Open kelvinqin opened 3 years ago
I found two defects inside Supervised_trainer.py, which is all about CER ---
(_train_epoches) line 267, cer = self.metric(targets, y_hats) the returned CER is just the CER on the last batch, not for the whole dataset
(_validate) line 365, cer = self.metric(targets, y_hats) the returned CER is just the CER on the last batch, not for the whole dataset
Thanks for ur attention,
Title
Description
I found two defects inside Supervised_trainer.py, which is all about CER ---
(_train_epoches) line 267, cer = self.metric(targets, y_hats) the returned CER is just the CER on the last batch, not for the whole dataset
(_validate) line 365, cer = self.metric(targets, y_hats) the returned CER is just the CER on the last batch, not for the whole dataset
Thanks for ur attention,
Linked Issues