GitLab

Merge branch 'master' of https://gricad-gitlab.univ-grenoble-alpes.fr/dinarelm/tarc-multi-task-system

Added OO prediction management at test phase when using both token and character level information (due most probably to the fact that the model has not been learned properly). TODO: code clean-up

Added OO prediction management at test phase when using both token and character level information (due most probably to the fact that the model has not been learned properly). TODO: code clean-up

Corrected a mistake in the double-signal learning op flow. Using 2 different additional_fc for tokens and chars in the decoders. TODO: test this solution now

Sur la branche master
Votre branche est en avance sur 'origin/master' de 1 commit.
   (utilisez "git push" pour publier vos commits locaux)

Modifications qui seront validées :
	modifié :         TarcLSTMModules.py
	modifié :         TarcMultiTaskModels.py
	modifié :         tarc_multitask_sequence_generator.py

Added the possibility to train the model based on both token and character sequences (that is two losses for each task). Inference phase has not been tested yet.

Added the possibility to train the model based on both token and character sequences (that is two losses for each task). Inference phase has not been tested yet.

Few more modification for integrating the use of tokens and chars together, managed through the use of granularity_merging_flags, instead of detecting if the merging needed to be performed at running time

Latest version for paper WANLP 2020. Added the possibility to train with token, char, and token+char features at any level.

Added the possibility to generate hypotheses in the original order of the read data (useful for annotation), added the functionality for saving hypotheses in tabular and parallel format. Still TODO: code clean-up

Added the possibility to ignore output items in test data for annotation purposes. Added error rate evaluation in compute_mt_accuracy.py. Still TODO: code clean-up

Added the possibility to ignore output items in test data for annotation purposes. Added error rate evaluation in compute_mt_accuracy.py. Still TODO: code clean-up