TensorFlow之双隐含层多层感知器（MLP）

程序改自上一篇博客，使用了双隐含层，第二层隐含层初始w需要和第一层类似，否则程序正确率一直在0.1左右。修改后的程序正确率也在98%左右。

# -*- coding:utf-8 -*-
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf

# data
dir='/home/kaka/Documents/input_data'
mnist = input_data.read_data_sets(dir, one_hot=True)

# model
sess = tf.InteractiveSession()

hd1in_units = 784
hd1out_units = 500
hd2out_units = 300

w1 = tf.Variable(tf.truncated_normal([hd1in_units, hd1out_units], stddev=0.1))
b1 = tf.Variable(tf.zeros(hd1out_units))
# w2 = tf.Variable(tf.zeros([hd1out_units, hd2out_units]))
w2 = tf.Variable(tf.truncated_normal([hd1out_units, hd2out_units], stddev=0.1))
b2 = tf.Variable(tf.zeros([hd2out_units]))
w3 = tf.Variable(tf.zeros([hd2out_units, 10]))
b3 = tf.Variable(tf.zeros([10]))


x = tf.placeholder(tf.float32, [None, hd1in_units])

keep_prob = tf.placeholder(tf.float32)   # dropout proportion
hidden1 = tf.nn.relu(tf.matmul(x, w1) + b1)
hidden1_drop = tf.nn.dropout(hidden1, keep_prob)

hidden2 = tf.nn.relu(tf.matmul(hidden1_drop, w2) + b2)
hidden2_drop = tf.nn.dropout(hidden2, keep_prob)

y = tf.nn.softmax(tf.matmul(hidden2_drop, w3) + b3)

# loss
y_ = tf.placeholder(tf.float32, [None, 10])
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y))

correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

train_step = tf.train.AdagradOptimizer(0.3).minimize(cross_entropy)
# train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

tf.global_variables_initializer().run()
for i in range(100000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 0.8})
    if i % 100 == 0:
        train_accuracy = accuracy.eval(feed_dict={
            x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})
        print('step %d, training accuracy %g' % (i, train_accuracy))

print(sess.run(accuracy, feed_dict={x: mnist.test.images,
                                    y_: mnist.test.labels,
                                    keep_prob: 1.0}))