1.CNN_my_test.py
import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets('data/', one_hot=True) print('数据ok') print(mnist.train.images[0].shape) def weight_initializer(shape): initializer = tf.truncated_normal(shape, stddev= 0.1) return tf.Variable(initializer) def biases_initializer(shape): initializer = tf.constant(0.1, shape=shape) return tf.Variable(initializer) x = tf.placeholder(tf.float32, shape=[None, 784]) y = tf.placeholder(tf.float32, shape=[None, 10]) x_image = tf.reshape(x, [-1, 28, 28, 1]) tf.summary.image('input', x_image, 1) wc1 = weight_initializer([5, 5, 1, 32]) bc1 = biases_initializer([32]) hc1 = tf.nn.relu(tf.nn.conv2d(x_image, wc1, strides=[1, 1, 1, 1], padding='SAME') + bc1) pool_hc1 = tf.nn.max_pool(hc1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') wc2 = weight_initializer([5, 5, 32, 64]) bc2 = biases_initializer([64]) hc2 = tf.nn.relu(tf.nn.conv2d(pool_hc1, wc2, strides=[1, 1, 1, 1], padding='SAME') + bc2) pool_hc2 = tf.nn.max_pool(hc2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') wd1 = weight_initializer([7*7*64, 1024]) bd1 = biases_initializer([1024]) hc2_flat = tf.reshape(pool_hc2, [-1, 7*7*64]) hd1 = tf.nn.relu(tf.matmul(hc2_flat, wd1) + bd1) hd1_dp = tf.nn.dropout(hd1, keep_prob=0.7) wd2 = weight_initializer([1024, 10]) bd2 = biases_initializer([10]) y_conv = tf.nn.softmax(tf.matmul(hd1_dp, wd2) + bd2) cross_entropy = tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=y)) tf.summary.scalar('cross entropy', cross_entropy) train_step = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(cross_entropy) corr = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y, 1)) acc = tf.reduce_mean(tf.cast(corr, tf.float32)) sess = tf.Session() sess.run(tf.global_variables_initializer()) merged = tf.summary.merge_all() log_dir = './log' train_writer = tf.summary.FileWriter(log_dir + '/train', sess.graph) for i in range(2000): if i % 100 != 0: batch = mnist.train.next_batch(50) train_step.run(session=sess, feed_dict={x: batch[0], y: batch[1]}) summary, _ = sess.run([merged, train_step], feed_dict={x: batch[0], y: batch[1]}) train_writer.add_summary(summary, i) else: batch = mnist.train.next_batch(50) train_accuracy = acc.eval(session=sess, feed_dict={x: batch[0], y: batch[1]}) test_accuracy = acc.eval(session=sess, feed_dict={x: mnist.test.images[0:50], y: mnist.test.labels[0:50]}) print('train_acc: %.5f, test_acc: %.5f' % (train_accuracy, test_accuracy)) run_metadata = tf.RunMetadata() train_writer.add_run_metadata(run_metadata, 'step%03d' % i) summary, _ = sess.run([merged, train_step], feed_dict={x: batch[0], y: batch[1]}) train_writer.add_summary(summary, 1) print('训练完成!!') train_writer.close()
分3个部分
1.将需要记录的变量用一下函数记录
图像
tf.summary.image('input', x_image, 1)
散点图
tf.summary.scalar('cross entropy', cross_entropy)
2.生成实现变量记录的对象,和记录文件路径
merged = tf.summary.merge_all() log_dir = './log' train_writer = tf.summary.FileWriter(log_dir + '/train', sess.graph)
3.训练时进行记录
summary, _ = sess.run([merged, train_step], feed_dict={x: batch[0], y: batch[1]}) train_writer.add_summary(summary, 1)