fix main to take into consideration value of batches_per_epoch argument + removed batches_per_epoch from template since it is not used

added option for passing kweargs when predicting + fixed default_model to be consistent wiht config.template.json

# Conflicts:
#	requirements.txt
#	setup.py