Tensorflow是一個非常好用的deep learning框架

學完了cs231n，大概就可以寫一個CNN做一下MNIST了

tensorflow具體原理可以參見它的官方文檔

然后CNN的原理可以直接學習cs231n的課程。

?

另外這份代碼本地跑得奇慢。。估計用gpu會快很多。

?

import loaddata import tensorflow as tf#生成指定大小符合標準差為0.1的正態分布的矩陣 def weight_variable(shape):initial = tf.truncated_normal(shape, stddev = 0.1)return tf.Variable(initial)#生成偏移變量 def bias_variable(shape):initial = tf.constant(0.1, shape=shape)return tf.Variable(initial)#做W與x的卷積運算，跨度為1，zero-padding補全邊界（使得最后結果大小一致） def conv2d(x, W):return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')#做2x2的max池化運算，使結果縮小4倍（面積上） def max_pool_2x2(x):return tf.nn.max_pool(x, ksize = [1, 2, 2, 1],strides=[1, 2, 2, 1], padding = 'SAME')#導入數據 mnist = loaddata.read_data_sets('MNIST_data', one_hot=True)x = tf.placeholder("float", shape=[None, 784]) y_ = tf.placeholder("float", shape=[None, 10])#filter取5x5的范圍，因為mnist為單色，所以第三維是1，卷積層的深度為32 W_conv1 = weight_variable([5, 5, 1, 32]) b_conv1 = bias_variable([32])#將輸入圖像變成28*28*1的形式，來進行卷積 x_image = tf.reshape(x, [-1, 28, 28, 1])#卷積運算，activation為relu h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)#池化運算 h_pool1 = max_pool_2x2(h_conv1)#第二個卷積層，深度為64，filter仍然取5x5 W_conv2 = weight_variable([5, 5, 32, 64]) b_conv2 = bias_variable([64])#做同樣的運算 h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) h_pool2 = max_pool_2x2(h_conv2)#full-connected層，將7*7*64個神經元fc到1024個神經元上去 W_fc1 = weight_variable([7*7*64, 1024]) b_fc1 = bias_variable([1024])#將h_pool2（池化后的結果）打平后，進行fc運算 h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64]) h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)#防止過擬合，fc層進行dropout處理，參數為0.5 keep_prob = tf.placeholder("float") h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)#第二個fc層，將1024個神經元fc到10個最終結果上去（分別對應0~9） W_fc2 = weight_variable([1024, 10]) b_fc2 = bias_variable([10])#最后結果 y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)#誤差函數使用交叉熵 cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))#梯度下降使用adam算法 train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)#正確率處理 correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))#初始化 sess = tf.Session() sess.run(tf.initialize_all_variables())#進行訓練 for i in range(20000):batch = mnist.train.next_batch(50)if i%100 == 0:train_accuracy = sess.run(accuracy, feed_dict = {x:batch[0], y_:batch[1], keep_prob : 1.0})print("step %d, accuracy %g" % (i, train_accuracy))sess.run(train_step, feed_dict={x:batch[0], y_:batch[1], keep_prob:0.5})#輸出最終結果 print(sess.run(accuracy, feed_dict={x:mnist.test.images, y_:mnist.test.labels, keep_prob:1.0}))

?

轉載于:https://www.cnblogs.com/Saurus/p/7487720.html

總結

以上是生活随笔為你收集整理的Tensorflow框架初尝试————搭建卷积神经网络做MNIST问题的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。