@graykode Can you explain how batch training would be conducted? For example, what if we had multiple input files for training data?
Currently, training is done using only a single data file. For multiple data files, would data_utils._create_data have to return a batch of features?
