“Jt-chinese”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
Lr讨论 | 贡献
第1行: 第1行:
==data and model==
+
=data and model=
 
* train
 
* train
 
:* size: 62M
 
:* size: 62M
第21行: 第21行:
 
* learning rate
 
* learning rate
 
:* 0.1-0.00625
 
:* 0.1-0.00625
 +
 +
=sample data from rnnlm=
 +
* different size of simple data
 +
{| border="2px"
 +
|+ different size of simple data
 +
|-
 +
! size  !! ppl !! mix0.3 !! mix0.5 !! mix0.7
 +
|-
 +
!50M
 +
| 105.457 || 86.7 || 87.5 || 89.7
 +
|-
 +
!100M
 +
| ||
 +
|-
 +
!150
 +
|}

2014年12月1日 (一) 01:16的版本

data and model

  • train
  • size: 62M
  • 8k-sentence from jt(about dianxin)
  • dev
  • 1000 row from train data
  • dict
  • chn_150576.txt(15w)
  • model
Train Set Environment
Parameters hidden class direct bbt bptt_block threads direct-order rand_seed nwords time(min) iter
set1 320 300 2000 5 20 1 3 1 10000 (31h) 8
  • ppl
  • dev:86-66(ppl)
  • learning rate
  • 0.1-0.00625

sample data from rnnlm

  • different size of simple data
different size of simple data
size ppl mix0.3 mix0.5 mix0.7
50M 105.457 86.7 87.5 89.7
100M
150