chenzimin

Hi @Ringbo and @dangnguyenngochai, I believe the training script for all models (the best and ablation models) is included. For example, in each `*_models/*_parameter_sweep` directory, `train_config.yml` and `vocab_config.yml` contain the...

Hi @dangnguyenngochai, I could not open the log file.

I can see the log with the new link. The training accuracy is really high (almost 100%), but the validation accuracy is only `[2022-05-06 01:14:23,334 INFO] Validation accuracy: 49.948`. So...

I believe that OpenNMT-py reports per token accuracy, whereas we calculate the sequence accuracy.

One more thing, you can use [src/find_best_model_and_translate_config.py](https://github.com/SteveKommrusch/VRepair/blob/main/src/find_best_model_and_translate_config.py) to generate command to generate the predictions from the model with the best validation accuracy.

We used NVIDIA Tesla T4 to train all the models. > transfer learning in rq2 source domain training (training on a bug fix corpus) takes ~3 - 7 days, target...

I am now contacting Zenodo to host all the models (compressed version is 273GB). I have them currently uploaded to OneDrive for now: * https://kth-my.sharepoint.com/:u:/g/personal/zimin_ug_kth_se/EV9uVzTRIF5HlMrRNHPHjfoBOivr63Is023EFzNe1Ax9zg?e=dEck8z * https://kth-my.sharepoint.com/:u:/g/personal/zimin_ug_kth_se/EVy6M7UZ0rVPr3c28wxKH2cBYZPKGSPqb5_lqLOrduzvlQ?e=lgn8TV * https://kth-my.sharepoint.com/:u:/g/personal/zimin_ug_kth_se/EVTHzJRcIW9Ak5PVn38RYX8BEWH5zyKlSlJh9KgKXU3SYw?e=WgSkHp *...

The link will be valid until Sep 2022. The models can be stored at OneDrive at least until Oct 2023, in the meantime I will try to find a location...

I believe that we used `BugFixNoDup_` to generate the dataset. But if you are interested in using the dataset with context size 3, they are already in the `data` folder...

Impressive result! > NB: I believe, that the metric is called mean reciprocal rank (MRR), not mean average precision. Thanks, I did not know that. It seems that MRR is...