ltgoslo / multitask_negation_for_sa Goto Github PK

Hi!

Thank you for making the code available!
I have a question regarding the hierarchical_training model. I'm currently running it and I observed that the file only contains the train_model function. For getting the results (precision, recall, f1) do I have to implement another function or is there any other file that I didn't see and already contains the implementation?

Thank you for your time.
Best regards.

Thanks for the very insightful publication and for making the code available here! Two minor things/questions:

• I couldn't quite figure out which embeddings were used for training/validating. The model/training both refer to "../../embeddings/google.txt" as a default for the embeddings and I used Google News 2013 from here which seems to work fine. Could you maybe clarify what exactly was used? And sorry if this was mentioned somewhere and I just missed it.
• In addition to the dependencies mentioned in the README (scikit-learn, torch, torchvision and nltk) I also had to explicitly install (i.e. those weren't pulled as requirements for something else) the following in a fresh environment to get everything up and running: matplotlib, tqdm and torchtext.

Thanks again!

Christoph