Batch

[hash2430]

when you give input to the model, do you give one npz to the model at a time?
It seems

1. train() is called for every epoch
2. "for full_txt, full_feat, spkr in train_enum" is called for every batch
3. "for txt, feat, spkr, start in batch_iter:" is called for every npz
   ==> model.forward() is called here then the loss is sumed up for batch
   Then what is the meaning of having batch at all when model is not called batch-wise?