深度学习原理与框架-递归神经网络-RNN_exmaple(代码) 1.rnn.BasicLSTMCell(构造基本网络) 2.tf.nn.dynamic_rnn(执行rnn网络) 3.tf.expand_dim(增加输入数据的维度) 4.tf.tile(在某个维度上按照倍数进行平铺迭代) 5.tf.squeeze(去除维度上为1的维度)

sess = tf.Session()
x = tf.constant([[1,2,1],[1,2,1]])
y = tf.tile(x, [2,1])
print(sess.run(y))
# 代码结果

[[1 2 1]
[1 2 1]
[1 2 1]
[1 2 1]]

import tensorflow as tf
import numpy as np
from tensorflow.contrib import rnn


class SeriesPredictor:

    def __init__(self, input_size, seq_size, hidden_num):

        self.input_size = input_size
        self.seq_size = seq_size
        self.hidden_num = hidden_num

        # 使用tf.placeholder设置x和y， 并且设置W, b
        self.W_out = tf.Variable(tf.random_normal([hidden_num, 1]), name=‘W_out‘)
        self.b_out = tf.Variable(tf.constant(0.1, shape=[1]), name=‘b_out‘)
        self.x = tf.placeholder(tf.float32, [None, seq_size, input_size])
        self.y = tf.placeholder(tf.float32, [None, seq_size])

        self.cost = tf.reduce_mean(tf.square(self.model() - self.y))
        self.train_op = tf.train.AdamOptimizer().minimize(self.cost)

        self.saver = tf.train.Saver()


    def model(self):

        cell = rnn.BasicLSTMCell(self.hidden_num)
        outputs, states = tf.nn.dynamic_rnn(cell, self.x, dtype=tf.float32)
        num_example = tf.shape(self.x)[0]

        tf_expand = tf.expand_dims(self.W_out, axis=0)
        tf_tile = tf.tile(tf_expand, [num_example, 1, 1])
        out = tf.matmul(outputs, tf_tile) + self.b_out
        out = tf.squeeze(out)

        return out

    def train(self, train_x, train_y):
        # 第一步：构建sess， 
        with tf.Session() as sess:
            # 模型参数的复用
            tf.get_variable_scope().reuse_variables()
            # 模型初始化
            sess.run(tf.global_variables_initializer())
            for i in range(1000):
                  # 第二步：执行降低损失值和计算cost
                _, mse = sess.run([self.train_op, self.cost], feed_dict={self.x:train_x, self.y:train_y})
                  # 第三步：每循环200次，打印结果
                if i % 200 == 0:
                    print(i, mse)
            # 第四步：保存sess的参数
            save_path = self.saver.save(sess, ‘./model‘)
            print(‘Model saved to {}‘.format(save_path))

    def test(self, test_x):
        # 第一步：构建sess，并进行模型参数的复用
        with tf.Session() as sess:
            tf.get_variable_scope().reuse_variables()
            # 第二步：加载sess模型的参数
            self.saver.restore(sess, ‘./model‘)
            # 第三步：输出模型的预测结果self.model, 返回结果
            out = sess.run(self.model(), feed_dict={self.x:test_x})
            return out

if __name__ == ‘__main__‘:
    # 实例化模型，input_size表示每一次输入，seq_size表示一共有几个单元，hidden_num表示隐藏层的个数
    predictor = SeriesPredictor(input_size=1, seq_size=4, hidden_num=10)
    # 训练集的X
    train_x = [[[1], [2], [5], [6]],
              [[5], [7], [7], [8]],
               [[3], [4], [5], [7]]]
     # 训练集的y
    train_y = [[1, 3, 7, 11],
               [5, 12, 14, 15],
               [3, 7, 9, 12]]
     # 调用实例化函数的train进行模型训练
    predictor.train(train_x, train_y)
     # 测试集的x
    test_x = [[[1], [2], [3], [4]],  # 1, 3, 5, 7
              [[4], [5], [6], [7]]]
     # 测试集的y
    actual_y = [[[1], [3], [5], [7]],
                [[4], [9], [11], [13]]]
     # 调用实例化的y进行模型的预测
    pred_y = predictor.test(test_x)
     # 打印输出的结果
    for i, x in enumerate(test_x):
        print(‘the input is {}‘.format(x))
        print(‘the actual y is {}‘.format(actual_y[i]))
        print(‘the pred y is {}‘.format(pred_y[i]))