Search found 295 matches

by Fabien Letouzey
Thu May 28, 2020 08:49
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

Especially the Kata-go paper link is interesting to read: it emphasizes using human ingenuity to come up with better NN primitives to accelerate learning. It should be right up your ally to see how this can be applied to draughts. The way I see it, DeepMind's MO has always been (so far?) to not opt...
by Fabien Letouzey
Wed May 27, 2020 08:46
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

There are no high-performance AlphaZero draughts programs yet. Last year, there were two candidate projects I was associated with (mostly at the advice level): one by Facebook, and one by the author of Galvanise-zero (G0) that I already mentioned on this forum. Unfortunately they were both terminat...
by Fabien Letouzey
Tue May 26, 2020 13:29
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

So it is up to the TD, in our case that means Krzysztof. Not sure I still like to compete myself. I have a magic trick to cheer people up, I just show them a part of my life. Years ago, Sidiki and I (independently) told everyone that Edeon Sport was using Scan 2's eval. What did Krzysztof do? He pa...
by Fabien Letouzey
Tue May 26, 2020 08:22
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

... I am surprised that this would be allowed, i.e. to call the result your original work. But if the community and in particular the lead programmer think it is fair then who am I to complain? Since the topic of fairness was brought up I wanted to include ML in the discussion. So to make it concre...
by Fabien Letouzey
Mon May 25, 2020 13:10
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

Just to establish some rules, would you say it is allowed if I take Scan's eval, change some features but still use your eval file, put that in my own search and use that to play my first million training games? I consider that using ML at all, especially when home-written, is enough to justify som...
by Fabien Letouzey
Sat May 23, 2020 20:05
Forum: Draughts, Computer, Internet
Topic: Flits self-learning mode
Replies: 83
Views: 4478

Re: Flits self-learning mode

For me, the key is how much Ed's code is actually contributing to Bert's evaluation weights. From the following quote from Bert (that I have no reason to doubt), it's focusing on the maths part. I feel that it's fine in this case, and this also goes in the direction Rein is advocating, that many 3rd...
by Fabien Letouzey
Sat Apr 04, 2020 14:24
Forum: Draughts, Computer, Internet
Topic: Computer Draughts Rating Lists
Replies: 4
Views: 831

Re: Computer Draughts Rating Lists

For instance : how do the current best engine versions (of the strongest engines) compare to Scan 2.0 ? I first thought your post was focused on human/computer comparisons (which I don't think exist), and missed that last line. Scan 3 was a big jump in 2017, so programs are still making progress. I...
by Fabien Letouzey
Sat Mar 28, 2020 09:33
Forum: Draughts, Computer, Internet
Topic: Damage 15.3
Replies: 139
Views: 19369

Re: Damage 15.3

Also find out if I can improved the speed for the AMD (so replace PEXT). Scan doesn't use PEXT. Not only it's crippled on AMD, as you just noticed, but it won't scale in a world that will increasingly use ARM (which has bit reverse BTW). Here is an explanation that I sent to Rein a few years ago; h...
by Fabien Letouzey
Tue Feb 25, 2020 08:40
Forum: Draughts, Computer, Internet
Topic: Damage 15.3
Replies: 139
Views: 19369

Re: Damage 15.3

Do I understand correctly Bert that your training games were played with the handcrafted Damage evaluation function? I used two batches, the first with material-only evaluation and the second with the evaluation trained from the first batch, so bootstrapping from zero knowledge. What I usually do i...
by Fabien Letouzey
Sun Feb 23, 2020 12:04
Forum: Draughts, Computer, Internet
Topic: Damage 15.3
Replies: 139
Views: 19369

Re: Damage 15.3

Hi Bert, An interesting research question is , if there is further improvement when one further increases the win/lose ratio (for example do not include any draws), or if there is an optimum. I remember trying skipping draws, calling the label target "Wilo" (win/loss), in hope that this component wo...
by Fabien Letouzey
Mon Sep 09, 2019 12:57
Forum: Draughts, Computer, Internet
Topic: Unofficial World Championships Of Computer Programs Rapid 2019
Replies: 71
Views: 20010

Re: Unofficial World Championships Of Computer Programs Rapid 2019

Ed Gilbert wrote:
Mon Sep 09, 2019 12:09
What is the end position of that game?
I don't have it; the game has been replaced.
by Fabien Letouzey
Mon Sep 09, 2019 08:56
Forum: Draughts, Computer, Internet
Topic: Unofficial World Championships Of Computer Programs Rapid 2019
Replies: 71
Views: 20010

Re: Unofficial World Championships Of Computer Programs Rapid 2019

While I was also surprised by the result, Krzysztof might be correct in his decision. Who knows, maybe 5 kings are not enough against 2 (!); I didn't know that. Scan's score alone cannot be trusted in the specific case of endgames with many kings. The reason is that Scan has learned from improved dr...
by Fabien Letouzey
Wed Jul 24, 2019 18:41
Forum: Draughts, Computer, Internet
Topic: "pretend solving" 10x10 variants on a 8x8 board
Replies: 6
Views: 2435

Re: "pretend solving" 10x10 variants on a 8x8 board

How big are your books for BT and LD? With 8 piece dbs you might see to the end of the game. They are tiny in comparison: about 5k for BT and 20k for LD (internal nodes). But I think that it can be misleading: you can compensate for a small book by having a better leaf "evaluation". Last week I act...
by Fabien Letouzey
Wed Jul 24, 2019 17:36
Forum: Draughts, Computer, Internet
Topic: "pretend solving" 10x10 variants on a 8x8 board
Replies: 6
Views: 2435

Re: "pretend solving" 10x10 variants on a 8x8 board

(losing draughts only) Usually a single computation step, expanding a leaf (whose position is displayed), takes about a minute. And, while I was watching the "alternative move" search (which excludes 24-20), many steps in succession only took a second, with long lines (about 40 plies, and normal sea...
by Fabien Letouzey
Wed Jul 24, 2019 09:47
Forum: Draughts, Computer, Internet
Topic: "pretend solving" 10x10 variants on a 8x8 board
Replies: 6
Views: 2435

Re: "pretend solving" 10x10 variants on a 8x8 board

Interesting stuff! Can you post the PVs for both games? Yes. BT: 22-18 11-15 18x11 8x15 21-17 7-11 17-13 9-14 23-19 11-16 LD: 24-20 9-13 28-24 13-17 21x14 10x17 22x13 5-9 20-16 12x28 25-22 8-12 22-18 11-16 18-14 9x18 23x14 6-10 27-23 10x17 13x22 1-5 I stopped showing them, because I think that the ...