/RNN

循环神经网络

Primary LanguagePython

RNN

循环神经网络

利用循环神经网络实现语言模型。通过训练过程,将选择下一个单词范围从1W减到180个,程序结果如下

In iteration: 1
After 0 steps,perplexity is 10007.465
After 100 steps,perplexity is 1444.825
After 200 steps,perplexity is 1072.107
After 300 steps,perplexity is 896.869
After 400 steps,perplexity is 786.424
After 500 steps,perplexity is 710.617
After 600 steps,perplexity is 654.160
After 700 steps,perplexity is 604.131
After 800 steps,perplexity is 559.369
After 900 steps,perplexity is 524.125
After 1000 steps,perplexity is 497.001
After 1100 steps,perplexity is 471.041
After 1200 steps,perplexity is 449.679
After 1300 steps,perplexity is 430.657
Epoch : 1 Validation perplexity:241.407
In iteration: 2
After 0 steps,perplexity is 367.913
After 100 steps,perplexity is 247.199
After 200 steps,perplexity is 251.614
After 300 steps,perplexity is 252.439
After 400 steps,perplexity is 248.990
After 500 steps,perplexity is 246.383
After 600 steps,perplexity is 245.405
After 700 steps,perplexity is 242.728
After 800 steps,perplexity is 237.978
After 900 steps,perplexity is 235.127
After 1000 steps,perplexity is 233.300
After 1100 steps,perplexity is 229.707
After 1200 steps,perplexity is 227.011
After 1300 steps,perplexity is 224.152
Epoch : 2 Validation perplexity:184.539
test Perplexity:179.757

预测正弦函数,预测值与准确的差值

''' Mean Square Error is: 0.007281 '''