51. cifar10_cnn.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"51_cifar10_cnn.ipynb","version":"0.3.2","views":{},"default_view":{},"provenance":[],"collapsed_sections":[]},"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"accelerator":"GPU"},"cells":[{"metadata":{"id":"6MQom9kBVYOD","colab_type":"text"},"cell_type":"markdown","source":["# 51. CIFAR-10 CNN\n","\n","<p style=\"text-align: right;\">\n","blackdew7@gmail.com<br>\n","Your name :\n","</p>\n","\n","#### 선행지식\n","1. TensorFlow 다루기 기초\n","2. 모델링을 한다는 것에 대한 이해.\n","3. Supervised Learning 중 Classification에 대한 기본 지식.\n","4. CNN에 대한 기본 구조와 개념\n","\n","#### 실습목표\n","1. CNN의 구조를 Graph로 그려낼 수 있다.\n","2. 그려낸 Graph를 텐서플로우를 이용해 코딩할 수 있다.\n","\n","#### 사용데이터.\n","\n","01. The CIFAR-10 dataset : https://www.cs.toronto.edu/~kriz/cifar.html\n","\n","\n","![image](https://github.com/blackdew/dplus-tensorflow/blob/master/img/06_network_flowchart.png?raw=true)\n"]},{"metadata":{"id":"JOyTNcJuVYOG","colab_type":"text"},"cell_type":"markdown","source":["## 00. 라이브러리 불러오기"]},{"metadata":{"id":"bxWrtPtOM3Qi","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["import tensorflow as tf\n","tf.test.gpu_device_name()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"LXTdshEdVYOL","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["import tensorflow as tf\n","import numpy as np\n","import random"],"execution_count":0,"outputs":[]},{"metadata":{"id":"r7uEvxD0VYOP","colab_type":"text"},"cell_type":"markdown","source":["## 01. 데이터 불러오기 & 전처리"]},{"metadata":{"id":"kE0oWkXcVYOU","colab_type":"text"},"cell_type":"markdown","source":["### train, test 데이터 셋 분리 "]},{"metadata":{"id":"ucfVrG3EVYOV","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# cifar10\n","\n","(train_x, train_y), (test_x, test_y) = tf.keras.datasets.cifar10.load_data()\n","class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']\n","\n","train_y = train_y.reshape(-1,)\n","test_y = test_y.reshape(-1,)\n","\n","'''\n","트레이닝 데이터로 활용할 50000개 이미지, 테스트용 10000개 이미지\n","32 * 32 * 3 사이즈의 이미지가 어레이에 담겨있다.\n","레이블은 one-hot encoding이 되어 있지 않다.\n","'''\n","print(train_x.shape)\n","print(train_y.shape)\n","print(test_x.shape)\n","print(test_y.shape)\n","\n","print(train_x[0].shape)\n","print(train_y[0])"],"execution_count":0,"outputs":[]},{"metadata":{"id":"BV2QswnpVYOX","colab_type":"text"},"cell_type":"markdown","source":["### 데이터가 어떻게 생겼는지 확인해보자."]},{"metadata":{"id":"jqZWWyjCVYOZ","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["%matplotlib inline\n","%config InlineBackend.figure_format = 'retina'\n","\n","import matplotlib.pyplot as plt\n","\n","class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer', \n","               'dog', 'frog', 'horse', 'ship', 'truck']\n","\n","# 이미지 출력을 위한 함수\n","def cifar10_plot(i):\n","    pixels = test_x[i]\n","    a = int(test_y[i])\n","    plt.title('Label: {} ({})'.format(class_names[a], a))\n","    plt.imshow(pixels, cmap='gray')\n","    return plt.show()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"tkoj2UqUVYOf","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["cifar10_plot(np.random.randint(0, high=9999))"],"execution_count":0,"outputs":[]},{"metadata":{"id":"IHBRilAcVYOm","colab_type":"text"},"cell_type":"markdown","source":["### CIFAR-10 dataset\n","* 총 60,000장\n","* 가로, 세로 길이 = 32, RGB\n","* 데이터는 32 * 32 * 3 의 배열로 담겨있음"]},{"metadata":{"id":"4uenK7v-VYOo","colab_type":"text"},"cell_type":"markdown","source":["## 02. 실습 - Prepare Variables for a CNN Graph with TF "]},{"metadata":{"id":"a8YW5kaaVYOq","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","Prepare Variables 코드를 완성해보자.\n","\n","여러분이 직접 스크립트를 짜야 합니다.