Open prasad4fun opened 5 years ago
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.
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?
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.
Keeping existing knowledge of Lexsub/Context2vec, how to further train the specific domain corpus?