fixed double logged train loss and boolean parser

[MarognaLorenzo]

1. train loss was logged twice at training time.

The bug was not creating any malfunctioning in the training process, but the train loss was read as double its actual amount in the `train_log.txt'.

2. Boolean parser was not working properly

Boolean parameters can now be passed in the form:
python train.py --<numeric_parameter>=256 --<boolean_parameter_true>=True --<boolean_parameter_false>=False --<boolean_parameter_true>=1 --<boolean_parameter_false>=0 --<boolean_parameter_true>=y --<boolean_parameter_false>=n
and will be processed correctly. Before, any value passed was evaluated as True

[fpaissan]

Seems good to me, already checked and discussed together.