\n","복사하여 붙여넣기를 하지 마시고, 직접 타이핑 하는 것을 권장합니다.\n","'''\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"FxZJ_rt_VYOt","colab_type":"text"},"cell_type":"markdown","source":["## 03. 실습 - Make a CNN Graph with TF"]},{"metadata":{"id":"AZ-j6PfSVYOu","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","CNN Graph 코드를 완성해보자. \n","\n","여러분이 직접 스크립트를 짜야 합니다.\n","복사하여 붙여넣기를 하지 마시고, 직접 타이핑 하는 것을 권장합니다.\n","'''\n","# First Hidden Layer 코드를 완성해보자\n","\n","\n","\n","# Second Hidden Layer 코드를 완성해보자\n","\n","\n","\n","# Fully Connected Layer 코드를 완성해보자\n","\n","\n","\n","# Cost & Optimizer 코드를 완성해보자\n","\n","\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"43EzOTFKVYOy","colab_type":"text"},"cell_type":"markdown","source":["## 04. 실습 - Traning & Evaluation"]},{"metadata":{"id":"wVUbbJehVYO0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","Traning & Evaluation 코드를 완성해보자\n","'''\n","\n","# Initialize\n","\n","\n","# Training\n","\n","print('Learning started. It takes sometime.')\n","\n","\n","print(\"Training Complete\")\n","\n","# Evaluation\n","# Test model and check accuracy\n","\n","\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"1drbRvfIVYO4","colab_type":"text"},"cell_type":"markdown","source":["## 05. 실습 해답"]},{"metadata":{"id":"Boh1N7VoVYO7","colab_type":"text"},"cell_type":"markdown","source":["### 05.1 Import Library & Prepare CIFAR-10 Data"]},{"metadata":{"id":"BKhshq7NVYO8","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["import tensorflow as tf\n","import numpy as np\n","import random\n","\n","# cifar10\n","class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer', \n","               'dog', 'frog', 'horse', 'ship', 'truck']\n","\n","(train_x, train_y), (test_x, test_y) = tf.keras.datasets.cifar10.load_data()\n","\n","train_y = train_y.reshape(-1,)\n","test_y = test_y.reshape(-1,)\n","\n","'''\n","트레이닝 데이터로 활용할 50000개 이미지, 테스트용 10000개 이미지\n","32 * 32 * 3 사이즈의 이미지가 어레이에 담겨있다.\n","레이블은 one-hot encoding이 되어 있지 않다.\n","'''\n","\n","print(train_x.shape)\n","print(train_y.shape)\n","print(test_x.shape)\n","print(test_y.shape)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Dkkt9zvKVYPB","colab_type":"text"},"cell_type":"markdown","source":["### 05.2 Prepare Variables for a CNN Graph with TF"]},{"metadata":{"id":"Cv98D68iVYPC","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# Graph Clear\n","tf.reset_default_graph()\n","\n","############################\n","# Place Holders\n","\n","X = tf.placeholder(tf.float32, [None, 32, 32, 3])\n","Y = tf.placeholder(tf.float32, [None, 10])\n","\n","lr = tf.placeholder(tf.float32)\n","kp = tf.placeholder(tf.float32)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"d52opL74VYPF","colab_type":"text"},"cell_type":"markdown","source":["### 05.3 Make a CNN Graph & Run Session with TF"]},{"metadata":{"id":"c-xUyARiVYPG","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["#########################\n","# ConvLayer 01 \n","\n","# for Conv Layer 01 filter\n","w1 = tf.get_variable(shape=[5, 5, 3, 16], initializer=tf.contrib.layers.xavier_initializer(), name=\"w1\")\n","b1 = tf.Variable(tf.random_normal([16], stddev=0.01))\n","\n","# Convolution Layer -> (?, 32, 32, 16)\n","conv1 = tf.add(tf.nn.conv2d(X, w1, strides=[1, 1, 1, 1], padding='SAME'), b1)\n","conv1 = tf.nn.relu(conv1)\n","\n","# Pooling Layer -> (?, 16, 16, 16)\n","pool1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME')\n","\n","#########################\n","# ConvLayer 02\n","\n","# for Conv Layer 02 filter\n","w2 = tf.get_variable(shape=[3, 3, 16, 16], initializer=tf.contrib.layers.xavier_initializer(), name=\"w2\")\n","b2 = tf.Variable(tf.random_normal([16], stddev=0.01))\n","\n","# Convolution