Transfer learning for Lexsub

[tim-hari]

Keeping existing knowledge of Lexsub/Context2vec, how to further train the specific domain corpus?

[orenmel]

This lexsub repository is only about evaluation, not training of models.
Regarding context2vec (https://github.com/orenmel/context2vec), fine-tuning of an existing model is not supported.

[tim-hari]

This lexsub repository is only about evaluation, not training of models.
Regarding context2vec (https://github.com/orenmel/context2vec), fine-tuning of an existing model is not supported.

Any directions on how can we do that from existing context2vec?

[orenmel]

In train_context2vec.py, you can add an option to load an existing model as done in explore_context2vec.py. Then, you'll need sentence_reader.py to be able to read new text using the same word2ind mappings as were used in that existing model. With that, you should be able to resume training.