Weird Validation behavior

[Shumpei-Kikuta]

Thank you for your great work!

I made an experiment with shuffle True when validation to make sure it never affects performance.
But the result I got was significantly lower than shuffle False.
I don't think that AUC is never affected by the order in the validation datasets.

Would you have any ideas?

[24hours]

this issue is related to samet-akcay/ganomaly#47