search

qmpham / Dual_Learning

7 stars 4 forks source link

How to run the code now? #2

Open 0uO opened 6 years ago

0uO commented 6 years ago

I think the description in the README.md is currently not match to code……where is the script/train_dual_Q_value_shared.sh ? Any DIR configuration in the code need I modify?

qmpham commented 5 years ago

Well, sorry about that. I may have removed the script by accident. I will update the code source.

qmpham commented 5 years ago

You have to modify the config___.py file where you will put the location of MT model, LM model, parallel corpora, monolingual corpora. About configuration of theano, there is nothing special. The script train_dual_Q_value_shared.sh should be the same of train_dual_shared.sh except launching config_Q_value_sample.py rather than config_sample.py. The variable $index in the script can be omited.

qmpham commented 5 years ago

One more thing about evaluation scripts validate_dual_en_fr.sh and validate_dual_fr_en.sh that actually you can modify the prefix, dev, ref as you want where prefix is prefix of the name of file containing MT parameters, dev and ref are validation datasets (source and inference).

The evaluation script that is launched during training is Popen([external_validation_script_en_fr, args_en_fr_1, args_en_fr_2, str(training_progress_en_fr.uidx)]) where args_en_fr_1, args_en_fr_2 are index of experiments that i did (args_en_fr_1 is index of config; args_en_fr_2 is the name of parallel corpora); however you can leave them off or create another script if you have another idea.

Sorry, my source code is kind of messy. I will rearrange it if I have time. Thanks for your interests!