train script

[pankhuriVarshney]

prepare the train.py script.
fix the learning rate and batch size (find a more appropriate figure for them)
use float states instead of booleans in the get_state() function