TensorFlow【极简】RNN

简洁单层RNN

import tensorflow as tf
from tensorflow.contrib.rnn import LSTMCell
from tensorflow.examples.tutorials.mnist import input_data

"""加载样本集：手写数字"""
mnist = input_data.read_data_sets('data', one_hot=True)

"""网络结构"""
num_units = 50  # 隐层神经元数量
batch_size = 500  # 批量

# 输入层
X = tf.placeholder(tf.float32, [batch_size, 28, 28])  # 批量、高、宽
y = tf.placeholder(tf.float32, [batch_size, 10])  # 10分类（0~9）

# LSTM
lstm_cell = LSTMCell(num_units)

# 获取最后一个隐层
LSTMStateTuple = lstm_cell.zero_state(batch_size, tf.float32)  # c&h
for time_step in range(28):
    h, LSTMStateTuple = lstm_cell(X[:, time_step, :], LSTMStateTuple)

# 输出层
W = tf.Variable(tf.truncated_normal([num_units, 10], stddev=.1))
b = tf.Variable(tf.constant(.1, tf.float32, [10]), dtype=tf.float32)
o = tf.matmul(h, W) + b  # output

# softmax处理、交叉熵损失
loss = tf.nn.softmax_cross_entropy_with_logits(lab