Feature request: have more information returned with the action chosen about mcts HOT 7 CLOSED

I add a parameter needNodeValue=False in search. If it is set to True, search will return the best action and NodeValue, which is the best node's average value. If it is kept False, as default, search will keep its original behavior, making it backward compatible. I will push from my fork to this repo as soon as I finish the tests.

from mcts.

I am going to add this feature very soon. I am writing a card AI where I need this feature.

from mcts.

I have added this feature in my fork and pushed my branch to pull request #13.

from mcts.

Let totalReward / numVisit be avg_reward_per_visit
Do you think that best child avg_reward_per_visit / sum of all childs avg_reward_per_visit
can give some sort of confidence in the choice ?
Thanks

from mcts.

Hi, I hope my comment is clear, or maybe a bit more explanation is needed?

from mcts.

Hi Dario,

Thanks for the request, aye, it's clear. I'm afraid I don't have the bandwidth to work on this myself right now though, but feel free to make a pull request and I'll take a look :)

from mcts.

Yeah I can do that, no problem, but do you think the logic I've mentioned makes sense? it is a bit of a hack, and it is not backed by any theory :)

from mcts.