Nlp-progress 2017/03

From cslt Wiki
Jump to: navigation, search

Daily Report

Date Person start leave hours status
2017/3/01 Andy Zhang 9:30 18:30 8
  • tested the effect of masks on baseline_beam, turns out it is a little better than without masks
Peilun Xiao
2017/3/02 Andy Zhang 9:30 18:30 8
  • try to use the 'sequence_length' instead of masks
Peilun Xiao
2017/3/03 Andy Zhang 9:30 18:30 8
  • finished coding, needed to be further tested
Peilun Xiao
2017/3/06 Andy Zhang 9:30 18:30 8
  • abandoned the thought from last week
  • talked about graduation design with Pro. Wang
Peilun Xiao
2017/3/07 Andy Zhang 9:30 18:30 8
  • try to use memory to solve oov problem
Peilun Xiao
2017/3/08 Andy Zhang 9:30 18:30 8
  • try to use memory to solve oov problem, first try word that is not in source dict but in target dict
Peilun Xiao
2017/3/09 Andy Zhang 9:30 18:30 8
  • it worked
  • try word that is in neither source dict nor target dict
Peilun Xiao
2017/3/10 Andy Zhang 9:30 18:30 8
  • met with bug that the model stopped predicting <EOS> symbol
Peilun Xiao
2017/3/13 Andy Zhang
Peilun Xiao
2017/3/14 Andy Zhang 9:30 18:30 8
  • try to fix the bug
Peilun Xiao
2017/3/15 Andy Zhang 9:30 18:30 8
  • try to fix the bug
Peilun Xiao
2017/3/16 Andy Zhang 9:30 18:30 8
  • try to fix the bug
Peilun Xiao
2017/3/17 Andy Zhang 9:30 18:30 8
  • try to fix the bug
Peilun Xiao
2017/3/20 Andy Zhang 9:30 18:30 8
  • try to fix the bug
  • turns out the bug rises from the unfamiliarity with numpy.resize() function
  • the model can now deal with oov problem
Peilun Xiao
2017/3/21 Andy Zhang 9:30 18:30 8
  • working on graduation design document
Peilun Xiao
2017/3/22 Andy Zhang 9:30 18:30 8
  • working on graduation design document
Peilun Xiao
2017/3/23 Andy Zhang 9:30 18:30 8
  • working on graduation design document
Peilun Xiao
2017/3/24 Andy Zhang 9:30 18:30 8
  • working on graduation design document
Peilun Xiao
2017/3/27 Andy Zhang 9:30 18:30 8
  • rerun cs2en data on theano
  • run wmt17 zh2en data on theano
Peilun Xiao
2017/3/28 Andy Zhang 9:30 18:30 8
  • run zh2en data from Lihang's on theano
  • test the best bleu models on test set
Peilun Xiao
2017/3/29 Andy Zhang 9:30 18:30 8
  • rerun experiments as on Monday
  • tested results
Peilun Xiao
2017/3/30 Andy Zhang
Peilun Xiao
2017/3/31 Andy Zhang
Peilun Xiao