Cool, thanks Graham.
This version is a bit experimental. It has pretty drastic futility pruning - 50cp*depth - but that seems to work out best over a large number of games - with some howlers thrown in for good measure. Ugly chess but until the eval fn is more mature it seems the best way to get results. Totally non-intuitive to me, though thinking about it, I can kindasortaish see why it works in a PVS context - essentially a hypothesis testing context.
Silvian, I added the UCI harness to Gary's GarboChessJS and I may have done it incorrectly. If you test it, I'd be interested to hear what you think. I saw some wild eval figures being reported as I recall but I don't know if that's me or not...
Lozza website updated to 1.15
Moderator: Ras
-
op12no2
- Posts: 561
- Joined: Tue Feb 04, 2014 12:25 pm
- Location: Gower, Wales
- Full name: Colin Jenkins
-
Sylwy
- Posts: 5289
- Joined: Fri Apr 21, 2006 4:19 pm
- Location: IAȘI - the historical capital of MOLDOVA
- Full name: Silvian Rucsandescu
Re: Very nice !
To be honest - the play of GarboChess was strangely at the first attempt: incorrect moves and a quick defeat. Something is not OK, for sure !op12no2 wrote:
Silvian, I added the UCI harness to Gary's GarboChessJS and I may have done it incorrectly. If you test it, I'd be interested to hear what you think. I saw some wild eval figures being reported as I recall but I don't know if that's me or not...
SilvianR
-
op12no2
- Posts: 561
- Joined: Tue Feb 04, 2014 12:25 pm
- Location: Gower, Wales
- Full name: Colin Jenkins
Re: Very nice !
Ah OK thanks. I've probably screwed up. Gary, if you see this, can you have a quick look at:-Sylwy wrote:
To be honest - the play of GarboChess was strangely at the first attempt: incorrect moves and a quick defeat. Something is not OK, for sure !
SilvianR
http://op12no2.me/toys/lozza/garbo.js
The bit at the end where I invoke your primitives from my UCI stuff.