Difference between revisions of "NLP Status Report 2016-12-26"

From cslt Wiki
Jump to: navigation, search
(以“{| class="wikitable" !Date !! People !! Last Week !! This Week |- | rowspan="6"|2016/12/19 |Yang Feng || *s2smn: wrote the manual of s2s with tensorflow http:/...”为内容创建页面)
 
Line 29: Line 29:
 
* tried different scale vectors, and found setting >=-5000 is good
 
* tried different scale vectors, and found setting >=-5000 is good
 
* tried to change cos to only inner product, and inner product is better than cos.
 
* tried to change cos to only inner product, and inner product is better than cos.
* [[http://cslt.riit.tsinghua.edu.cn/mediawiki/images/9/9f/RNNG%2Bmm_experiment_report.pdf]]
+
* [[http://cslt.riit.tsinghua.edu.cn/mediawiki/images/9/9f/RNNG%2Bmm_experiment_report.pdf report]]
 
* read a paper [[http://cslt.riit.tsinghua.edu.cn/mediawiki/images/9/92/DEEP_BIAFFINE_ATTENTION_FOR_NEURAL_DEPENDENCY_PARSING.pdf]]
 
* read a paper [[http://cslt.riit.tsinghua.edu.cn/mediawiki/images/9/92/DEEP_BIAFFINE_ATTENTION_FOR_NEURAL_DEPENDENCY_PARSING.pdf]]
 
* trying the joint training, which got a problem of optimization.  
 
* trying the joint training, which got a problem of optimization.  
 
||
 
||
 
+
* try the joint training
 +
* read more papers [[]]
 
|-
 
|-
 
|Guli ||
 
|Guli ||

Revision as of 01:19, 26 December 2016

Date People Last Week This Week
2016/12/19 Yang Feng
  • s2smn: wrote the manual of s2s with tensorflow [nmt-manual]
  • wrote part of the code of mn.
  • wrote the manual of Moses [moses-manual]
  • Huilan: fixed the problem of syntax-based translation.
  • sort out the system and corresponding documents.
Jiyuan Zhang
  • coded tone_model,but had some trouble
  • run global_attention_model that decodes four sentences, fourfivegenerated by local_attention model
  • improve poem model
Andi Zhang
  • coded to output encoder outputs and correspoding source & target sentences(ids in dictionaries)
  • coded a script for bleu scoring, which tests the five checkpoints auto created by training process and save the one with best performance
  • extract encoder outputs
Shiyue Zhang
  • tried to add true action info when training gate, which got better results than no true actions, but still not very good.
  • tried different scale vectors, and found setting >=-5000 is good
  • tried to change cos to only inner product, and inner product is better than cos.
  • [report]
  • read a paper [[1]]
  • trying the joint training, which got a problem of optimization.
  • try the joint training
  • read more papers [[]]
Guli
  • read papers about Transfer learning and solving OOV
  • conducted comparative test
  • writing survey
  • complete the first draft of the survey
Peilun Xiao
  • use LDA to generate 10-500 dimension document vector in the rest datasets
  • write a python code about a new algorithm about tf-idf
  • debug the code