I have my model defined with certain structure and now just permuting between filter counts and number of layers of structure.

I am watching the output of model.fit() such as

Epoch 52
65432/85386 [=============================] - 2435s 47ms/step - loss: 0.0071 - val loss: 0.0151

both training loss and validation loss are heading downward. But, as training continues I see the same numbers for more and more epochs so for instance .0072 appeared for 12 epochs and .0073 for 11. It would be nice to see the action in the lower digits while ever more slowly dropping.

I've used tensorboard before and it's kind of overkill. I don't recall being able to see more precision in there either. It would be sufficient for me at this point to just be able to display more precision in the losses. I searched but didn't find anything. Is there an easy way to do this? A hook or something? Scientific Notation? (which doesn't automatically occur until loss gets below 0.0010) BTW the model is in float-32.




Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.