search

francoisgermain / SpeechDenoisingWithDeepFeatureLosses

Speech Denoising with Deep Feature Losses
MIT License
185 stars 53 forks source link

Unable to understand 1e9 in the validation loss #5

Open thak123 opened 5 years ago

thak123 commented 5 years ago

why there is a 1e9 multiplied in the loss format string and it is absent for the train loss

`

PRINT EPOCH TRAINING LOSS AVERAGE

str = "T: %d\t " % (epoch)
if SE_LOSS_TYPE == "FL":
    for j in range(SE_LOSS_LAYERS + 1):
        str += ", %10.6e" % (np.mean(loss_train, axis=0)[j])
else:
    str += ", %10.6e" % (np.mean(loss_train, axis=0)[0])

`

`

PRINT VALIDATION EPOCH LOSS AVERAGE

str = "V: %d " % (epoch)
if SE_LOSS_TYPE == "FL":
    for j in range(SE_LOSS_LAYERS + 1):
         str += ", %10.6e" % (np.mean(loss_val, axis=0)[j] * 1e9) 
else:
    str += ", %10.6e" % (np.mean(loss_val, axis=0)[0] * 1e9)`