Commits · 9108fc2ba64ccd280bb1ca2cc8989084c54c535b · Franck Dary / macaon · GitLab

Snippets Groups Projects

May 06, 2020
- Do not compute grad with newer modules · 9108fc2b
  Franck Dary authored 4 years ago
  
  9108fc2b
- Added NumericColumnModule · d2655230
  Franck Dary authored 4 years ago
  
  d2655230
- Added module UppercaseRate · 014779b1
  Franck Dary authored 4 years ago
  
  014779b1
May 05, 2020
- Each SubModule have its own Dict · d08bf04c
  Franck Dary authored 4 years ago
  
  d08bf04c
May 04, 2020
- Increased default value for rarityThreshold · ea3b87d6
  Franck Dary authored 4 years ago
  
  ea3b87d6
May 03, 2020
- EOS empty stack · 96509501
  Franck Dary authored 4 years ago
  
  96509501
- Made sentid into an extra column · 7e8b84ee
  Franck Dary authored 4 years ago
  
  7e8b84ee
- New Strategy class · 23b452ce
  Franck Dary authored 4 years ago
  
  23b452ce
- modernized EOS · d377af89
  Franck Dary authored 4 years ago
  
  d377af89
Apr 30, 2020
- Fixed bug in SplitTransModule where a member was left uninitialised · 5ed6be65
  Franck Dary authored 4 years ago
  
  5ed6be65
- Added module StateName · 10d4492e
  Franck Dary authored 4 years ago
  
  10d4492e
- Changed name of optimizer checkpoints to avoid confusion with models checkpoints · 47cbed2e
  Franck Dary authored 4 years ago
  
  47cbed2e
- Making sure optimizer have access to embeddings parameters · b4c4cd4c
  Franck Dary authored 4 years ago
  
  b4c4cd4c
- Corrected bug where embeddings were not loaded when training resumed · 93b2c58c
  Franck Dary authored 4 years ago
  
  93b2c58c
Apr 29, 2020
- Corrected a bug where dict was modified if training was resumed · 29907cb5
  Franck Dary authored 4 years ago
  
  29907cb5
- Corrected a bug where embeddings module was not sent to cuda · 431fab99
  Franck Dary authored 4 years ago
  
  431fab99
- Rare values are now treated as unknown values. Embeddings sizes now exactly match dict size · d217cf58
  Franck Dary authored 4 years ago
  
  d217cf58
- Fixed error message · 7982090e
  Franck Dary authored 4 years ago
  
  7982090e
- Implemented all modules · ca72f53e
  Franck Dary authored 4 years ago
  
  ca72f53e
- First draft of modular neural network · b4228d7b
  Franck Dary authored 4 years ago
  
  b4228d7b
Apr 27, 2020
- Updated minimum required cmake version · 5a2ea279
  Franck Dary authored 4 years ago
  
  5a2ea279
- added safety check on number of examples extracted to avoid disk filling · 2996bbfd
  Franck Dary authored 4 years ago
  
  2996bbfd
Apr 26, 2020
- Fixed dynamic oracle · 59751292
  Franck Dary authored 4 years ago
  
  59751292
Apr 24, 2020
- Updated syntax for pytorch 1.5 · bfd53a17
  Franck Dary authored 4 years ago
  
  bfd53a17
Apr 23, 2020
- Fixed bugs when there was no HEAD col in mcd · b43d29d9
  Franck Dary authored 4 years ago
  
  b43d29d9
Apr 22, 2020
- Fixed a problem where ConfigDataset was not giving batches · 2cd71b60
  Franck Dary authored 4 years ago
  
  2cd71b60
Apr 21, 2020
- better repartition of batches between states · a81d26c9
  Franck Dary authored 4 years ago
  
  a81d26c9
Apr 20, 2020
- Each classifier can have a different output layer in neural networks · a31419f9
  Franck Dary authored 4 years ago
  
  a31419f9
- Do not split for unknown tokens but replace instead · ce0503c8
  Franck Dary authored 4 years ago
  
  ce0503c8
Apr 17, 2020
- split unknown only when extracting train dataset · 28045459
  Franck Dary authored 4 years ago
  
  28045459
- Excluded parser from dynamic oracle. Set train mode to false when extracting... · 183f0297
  Franck Dary authored 4 years ago
  
  Excluded parser from dynamic oracle. Set train mode to false when extracting examples. Set dicts to closed when extracting dev examples.
  183f0297
- Setting cudann bench to true · dca3ce69
  Franck Dary authored 4 years ago
  
  dca3ce69
- Added option to chose if embeddings dropout is 2d or not · e4e9d906
  Franck Dary authored 4 years ago
  
  e4e9d906
- Added total input dropout to LSTMNetwork · 70896a64
  Franck Dary authored 4 years ago
  
  70896a64
Apr 16, 2020
- Optimizer is now a member of Classifier, and its parameters can be given in rm file · e350f836
  Franck Dary authored 4 years ago
  
  e350f836
- embeddings dropout is now a LSTM classifier parameter · 979104ff
  Franck Dary authored 4 years ago
  
  979104ff
Apr 15, 2020
- Fixed unknownValueThreshold usage · 219be1d7
  Franck Dary authored 4 years ago
  
  219be1d7
- Fixed bug in RawInputLSTM where only last context was added to. Applying dropout after relu in MLP · 2261c98b
  Franck Dary authored 4 years ago
  
  2261c98b
- Trainer now returns loss per example · f6de0f30
  Franck Dary authored 4 years ago
  
  f6de0f30
Apr 14, 2020
- do not close dict during example extraction · 0daae795
  Franck Dary authored 4 years ago
  
  0daae795