ZeroDivisionError: float division by zero

[eternaldolphin]

It seems that the code lacks some data filtration?

[tgxs002]

You should be able to fix it with larger batch size.

[eternaldolphin]

Thank you (and voice out poverty

[davidyang180]

You should be able to fix it with larger batch size.

If the memory is limited, does adding a very small number (float("1e-8")) compensation to the denominator part affect the model training?