Reset metrics before validation

[jrzaurin]

At the moment the metrics are not reseted before evaluation which means that if you print metrics as well as losses, the metrics will be computed considering the entire dataset (this does not affect losses).

A fix will happen asap in the current version (will only be available in github) and there is a new version coming where it will also be fixed (available in both, github and pypi)

[jrzaurin]

Bug Fixed.

IMPORTANT: fix only available if package installed from github.

pip install git+https://github.com/jrzaurin/pytorch-widedeep.git

New version available via both github and pip coming soon.