Training accuracy

[francescamanni1989]

Thanks for your implementation, very insightful!
Anyway, why training accuracy is so low?

I have been choosing your default hyperparameters and setting.

Thanks!

[xuyxu]

I think there should be some problems with your way of calculating the training accuracy, could you show me the code?

[francescamanni1989]

yes, ofc.

[xuyxu]

Please replace the len(train_loader.dataset) in the third line with data.size(0) , this should fix your problem.

[xuyxu]

len(train_loader.dataset) indicate the total number of batches in the dataset, instead of the number of samples in the current batch.

[francescamanni1989]

Thank you! In case I want to implement the F.softmax in the forward step of the model, is it ok to do so after y_pred = self.leaf_nodes(mu)?

[xuyxu]

Of course, this should be fine.