macabdul9/CASA-Dialogue-Act-Classifier

incompatible dataset labels

Opened this issue · 1 comments

Labels for the train, valid and test datasets are created independently of each other. See

classes = sorted(set(self.acts))

Even if sorted, they won't be compatible if the validate/test splits don't contain all the labels.

For the switchboard dataset, the test data contains 5 less labels than the train data, so the predictions will be off.

Fix proposed in PR #11