Research best layer count and size for Q-Function

[krutovsky-danya]

In PR#27 new layer has been added to Q-Function It could be helpfull to research influence of net params to final result

• May be net is to big and smaller nets are no worse
• May be sizes of layer better be different