \n", " max_pooling2d_8 (MaxPooling2D) (None, 4, 4, 512) 0 \n", " \n", " flatten_2 (Flatten) (None, 8192) 0 \n", " \n", " dense_6 (Dense) (None, 1024) 8389632 \n", " \n", " dense_7 (Dense) (None, 128) 131200 \n", " \n", " dense_8 (Dense) (None, 10) 1290 \n", " \n", "===========================================================================\n", "Total params: 9,150,346\n", "Trainable params: 9,150,346\n", "Non-trainable params: 0\n", "___________________________________________________________________________\n" ] } ] }, { "cell_type": "code", "source": [ "" ], "metadata": { "id": "Me5NQzpaxbOH" }, "execution_count": null, "outputs": [] } ], "metadata": { "accelerator": "GPU", "colab": { "collapsed_sections": [], "machine_shape": "hm", "name": "冒險32_神經網路的另一個打造方式.ipynb", "provenance": [], "include_colab_link": true }, "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": "3.7.3" } }, "nbformat": 4, "nbformat_minor": 0 }