In the fine-tuning part in the meta test process, I want to make sure what's the update_step_test.
Does update_step_test only apply to the support set right?
Then we can train and update the test support set 10 times and save the loss.
And then we use this loss with the test query set for the real test?
Am I understanding correctly? I want to make clear about the testing process :)
Can you explain more specifically about this issue?
Can you recommend some documents or papers that describe the meta testing process you implemented in the code in detail?
Thank you for reading my questions. Have a great day!
Hello there, thanks for your answerings.
While running and studying your code last week, I have one more question in meta test part.
https://github.com/YuhaoWong0103/fsl_ts/blob/6a5fe9fb8376c935e7415b96ead499906695fe4a/maml_higher_learner.py#L106
In the fine-tuning part in the meta test process, I want to make sure what's the update_step_test. Does update_step_test only apply to the support set right? Then we can train and update the test support set 10 times and save the loss. And then we use this loss with the test query set for the real test?
Also, Is it okay to update the test query set more than 1 time? like in original MAML code. (https://github.com/cbfinn/maml/blob/master/main.py)
Am I understanding correctly? I want to make clear about the testing process :) Can you explain more specifically about this issue? Can you recommend some documents or papers that describe the meta testing process you implemented in the code in detail?
Thank you for reading my questions. Have a great day!