search

WentaoChen0813 / SemanticPrompt

GNU General Public License v3.0
35 stars 4 forks source link

Test accuracy always equals 100% #4

Closed zhouchunpong closed 1 year ago

zhouchunpong commented 1 year ago

During the Pre-train the feature extractor，Fine-tune the model with SP, and the test stage, Test accuracy always equals 100% Anyone else has the same situation?

LoveMiki commented 1 year ago

During the Pre-train the feature extractor，Fine-tune the model with SP, and the test stage, Test accuracy always equals 100% Anyone else has the same situation?

I did not encounter this situation, have you checked the dataset? Additionally, I think the author has set the 'split' argument to 'test' by default while validation, which should be 'split =val' in order to use the validation set to choose the best model after each training epoch.