Tensorflow 之模型内容可视化
1. tensorflow实现
![](https://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif)
# 卷积网络的训练数据为MNIST(28*28灰度单色图像) import tensorflow as tf import numpy as np import matplotlib.pyplot as plt from tensorflow.examples.tutorials.mnist import input_data train_epochs = 100 # 训练轮数 batch_size = 100 # 随机出去数据大小 display_step = 1 # 显示训练结果的间隔 learning_rate= 0.0001 # 学习效率 drop_prob = 0.5 # 正则化,丢弃比例 fch_nodes = 512 # 全连接隐藏层神经元的个数 # 网络模型需要的一些辅助函数 # 权重初始化(卷积核初始化) # tf.truncated_normal()不同于tf.random_normal(),返回的值中不会偏离均值两倍的标准差 # 参数shpae为一个列表对象,例如[5, 5, 1, 32]对应 # 5,5 表示卷积核的大小, 1代表通道channel,对彩色图片做卷积是3,单色灰度为1 # 最后一个数字32,卷积核的个数,(也就是卷基层提取的特征数量) # 显式声明数据类型,切记 def weight_init(shape): weights = tf.truncated_normal(shape, stddev=0.1,dtype=tf.float32) return tf.Variable(weights) # 偏置的初始化 def biases_init(shape): biases = tf.random_normal(shape,dtype=tf.float32) return tf.Variable(biases) # 随机选取mini_batch def get_random_batchdata(n_samples, batchsize): start_index = np.random.randint(0, n_samples - batchsize) return (start_index, start_index + batchsize) # 全连接层权重初始化函数xavier def xavier_init(layer1, layer2, constant = 1): Min = -constant * np.sqrt(6.0 / (layer1 + layer2)) Max = constant * np.sqrt(6.0 / (layer1 + layer2)) return tf.Variable(tf.random_uniform((layer1, layer2), minval = Min, maxval = Max, dtype = tf.float32)) # 卷积 def conv2d(x, w): return tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME') # 源码的位置在tensorflow/python/ops下nn_impl.py和nn_ops.py # 这个函数接收两个参数,x 是图像的像素, w 是卷积核 # x 张量的维度[batch, height, width, channels] # w 卷积核的维度[height, width, channels, channels_multiplier] # tf.nn.conv2d()是一个二维卷积函数, # stirdes 是卷积核移动的步长,4个1表示,在x张量维度的四个参数上移动步长 # padding 参数'SAME',表示对原始输入像素进行填充,卷积后映射的2D图像与原图大小相等 # 填充,是指在原图像素值矩阵周围填充0像素点 # 如果不进行填充,假设 原图为 32x32 的图像,卷积和大小为 5x5 ,卷积后映射图像大小 为 28x28 # 池化 def max_pool_2x2(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') # 池化跟卷积的情况有点类似 # x 是卷积后,有经过非线性激活后的图像, # ksize 是池化滑动张量 # ksize 的维度[batch, height, width, channels],跟 x 张量相同 # strides [1, 2, 2, 1],与上面对应维度的移动步长 # padding与卷积函数相同,padding='VALID',对原图像不进行0填充 # x 是手写图像的像素值,y 是图像对应的标签 x = tf.placeholder(tf.float32, [None, 784]) y = tf.placeholder(tf.float32, [None, 10]) # 把灰度图像一维向量,转换为28x28二维结构 x_image = tf.reshape(x, [-1, 28, 28, 1]) # -1表示任意数量的样本数,大小为28x28深度为一的张量 # 可以忽略(其实是用深度为28的,28x1的张量,来表示28x28深度为1的张量) w_conv1 = weight_init([5, 5, 1, 16]) # 5x5,深度为1,16个 b_conv1 = biases_init([16]) h_conv1 = tf.nn.relu(conv2d(x_image, w_conv1) + b_conv1) # 输出张量的尺寸:28x28x16 h_pool1 = max_pool_2x2(h_conv1) # 池化后张量尺寸:14x14x16 # h_pool1 , 14x14的16个特征图 w_conv2 = weight_init([5, 5, 16, 32]) # 5x5,深度为16,32个 b_conv2 = biases_init([32]) h_conv2 = tf.nn.relu(conv2d(h_pool1, w_conv2) + b_conv2) # 输出张量的尺寸:14x14x32 h_pool2 = max_pool_2x2(h_conv2) # 池化后张量尺寸:7x7x32 # h_pool2 , 7x7的32个特征图 # h_pool2是一个7x7x32的tensor,将其转换为一个一维的向量 h_fpool2 = tf.reshape(h_pool2, [-1, 7*7*32]) # 全连接层,隐藏层节点为512个 # 权重初始化 w_fc1 = xavier_init(7*7*32, fch_nodes) b_fc1 = biases_init([fch_nodes]) h_fc1 = tf.nn.relu(tf.matmul(h_fpool2, w_fc1) + b_fc1) # 全连接隐藏层/输出层 # 为了防止网络出现过拟合的情况,对全连接隐藏层进行 Dropout(正则化)处理,在训练过程中随机的丢弃部分 # 节点的数据来防止过拟合.Dropout同把节点数据设置为0来丢弃一些特征值,仅在训练过程中, # 预测的时候,仍使用全数据特征 # 传入丢弃节点数据的比例 #keep_prob = tf.placeholder(tf.float32) h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob=drop_prob) # 隐藏层与输出层权重初始化 w_fc2 = xavier_init(fch_nodes, 