深度学习-multi-dimensional-tf

# -*- coding: utf-8 -*- import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data if __name__=='__main__': #下载mnist数据到本地data目录下 mnist = input_data.read_data_sets('data/', one_hot=True) #批处理数据量 batch_size = 200 #设置两个tensor，用来存储 x = tf.placeholder(shape=[None, 784], dtype=tf.float32) y = tf.placeholder(shape=[None, 10], dtype=tf.float32) #设置两个tensor，用来存储权重w和偏移量b init_W = tf.Variable(tf.random_normal(shape=[784, 10]),name="w") init_b = tf.Variable(tf.random_normal(shape=[1, 10]),name="b") #多分类逻辑回归算法 _y = tf.nn.softmax(tf.matmul(x, init_W) + init_b) with tf.name_scope('loss'): # 代价函数采用softmax逻辑回归函数，然后取均方误差 loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=_y, labels=y)) tf.summary.scalar('loss', loss) #定义梯度下降法，使用最速下降法求解目标函数 # optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01) #定义梯度下降法，使用Adam下降法求解目标函数 optimizer = tf.train.AdamOptimizer(learning_rate=0.01) target = optimizer.minimize(loss) init_variables = tf.global_variables_initializer() saver = tf.train.Saver(max_to_keep=0) merged = tf.summary.merge_all() with tf.Session() as sess: writer = tf.summary.FileWriter("logs/", sess.graph) sess.run(init_variables) #训练次数，每次训练都是全部数据训练完 for step in range(1000): #存储每次训练过程中的代价函数值 loss_avg = 0 #所有数据按照每次训练batch_size个，训练完需要的总次数 batch_nums = int(mnist.train.num_examples / batch_size) for i in range(batch_nums): x_batch, y_batch = mnist.train.next_batch(batch_size) feed_dict = {x: x_batch, y: y_batch} sess.run(target, feed_dict=feed_dict) loss_avg += sess.run(loss, feed_dict=feed_dict) #平均代价函数值 loss_avg /=batch_nums summary = sess.run(merged, feed_dict=feed_dict) if step % 10 == 0: print(step,loss_avg) saver.save(sess, save_path='ckpt/mnist_logistic.ckpt', global_step=step) writer.add_summary(summary, step)