search

SanchoGGP / ggp-base

The General Game Playing Base Package
8 stars 4 forks source link

Terrible checkers move #17

Open arr28 opened 10 years ago

arr28 commented 10 years ago

Move 43 (red) in http://www.ggp.org/view/tiltyard/matches/227994804ffdf7e0f4192646d04c860e26a12c13/ was totally daft. Need to add a regression case and then investigate.

Might be struggling from time. Game had 15s moves. I don't know how much we currently leave for network latency.

SteveDraper commented 10 years ago

We leave 2 seconds for latency I think (might be 3).

Also (and more to the point perhaps here) in English Draughts captures are forced, so it';s two moves earlier (when the king moves off the back rank) that is the bad move, but obviously that's harder to read though.

I suspect what we need here is a forced drill-down on forced sequences (i.e. - if you generate a node with a single response ALWAYS expand it). This would be great test case for that though

Steve

On Sat, Mar 15, 2014 at 8:26 AM, Andrew Rose notifications@github.comwrote:

Move 43 (red) in http://www.ggp.org/view/tiltyard/matches/227994804ffdf7e0f4192646d04c860e26a12c13/was totally daft. Need to add a regression case and then investigate.

Might be struggling from time. Game had 15s moves. I don't know how much we currently leave for network latency.

Reply to this email directly or view it on GitHubhttps://github.com/SteveDraper/ggp-base/issues/17 .

arr28 commented 10 years ago

As we discussed, forced drill-down wouldn't really have helped. Any rollout through a node with a single move is bound to go through the single move. (It does mean that you start keeping stats about the following moves sooner - rather than doing a completely random playout through them. However, in checkers, that's typically only a single extra select or sometimes two.)