Comment by flashfaffe2

Thanks for the inputs.

"Bandits work best for relatively simple optimization choices at very large scale"

Have you considered different methods to address this shortcoming?