not approaching group to die

[dsmic]

A game from cgos, where oakfoam does not kill because it does not play a9 or a7?!

[Francois van Niekerk]

This requires the addition of the "approach moves" heuristic, which can be seen in Section 3.3 of Chaslot et. al. "Adding expert knowledge and exploration in Monte-Carlo Tree Search"

[dsmic]

This rule would be nice, but I think there is a bug explicitly forbidding this moves, as they are not found at all. At least at the top of the tree search it should be found as far as I understand it?!

[dsmic]

I check this out, the version on my Laptop finds the winning move 84, I have to check if I made a mistake with the version on the MAC, sorry for the noise:(

[dsmic]

OK, we will get a lot of fun optimizing the parameters:) The parameters I use at the moment are stronger in average but they do often not find the move 84:(

Hopefully we get a feeling how to do the optimization, if we are at this point. Seems to be a little more difficult than just to find the optimum value. I already saw, that the parameters are a little weaker on cgos than my last ones,

but I think you can close this bug. We have to keep the parameter problem in mind!

[Francois van Niekerk]

Issue #14 was marked as a duplicate of this issue.

[Francois van Niekerk]

Changeset 8f4184d8362f adds the approach move heuristic. Enabling this with the playout_random_approach_p parameter fixes this bug.