TF之LSTM：基于tensorflow框架自定义LSTM算法实现股票历史(1990~2015数据集,6112预测后100+单变量最高)行情回归预测

1. def LSTM(batch):      
2.     w_in=weights['in']
3.     b_in=biases['in']
4.     input_rnn=tf.matmul(input,w_in)+b_in 
5.     input_rnn=tf.reshape(input_rnn,[-1,time_step,rnn_unit]) 
6.     cell=tf.nn.rnn_cell.BasicLSTMCell(rnn_unit)  
7.     init_state=cell.zero_state(batch,dtype=tf.float32) 
8.     output_rnn,final_states=tf.nn.dynamic_rnn(cell, input_rnn,initial_state=init_state, dtype=tf.float32)   
9.     output=tf.reshape(output_rnn,[-1,rnn_unit]) 
10.     w_out=weights['out']
11.     b_out=biases['out']
12.     pred=tf.matmul(output,w_out)+b_out 
13. return pred,final_states