About zero shot task

[W-rudder]

I think this is a fantastic work! And it's been incredibly interesting for me! However, I have a question that I'd like to trouble you with. I'm curious about the transferability of the model, for instance, training the model on the training set from arXiv in a supervised learning setup, and then testing it on the test sets from PubMed and Cora. Can the existing code achieve this? Specifically, how should I modify the configuration file? Looking forward to your reply! Wishing you smooth work and a happy life!

[W-rudder]

Is this the correct way to modify it? Forconfig/task_config.yaml

arxiv_transfer: &arxiv_transfer
  <<: *E2E-node
  dataset: arxiv
  eval_set_constructs:
    - stage: train
      split_name: train
      dataset: arxiv
    - stage: valid
      split_name: valid
      dataset: arxiv
    - stage: test
      split_name: test
      dataset: arxiv
    - stage: test
      split_name: test
      dataset: pubmed_node
    - stage: test
      split_name: test
      dataset: cora_node

[LechengKong]

Hi @W-rudder , yes, this is the right config.

[W-rudder]

Hi @W-rudder , yes, this is the right config.

Thanks for your reply! I learned a lot from your paper and codes. Best wishes to you^_^