Closed Wensi-Tang closed 1 year ago
Thanks so much! You are doing it well, but I just think this might be dangerous for someone to reproduce the result.
This is a good point. Maybe it would be best to add a warning, in case someone uses --test_only
option but does not use the --norm_from
option.
Hello, yes, the assumption is that the model will first be trained, or otherwise training set statistics will be provided, before proceeding to evaluation on the test set.