深度学习：详解循环神经网络RNN代码及注释

# -*- coding: utf-8 -*-  
""" 
Created on Thu t 08 17:36:23 2017
 
@author: Administrator 
"""  
  
import numpy as np  
#import codecs  

  
# dataSet = open('D:\\test.txt', 'r').read() #读取txt一整个文件的内容为字符串str类型
dataSet = open('./data.txt', encoding='utf-8',errors='ignore').read()
charSet = list(set(dataSet))#去除重复的字符  
print ('charSet=',charSet,'enumerate(charSet)=',enumerate(charSet))
#打印源文件中包含的字符个数、去重后字符个数  
dataSet_size, vocab_size = len(dataSet), len(charSet)  
print ('dataSet has %d characters, %d unique.' % (dataSet_size, vocab_size)  )
#创建字符的索引表  
char_to_ix = { ch:i for i,ch in enumerate(charSet) }  
ix_to_char = { i:ch for i,ch in enumerate(charSet) }  
print ('char_to_ix=',char_to_ix )
hiddenSize = 4 # 隐藏层神经元个数
seq_length = 6 #  取多少汉字训练
learning_rate = 1e-1#学习率  
  
#网络模型  
# Input_Hidden = np.random.randn(hiddenSize, vocab_size)*0.01 # 输入层到隐藏层

# 列数和字符集中字数相同

Input_Hidden = [[-0.01384185, 0.01011995,-0.00064651, 0.01972754,-0.00522446, 0.01649676, -0.00758941, -0.02056916],
                 [ 0.0076513 , 0.01162008,0.01684354, -0.0080239 , -0.02579813 , 0.00646917,  -0.00383144, -0.0138928 ],
                 [ 0.00058557 , 0.00626551 ,-0.0061216  , 0.0143277 , -0.00884462  ,0.00061148,  -0.01373899 , 0.00581686],
                 [ 0.00890854 , 0.00743596 ,-0.00321393 , 0.00155898, -0.00512344 , 0.00602878,   0.02082394 , 0.0011709 ]]


# Hidden_Hidden = np.random.randn(hiddenSize, hiddenSize)*0.01 # 隐藏层与隐藏层

Hidden_Hidden =np.array( [[-0.00190063,-0.00280163, -0.00700358, -0.01759825],
                        [ 0.02696621,-0.02045537, -0.00928934,  0.01104184],
                        [ 0.00104444,-0.00699445, -0.01231967, -0.01586002],
                        [-0.00578787,-0.01632094, -0.0158864 , -0.00531699]])

# Hidden_Output = np.random.randn(vocab_size, hiddenSize)*0.01 # 隐藏层到输出层，输出层预测的是每个字符的概率
# print (type(Hidden_Output))
Hidden_Output =np.array([[-0.00742529 , 0.00517462  ,0.02158405 , 0.00116323],
                         [ 0.00111516, -0.00713596, 0.00960925 ,-0.0050125 ],
                         [-0.01346228, -0.03331734, 0.00499212 , 0.00025231],
                        [-0.01305254, -0.03127021,-0.00979054 ,-0.00110666],
                        [ 0.00357137, -0.00624438,-0.022717   , 0.0127503 ],
                        [ 0.00186032, -0.00792605, 0.00663416 , 0.00173953],
                        [-0.00203968,  0.00247121,-0.01086589 , 0.00794997],
                        [-0.00856974, -0.00034612,-0.00675285 , 0.00286347]])
print (type(Hidden_Output))

# Hidden_Bias = np.zeros((hiddenSize, 1)) #隐藏层偏置项
# Output_Bias = np.zeros((vocab_size, 1)) #输出层偏置项
Hidden_Bias= [[ 0.], [ 0.], [ 0.], [ 0.]]
Output_Bias= [[ 0.], [ 0.], [ 0.], [ 0.], [ 0.], [ 0.], [ 0.], [ 0.]]
# print ('Input_Hidden=',Input_Hidden,'\nHidden_Hidden=',Hidden_Hidden,'\nHidden_Output=',Hidden_Output,'\nHidden_Bias=',Hidden_Bias,'\nOutput_Bias=',Output_Bias)