search

mbithy / Trevel

The smartest Bitcoin Gambling bot for freebitco.in.
88 stars 56 forks source link

improve after learning? #14

Closed panda2048 closed 6 years ago

panda2048 commented 6 years ago

I am expecting the DQN was there to improve the "dice guess", so over time, the max loose streak will be lower down (i.e. lower down the min. satoish required to start) But after running for 2 millions round. I do not actually seeing significant change Is my understanding correct? If so, when should we expect the improvement?

mbithy commented 6 years ago

yup, theoretically, this could be an indication that there is no advantage to be gained by using MA OR the ReinforceJS mathematical equations aren't capable of beating the randomness of provably fair OR we(me) from start are using the wrong "inputs" to try and get a MA algorithm to find a pattern... I'm leaning on the first one.

mbithy commented 6 years ago

@osavigne cool stuff, I would like to see