“NLP Status Report 2016-09-26”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(2位用户的2个中间修订版本未显示)
第9行: 第9行:
 
*read the code of rnng
 
*read the code of rnng
 
||
 
||
*run the experiments of MN grammar
+
*modify the code to MN grammar and run the experiments
 
|-
 
|-
 
|Jiyuan Zhang ||
 
|Jiyuan Zhang ||
 
*ran two versions of the code that qixin gave me,the last version of the result is good
 
*ran two versions of the code that qixin gave me,the last version of the result is good
 
*perfected my poem's code
 
*perfected my poem's code
*Results for all versions of the code[[http://cslt.riit.tsinghua.edu.cn/mediawiki/index.php/%E6%96%87%E4%BB%B6:Theme.pdf here]]
+
*Results for all versions of the code[[http://cslt.riit.tsinghua.edu.cn/mediawiki/images/7/72/Theme.pdf here]]
 
||
 
||
 
*perfect my code  according to up-to-date version of qixin's code
 
*perfect my code  according to up-to-date version of qixin's code
第35行: 第35行:
 
|-
 
|-
 
|Shiyue Zhang ||  
 
|Shiyue Zhang ||  
*ran rnng code successfully
+
*ran rnng code successfully: the result of discriminative model is a bit better than original paper; the generative model has not fully trained, so the performance is worse.
 
||
 
||
 
*get generative model f1 score
 
*get generative model f1 score

2016年9月26日 (一) 06:18的最后版本

Date People Last Week This Week
2016/09/26 Yang Feng
  • prepared the phrasal check of Huilan
  • prepared and discussed the main idea of next work
  • learned Lua and Torch and read the code of MemNN
  • read the code of rnng
  • modify the code to MN grammar and run the experiments
Jiyuan Zhang
  • ran two versions of the code that qixin gave me,the last version of the result is good
  • perfected my poem's code
  • Results for all versions of the code[here]
  • perfect my code according to up-to-date version of qixin's code
Aodong Li
Andi Zhang
  • read the sorce code of MemN2N
  • installed torch
  • try to fit the params of rnng into memn2n
Shiyao Li
Shiyue Zhang
  • ran rnng code successfully: the result of discriminative model is a bit better than original paper; the generative model has not fully trained, so the performance is worse.
  • get generative model f1 score
  • get the states of each timestep