Layer -> (?, 16, 16, 16)\n","conv2 = tf.add(tf.nn.conv2d(pool1, w2, strides=[1, 1, 1, 1], padding='SAME'), b2)\n","conv2 = tf.nn.relu(conv2)\n","\n","# Pooling Layer -> (?, 8, 8, 16)\n","pool2 = tf.nn.max_pool(conv2, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME')\n","    \n","##################################################\n","# Flatten layer & Fully Connected Layer\n","\n","###############################\n","# Flatten layer\n","flat = tf.reshape(pool2, [-1, 8 * 8 * 16])\n","\n","###############################\n","# Fully Connected Layer 01\n","\n","# for Final FC 8x8x64 inputs -> 256 outputs\n","wfc1 = tf.get_variable(shape=[8 * 8 * 16, 128], initializer=tf.contrib.layers.xavier_initializer(), name=\"wfc1\")\n","bfc1 = tf.Variable(tf.random_normal([128], stddev=0.01))\n","\n","# fc layer\n","fc1 = tf.add(tf.matmul(flat, wfc1), bfc1)\n","fc1 = tf.nn.relu(fc1)\n","\n","# dropout\n","fc1 = tf.nn.dropout(fc1, keep_prob=kp)\n","\n","###############################\n","# Fully Connected Layer 02\n","\n","# for Final FC 256 inputs -> 10 outputs\n","wfo = tf.get_variable(shape=[128, 10], initializer=tf.contrib.layers.xavier_initializer(), name=\"wfo\")\n","bfo = tf.Variable(tf.random_normal([10], stddev=0.01))\n","\n","# fc layer\n","logits = tf.add(tf.matmul(fc1, wfo), bfo)\n","\n","#########################\n","# Cost & Optimizer\n","\n","# Cost(loss) function & Optimizer\n","cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits, labels=Y))\n","optimizer = tf.train.AdamOptimizer(learning_rate=lr).minimize(cost)\n","    \n","# Test model and check accuracy\n","correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(Y, 1))\n","acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"udWQWhLCVYPJ","colab_type":"text"},"cell_type":"markdown","source":["### 05.4 Training & Evaluation"]},{"metadata":{"id":"l-ThLhQVVYPJ","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["#########################\n","# Session initialize\n","\n","sess = tf.Session()\n","sess.run(tf.global_variables_initializer())\n","\n","###########################\n","# Traning\n","\n","print('Learning started. It takes sometime.')\n","\n","epochs = 20\n","batch_size = 256\n","n_of_batches = int(len(train_x) / batch_size)\n","\n","for epoch in range(epochs): \n","    print(\"%dth epoch\" % (epoch + 1))\n","\n","    for i in range(n_of_batches):\n","        X_batch = train_x[(i * batch_size):((i + 1) * batch_size)]\n","        Y_batch = train_y[(i * batch_size):((i + 1) * batch_size)]\n","\n","        sess.run(optimizer, feed_dict={X: X_batch, Y: np.eye(10)[Y_batch], lr: 0.05, kp: 0.7})\n","\n","        # 학습 상황 디스플레이\n","        if ((i + 1) % 20 == 0): \n","            loss, accuracy = sess.run([cost, acc], feed_dict={X: X_batch, Y: np.eye(10)[Y_batch], lr: 0.05, kp: 1.0})\n","            print(\"%dth records, training cost: %.3f, accuracy: %.2f\" % ((i + 1) * batch_size, loss, accuracy * 100))\n","\n","print(\"Training Complete\")"],"execution_count":0,"outputs":[]},{"metadata":{"id":"6diN7CS-VYPN","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["###########################\n","# Evaluation\n","accuracy = sess.run(acc, feed_dict={X: test_x, Y: np.eye(10)[test_y], lr: 0.01, , kp: 1.0})\n","print('Accuracy: %.2f' % (accuracy * 100))"],"execution_count":0,"outputs":[]},{"metadata":{"id":"DZzf397RVYPQ","colab_type":"text"},"cell_type":"markdown","source":["### 성능을 한 번 확인해 보자. 제대로 되고 있는걸까?"]