💾 Evaluation datasets #

Folds #

Having completed the preprocessing steps to generate an oracle, one must generate the training 66-33 cross-validation folds and snippets. Once again, this is already provided by this distribution in src/main/resources/<lang-name>/folds. The Python’s subproject carries out such task, hence the following commands are to be run within this subproject’s root directory.

For this purpose, the following command should be run:

python oracle_cache_generator.py generate_folds <lang-name>

where <lang-name> is one of the three languages analyzed, as given above.

Snippets #

Snippets for each fold are generated through the command below and only after folds have been produced.

python oracle_cache_generator.py generate_snippets <lang-name>

Cache #

Cache is a more efficient representation of the training datasets, in the form of collections of PyTorch Tensors, especially performed for the training of Neural Networks. Although these are already provided in src/main/resources/<lang-name>/foldscach one can generate them by running:

python oracle_cache_generator.py generate_cache <lang-name>

Please note that for this to work all folds need to be generated first.

Run All #

All of the pipelines above can be run in once with the command:

python oracle_cache_generator.py all <lang-name>