keras 的 example 文件 lstm_text_generation.py 解析

该程序是学习现有的文章，然后学习预测下个字符，这样一个字符一个字符的学会写文章

 

先打印下char_indices

{'\n': 0, ' ': 1, '!': 2, '"': 3, "'": 4, '(': 5, ')': 6, ',': 7, '-': 8, '.': 9, '0': 10, '1': 11, '2': 12, '3': 13, '4': 14, '5': 15, '6': 16, '7': 17, '8': 18, '9': 19, ':': 20, ';': 21, '=': 22, '?': 23, '[': 24, ']': 25, '_': 26, 'a': 27, 'b': 28, 'c': 29, 'd': 30, 'e': 31, 'f': 32, 'g': 33, 'h': 34, 'i': 35, 'j': 36, 'k': 37, 'l': 38, 'm': 39, 'n': 40, 'o': 41, 'p': 42, 'q': 43, 'r': 44, 's': 45, 't': 46, 'u': 47, 'v': 48, 'w': 49, 'x': 50, 'y': 51, 'z': 52, 'ä': 53, 'æ': 54, 'é': 55, 'ë': 56}

 

然后构造训练数据，输入是 sentences，输出是 next_chars，构造成如下结构，sentences就是把句子拆分出来，next_chars，名字就看出来了，就是下一个字符

                                  sentences next_chars
preface\n\n\nsupposing that truth is a woma          n
face\n\n\nsupposing that truth is a woman--          w
e\n\n\nsupposing that truth is a woman--wha          t
  \nsupposing that truth is a woman--what t          h
   pposing that truth is a woman--what then          ?
   sing that truth is a woman--what then? i          s
   g that truth is a woman--what then? is t          h
   hat truth is a woman--what then? is ther          e
    truth is a woman--what then? is there n          o
   uth is a woman--what then? is there not           g
    is a woman--what then? is there not gro          u
    a woman--what then? is there not ground         \n
  woman--what then? is there not ground\nfo          r
  an--what then? is there not ground\nfor s          u
  -what then? is there not ground\nfor susp          e
  at then? is there not ground\nfor suspect          i
  then? is there not ground\nfor suspecting
  n? is there not ground\nfor suspecting th          a
  is there not ground\nfor suspecting that           a
  there not ground\nfor suspecting that all
  re not ground\nfor suspecting that all ph          i
  not ground\nfor suspecting that all philo          s
   ground\nfor suspecting that all philosop          h
  ound\nfor suspecting that all philosopher          s
  d\nfor suspecting that all philosophers,           i
   or suspecting that all philosophers, in           s
   suspecting that all philosophers, in so           f
   pecting that all philosophers, in so far
   ting that all philosophers, in so far as
   g that all philosophers, in so far as th          e
   hat all philosophers, in so far as they           h
    all philosophers, in so far as they hav          e
   l philosophers, in so far as they have b          e
   hilosophers, in so far as they have been         \n
  osophers, in so far as they have been\ndo          g
  phers, in so far as they have been\ndogma          t
  rs, in so far as they have been\ndogmatis          t
   in so far as they have been\ndogmatists,
   so far as they have been\ndogmatists, ha          v
   far as they have been\ndogmatists, have           f
  r as they have been\ndogmatists, have fai          l
  s they have been\ndogmatists, have failed
  hey have been\ndogmatists, have failed to
   have been\ndogmatists, have failed to un          d
  ve been\ndogmatists, have failed to under          s
  been\ndogmatists, have failed to understa          n
  n\ndogmatists, have failed to understand           w
   ogmatists, have failed to understand wom          e
   atists, have failed to understand women-          -
   sts, have failed to understand women--th          a

啊，有一点，就是上面的sentence，直接看起来好像不一样长，实际是一样长的，只不过前面三行，有两个\n，在打印的时候是两个字符，实际上\n是一个字符，导致的看起来不整齐

然后进行one-hot编码，这都是NLP的常规操作，然后输入输出数据shape为：

x.shape  (200285, 40, 57)
y.shape  (200285, 57)

神经网络模型为

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
lstm_1 (LSTM)                (None, 128)               95232
_________________________________________________________________
dense_1 (Dense)              (None, 57)                7353
=================================================================
Total params: 102,585
Trainable params: 102,585
Non-trainable params: 0
_________________________________________________________________

 

——————————————————————

总目录

keras的example文件解析