bug in train.py

[micuentadecasa]

dear all,

I have found a bug in the code, in train.py, in line 294, you call the function run without taking into account if the training is done in remote

_# after training is complete, we log the final loss on the validation set
run.log('final_val_loss', history.history['val_loss'][-1])_

it should be

_# after training is complete, we log the final loss on the validation set
if remote_execution:
      run.log('final_val_loss', history.history['val_loss'][-1])_

[uname0x96]

you just need try to defind 1 custom func for write it.

[wmpauli]

Thanks. The sdk should catch that and simply print a warning that it didn't log to aml because running locally.