Commits · ecf7290b97c4de62a4e8f647b67c66b4c98e98f0 · Franck Dary / macaon

Apr 13, 2020
- Added dynamical oracle and extracted examples are savec to the disk not to use too much memory · ecf7290b
  Franck Dary authored 4 years ago
  
  ecf7290b
Apr 09, 2020
- addded parameter machine to macaonTrain · 9144f997
  Franck Dary authored 4 years ago
  
  9144f997
- print number of parameters of the model · 86ded225
  Franck Dary authored 4 years ago
  
  86ded225
Apr 06, 2020
- Fixed detail in print · 79b3be6a
  Franck Dary authored 4 years ago
  
  79b3be6a
- reset strategy before extracting dev examples · 35a22b34
  Franck Dary authored 4 years ago
  
  35a22b34
Apr 05, 2020
- Made splitwords transition set optional in definition of reading machine · b9a3230c
  Franck Dary authored 4 years ago
  
  b9a3230c
- Sequential · 050839ed
  Franck Dary authored 4 years ago
  
  050839ed
Apr 03, 2020
- Corrected a bug where splitword had 0 cost even when it didn't match the size of the gold multiword · 03221c3f
  Franck Dary authored 4 years ago
  
  03221c3f
- Print time in extract examples · f75f941f
  Franck Dary authored 4 years ago
  
  f75f941f
Apr 02, 2020
- added transition split · d438d3c3
  Franck Dary authored 4 years ago
  
  d438d3c3
- Only transfer tensor to gpu the moment we serve them throught method get of ConfigDataset · 9c793241
  Franck Dary authored 4 years ago
  
  9c793241
Apr 01, 2020
- Checkpoints are created after each training epoch and it is possible to resume... · e45d45e6
  Franck Dary authored 4 years ago
  
  Checkpoints are created after each training epoch and it is possible to resume a training by training again on the same directory
  e45d45e6
Mar 31, 2020
- Print time during training · 9fb78420
  Franck Dary authored 4 years ago
  
  9fb78420
Mar 17, 2020
- macaon is now one executable · 05217688
  Franck Dary authored 5 years ago
  
  05217688
Mar 15, 2020
- NeuralNetwork::extractContext can now generate multiple variants of context · 26cc83e4
  Franck Dary authored 5 years ago
  
  26cc83e4
Mar 11, 2020
- During trainning, convert rare forms to unknownValue to train the corresponding embedding · 72c543b6
  Franck Dary authored 5 years ago
  
  72c543b6
Mar 10, 2020
- Close dict for decode · c8ea7e12
  Franck Dary authored 5 years ago
  
  c8ea7e12
Mar 09, 2020
- Added Dropout to CNNNetwork · 983dc489
  Franck Dary authored 5 years ago
  
  983dc489
- Neural network now sees multiwords, also ID can now be a focused column · 1efb791a
  Franck Dary authored 5 years ago
  
  1efb791a
Mar 08, 2020
- Improved prints · 182b62de
  Franck Dary authored 5 years ago
  
  182b62de
- Fixed problem where model would not be saved · 5d2ef7b9
  Franck Dary authored 5 years ago
  
  5d2ef7b9
Mar 07, 2020
- Added transitions for tokenizer · 7c5c406e
  Franck Dary authored 5 years ago
  
  7c5c406e
Mar 05, 2020
- Do not print advancement when decoding dev · 54867a47
  Franck Dary authored 5 years ago
  
  54867a47
- Added option to decide if dev must be evaluated or not · b743e4d3
  Franck Dary authored 5 years ago
  
  b743e4d3
- Printing decoding speed · a4fe4f0b
  Franck Dary authored 5 years ago
  
  a4fe4f0b
- Improved error messages in case of trying to use an unknown column · 764c2fc7
  Franck Dary authored 5 years ago
  
  764c2fc7
Mar 03, 2020
- Detect device (cpu or cuda) and use cuda if possible · f93305a5
  Franck Dary authored 5 years ago
  
  f93305a5
Feb 26, 2020
- Implemented RTLSTM · 5aafa66a
  Franck Dary authored 5 years ago
  
  5aafa66a
Feb 25, 2020
- Reseting strategy between different corpuses · d145be52
  Franck Dary authored 5 years ago
  
  d145be52
- Set amsgrad as default optimizer · b0bb4445
  Franck Dary authored 5 years ago
  
  b0bb4445
- Allow batch size 1 · d846104c
  Franck Dary authored 5 years ago
  
  d846104c
Feb 24, 2020
- Print speed during training · fb4e0869
  Franck Dary authored 5 years ago
  
  fb4e0869
Feb 23, 2020
- Set hyperparameters of adam · a7558853
  Franck Dary authored 5 years ago
  
  a7558853
- Removed distinction between dense and sparse parameters because it was hurting... · 97b8d2fc
  Franck Dary authored 5 years ago
  
  Removed distinction between dense and sparse parameters because it was hurting performances and the advantage in speed was not significant
  97b8d2fc
Feb 21, 2020
- Changed the loss to CrossEntropy, to avoid using log (caused exploding gradient) · a4844d76
  Franck Dary authored 5 years ago
  
  a4844d76
Feb 20, 2020
- Changed extract context · 096b59d9
  Franck Dary authored 5 years ago
  
  096b59d9
Feb 19, 2020
- Added base class to neural network · 82827388
  Franck Dary authored 5 years ago
  
  82827388
Feb 17, 2020
- When printing scores on dev, show metrics · 2465f80b
  Franck Dary authored 5 years ago
  
  2465f80b
Feb 16, 2020
- Forced EOS at the end of decode · b6fda44a
  Franck Dary authored 5 years ago
  
  b6fda44a
Feb 14, 2020
- Added movement in strategy · 9bc1ea8d
  Franck Dary authored 5 years ago
  
  9bc1ea8d