Hugo101/PLATINUM

Some doubts

prinshul opened this issue · 1 comments

prinshul commented

Thank you for sharing the codebase.

Can you please tell:

  1. How do you select the model to get the results reported in the paper?
    Is it "meta-test" accuracy corresponding to the highest "meta-val" accuracy in 400/600 epochs?

  2. The MAML paper reports around 63% acc in their paper in a fully supervised setting (when all labels are available).
    But your paper reports around 63% with only 20% labels? Can you please explain this?

  3. Does the performance of your method varies with different count of "test queries"?
    If we vary the test query to 5, 10,15, will the performance of PLATINUM vary?

prinshul commented

All questions are answered.

Thank you