10) b_fc2 = biases_init([10]) # 未激活的输出 y_ = tf.add(tf.matmul(h_fc1_drop, w_fc2), b_fc2) # 激活后的输出 y_out = tf.nn.softmax(y_) # 交叉熵代价函数 cross_entropy = tf.reduce_mean(-tf.reduce_sum(y * tf.log(y_out), reduction_indices = [1])) # tensorflow自带一个计算交叉熵的方法 # 输入没有进行非线性激活的输出值 和 对应真实标签 #cross_loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(y_, y)) # 优化器选择Adam(有多个选择) optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cross_entropy) # 准确率 # 每个样本的预测结果是一个(1,10)的vector correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_out, 1)) # tf.cast把bool值转换为浮点数 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) # 全局变量进行初始化的Operation init = tf.global_variables_initializer() # 加载数据集MNIST mnist = input_data.read_data_sets('MNIST/mnist', one_hot=True) n_samples = int(mnist.train.num_examples) total_batches = int(n_samples / batch_size) # 会话 with tf.Session() as sess: sess.run(init) Cost = [] Accuracy = [] for i in range(train_epochs): for j in range(100): start_index, end_index = get_random_batchdata(n_samples, batch_size) batch_x = mnist.train.images[start_index: end_index] batch_y = mnist.train.labels[start_index: end_index] _, cost, accu = sess.run([ optimizer, cross_entropy,accuracy], feed_dict={x:batch_x, y:batch_y}) Cost.append(cost) Accuracy.append(accu) if i % display_step ==0: print ('Epoch : %d , Cost : %.7f'%(i+1, cost)) print ('training finished') # 代价函数曲线 fig1,ax1 = plt.subplots(figsize=(10,7)) plt.plot(Cost) ax1.set_xlabel('Epochs') ax1.set_ylabel('Cost') plt.title('Cross Loss') plt.grid() plt.show() # 准确率曲线 fig7,ax7 = plt.subplots(figsize=(10,7)) plt.plot(Accuracy) ax7.set_xlabel('Epochs') ax7.set_ylabel('Accuracy Rate') plt.title('Train Accuracy Rate') plt.grid() plt.show() #----------------------------------各个层特征可视化------------------------------- # imput image fig2,ax2 = plt.subplots(figsize=(2,2)) ax2.imshow(np.reshape(mnist.train.images[11], (28, 28))) plt.show() # 第一层的卷积输出的特征图 input_image = mnist.train.images[11:12] conv1_16 = sess.run(h_conv1, feed_dict={x:input_image}) # [16, 28, 28 ,1] conv1_reshape = sess.run(tf.reshape(conv1_16, [16, 1, 28, 28])) fig3,ax3 = plt.subplots(nrows=1, ncols=16, figsize = (16,1)) for i in range(16): ax3[i].imshow(conv1_reshape[i][0]) # tensor的切片[batch, channels, row, column] plt.title('Conv1 16x28x28') plt.show() # 第一层池化后的特征图 pool1_16 = sess.run(h_pool1, feed_dict={x:input_image}) # [16, 14, 14, 1] pool1_reshape = sess.run(tf.reshape(pool1_16, [16, 1, 14, 14])) fig4,ax4 = plt.subplots(nrows=1, ncols=16, figsize=(16,1)) for i in range(16): ax4[i].imshow(pool1_reshape[i][0]) plt.title('Pool1 16x14x14') plt.show() # 第二层卷积输出特征图 conv2_32 = sess.run(h_conv2, feed_dict={x:input_image}) # [32, 14, 14, 1] conv2_reshape = sess.run(tf.reshape(conv2_32, [32, 1, 14, 14])) fig5,ax5 = plt.subplots(nrows=1, ncols=32, figsize = (32, 1)) for i in range(32): ax5[i].imshow(conv2_reshape[i][0]) plt.title('Conv2 32x14x14') plt.show() # 第二层池化后的特征图 pool2_32 = sess.run(h_pool2, feed_dict={x:input_image}) #[32, 7, 7, 1] pool2_reshape = sess.run(tf.reshape(pool2_32, [32, 1, 7, 7])) fig6,ax6 = plt.subplots(nrows=1, ncols=32, figsize = (32, 1)) plt.title('Pool2 32x7x7') for i in range(32): ax6[i].imshow(pool2_reshape[i][0]) plt.show()
2.keras实现
- Essentials of Deep Learning: Visualizing Convolutional Neural Networks in Python