},{"metadata":{"id":"-T5TgAjeVYPS","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["import random\n","\n","# Get one and predict\n","r = random.randint(0, len(test_x) - 1)\n","print(test_y[r:r + 1])\n","print(\"Label: \", class_names[int(test_y[r:r + 1])])\n","print(\"Prediction: \", class_names[int(sess.run(tf.argmax(logits, 1), feed_dict={X: test_x[r:r + 1]}))])\n","\n","import matplotlib.pyplot as plt\n","plt.imshow(test_x[r:r + 1].reshape(32, 32, 3), cmap='Greys', interpolation='nearest')\n","plt.show()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"d6NMZX52VYPX","colab_type":"text"},"cell_type":"markdown","source":["## 06. 대체 뭘 학습한 것인가? "]},{"metadata":{"id":"wXvCXaH6VYPZ","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["%matplotlib inline\n","import matplotlib.pyplot as plt\n","import math\n","import numpy as np\n","\n","def plot_conv_weights(weights, input_channel=0):\n","    \n","    with tf.Session() as sess:\n","        sess.run(tf.global_variables_initializer())\n","\n","        w = sess.run(weights)\n","\n","        w_min = np.min(w)\n","        w_max = np.max(w)\n","\n","        num_filters = w.shape[3]\n","        num_grids = math.ceil(math.sqrt(num_filters))\n","        fig, axes = plt.subplots(num_grids, num_grids)\n","\n","        for i, ax in enumerate(axes.flat):\n","            if i < num_filters:\n","                img = w[:, :, input_channel, i]\n","                ax.imshow(img, vmin=w_min, vmax=w_max, interpolation='nearest', cmap='seismic')\n","\n","            ax.set_xticks([])\n","            ax.set_yticks([])\n","\n","        plt.show()\n","        \n","def plot_conv_layer(layer, image):\n","\n","    with tf.Session() as sess:\n","\n","        sess.run(tf.global_variables_initializer())\n","\n","        feed_dict = {X: [image]}\n","        values = sess.run(layer, feed_dict=feed_dict)\n","\n","        num_filters = values.shape[3]\n","        num_grids = math.ceil(math.sqrt(num_filters))\n","        fig, axes = plt.subplots(num_grids, num_grids)\n","\n","        for i, ax in enumerate(axes.flat):\n","            if i < num_filters:\n","                img = values[0, :, :, i]\n","                ax.imshow(img, interpolation='nearest', cmap='binary')\n","                \n","            ax.set_xticks([])\n","            ax.set_yticks([])\n","\n","        plt.show()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"zyYCJfW_VYPc","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["plot_conv_weights(weights=w1)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"-TxbyGKTVYPf","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["plot_conv_weights(weights=w2, input_channel=0)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"URE4-Fy1VYPi","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["plot_conv_weights(weights=w2, input_channel=1)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"h_yLi-5OVYPl","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["img_shape = (32, 32, 3)\n","\n","def plot_image(image):\n","    plt.imshow(image.reshape(img_shape),\n","               interpolation='nearest',\n","               cmap='binary')\n","\n","    plt.show()\n","    \n","image1 = test_x[0]\n","plot_image(image1)\n","\n","image2 = test_x[1]\n","plot_image(image2)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"HBOmyO_2VYPn","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["plot_conv_layer(layer=pool1, image=image2)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"SVv_MZguVYPq","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["plot_conv_layer(layer=pool2, image=image2)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"cK7beT-VVYPs","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["session.close()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Xu9hDTjVVYPv","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":[""],"execution_count":0,"outputs":[]}]}