print traning_loss in the checkpoint filename

[zhenyulin]

trafficstars

As discussed in the char-nn repo document, the training_loss mark in the checkpoint filename is very useful for sampling.

would be great to bring this back for quicker exploration together with a -output_dir option to store checkpoint files trained from different parameters

[dgcrouse]

Output dir is already selectable via -checkpoint_name option.

If you would like to add loss to a checkpoint name, feel free to develop it yourself and submit a pull request.

[GabiGrin]

@zhenyulin after having the exact same need I found this PR https://github.com/jcjohnson/torch-rnn/pull/181 Until merged, I'm manually changing the file. Should work for you too :)

[InnovativeInventor]

I think adding training loss to the file name should be a flag option. Otherwise it might mess up some scripts that depend on having nice checkpoint names.