- https://github.com/keras-team/keras/blob/master/examples/conv_filter_visualization.py
- https://blog.keras.io/how-convolutional-neural-networks-see-the-world.html
- Neural network visualization toolkit for keras https://raghakot.github.io/keras-vis
![](https://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif)
'''Visualization of the filters of VGG16, via gradient ascent in input space. This script can run on CPU in a few minutes. Results example: http://i.imgur.com/4nj4KjN.jpg ''' from __future__ import print_function from scipy.misc import imsave import numpy as np import time from keras.applications import vgg16 from keras import backend as K # dimensions of the generated pictures for each filter. img_width = 128 img_height = 128 # the name of the layer we want to visualize # (see model definition at keras/applications/vgg16.py) layer_name = 'block5_conv1' # util function to convert a tensor into a valid image def deprocess_image(x): # normalize tensor: center on 0., ensure std is 0.1 x -= x.mean() x /= (x.std() + K.epsilon()) x *= 0.1 # clip to [0, 1] x += 0.5 x = np.clip(x, 0, 1) # convert to RGB array x *= 255 if K.image_data_format() == 'channels_first': x = x.transpose((1, 2, 0)) x = np.clip(x, 0, 255).astype('uint8') return x # build the VGG16 network with ImageNet weights model = vgg16.VGG16(weights='imagenet', include_top=False) print('Model loaded.') model.summary() # this is the placeholder for the input images input_img = model.input # get the symbolic outputs of each "key" layer (we gave them unique names). layer_dict = dict([(layer.name, layer) for layer in model.layers[1:]]) def normalize(x): # utility function to normalize a tensor by its L2 norm return x / (K.sqrt(K.mean(K.square(x))) + K.epsilon()) kept_filters = [] for filter_index in range(200): # we only scan through the first 200 filters, # but there are actually 512 of them print('Processing filter %d' % filter_index) start_time = time.time() # we build a loss function that maximizes the activation # of the nth filter of the layer considered layer_output = layer_dict[layer_name].output if K.image_data_format() == 'channels_first': loss = K.mean(layer_output[:, filter_index, :, :]) else: loss = K.mean(layer_output[:, :, :, filter_index]) # we compute the gradient of the input picture wrt this loss grads = K.gradients(loss, input_img)[0] # normalization trick: we normalize the gradient grads = normalize(grads) # this function returns the loss and grads given the input picture iterate = K.function([input_img], [loss, grads]) # step size for gradient ascent step = 1. # we start from a gray image with some random noise if K.image_data_format() == 'channels_first': input_img_data = np.random.random((1, 3, img_width, img_height)) else: input_img_data = np.random.random((1, img_width, img_height, 3)) input_img_data = (input_img_data - 0.5) * 20 + 128 # we run gradient ascent for 20 steps for i in range(20): loss_value, grads_value = iterate([input_img_data]) input_img_data += grads_value * step print('Current loss value:', loss_value) if loss_value <= 0.: # some filters get stuck to 0, we can skip them break # decode the resulting input image if loss_value > 0: img = deprocess_image(input_img_data[0]) kept_filters.append((img, loss_value)) end_time = time.time() print('Filter %d processed in %ds' % (filter_index, end_time - start_time)) # we will stich the best 64 filters on a 8 x 8 grid. n = 8 # the filters that have the highest loss are assumed to be better-looking. # we will only keep the top 64 filters. kept_filters.sort(key=lambda x: x[1], reverse=True) kept_filters = kept_filters[:n * n] # build a black picture with enough space for # our 8 x 8 filters of size 128 x 128, with a 5px margin in between margin = 5 width = n * img_width + (n - 1) * margin height = n * img_height + (n - 1) * margin stitched_filters = np.zeros((width, height, 3)) # fill the picture with our saved filters for i in range(n): for j in range(n): img, loss = kept_filters[i * n + j] stitched_filters[(img_width + margin) * i: (img_width + margin) * i + img_width, (img_height + margin) * j: (img_height + margin) * j + img_height, :] = img # save the result to disk imsave('stitched_filters_%dx%d.png' % (n, n), stitched_filters)
3.tflearn
![](https://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif)
""" An example showing how to save/restore models and retrieve weights. """ from __future__ import absolute_import, division, print_function import tflearn import tflearn.datasets.mnist as mnist # MNIST Data X, Y, testX, testY = mnist.load_data(one_hot=True) # Model input_layer = tflearn.input_data(shape=[None, 784], name='input') dense1 = tflearn.fully_connected(input_layer, 128, name='dense1') dense2 = tflearn.fully_connected(dense1, 256, name='dense2') softmax = tflearn.fully_connected(dense2, 10, activation='softmax') regression = tflearn.regression(softmax, optimizer='adam', learning_rate=0.001, loss='categorical_crossentropy') # Define classifier, with model checkpoint (autosave) model = tflearn.DNN(regression, checkpoint_path='model.tfl.ckpt') # Train model, with model checkpoint every epoch and every 200 training steps. model.fit(X, Y, n_epoch=1, validation_set=(testX, testY), show_metric=True, snapshot_epoch=True, # Snapshot (save & evaluate) model every epoch. snapshot_step=500, # Snapshot (save & evalaute) model every 500 steps. run_id='model_and_weights') # --------------------- # Save and load a model # --------------------- # Manually save model model.save("model.tfl") # Load a model model.load("model.tfl") # Or Load a model from auto-generated checkpoint # >> model.load("model.tfl.ckpt-500") # Resume training model.fit(X, Y, n_epoch=1, validation_set=(testX, testY), show_metric=True, snapshot_epoch=True, run_id='model_and_weights') # ------------------ # Retrieving weights # ------------------ # Retrieve a layer weights, by layer name: dense1_vars = tflearn.variables.get_layer_variables_by_name('dense1') # Get a variable's value, using model `get_weights` method: print("Dense1 layer weights:") print(model.get_weights(dense1_vars[0])) # Or using generic tflearn function: print("Dense1 layer biases:") with model.session.as_default(): print(tflearn.variables.get_value(dense1_vars[1])) # It is also possible to retrieve a layer weights through its attributes `W` # and `b` (if available). # Get variable's value, using model `get_weights` method: print("Dense2 layer weights:") print(model.get_weights(dense2.W)) # Or using generic tflearn function: print("Dense2 layer biases:") with model.session.as_default(): print(tflearn.variables.get_value(dense2.b))
C/C++基本语法学习
STL
C++ primer