World Draughts Forum

Reinforcement learning for International Draughts is a fun experiment. The biggest problem with this is that even games with a random factor included eventually all end in a draw, meaning the network no longer learns anything from them. An option to break the dead draw can be adding a few extra ...

After each iteration, a ballot match of 990 games is played to determine whether the new network is better than the old one. After 50 iterations, progress stalled because the learning rate decay was not functioning. Now that this has been fixed, it is progressing again.

To be honest, I believe that the current top 10 International draughts engines will never lose a match against the world champion. In fact, there is a very high chance these engines would win, even though the game is notoriously drawish. This shows that a computer can reach a level through just a ...

Bert,

I didn't expect it to play a reasonable game after just 5 iterations, which amounted to 24,750 games at 20k nodes per move. Naturally, the process is slowing down now because it doubles the nodes per move whenever the network's performance plateaus.

It has currently completed 80 iterations ...

An experiment I've wanted to try for a while now is training an international draughts network using reinforcement learning. Since it's a one-off project and coding everything from scratch takes quite a bit of time, I just never got around to it. However, with the help of Claude, it’s now actually ...

A review of the code by Claude Fable 5.1 and implementing the recommendations by ChatGPT Astra led to a factor two speedup. Rather than saying that Claude Fable 5.1 is better than ChatGPT Astra I think it shows it can be worthwhile to pitch models against each other. A review of the ChatGPT Sol ...

Joost (and Gijsbert might also find this useful), I was looking through some of my draughts doc files and discovered that I had created a stats file with all the counts for the 2 through 7 pieces DTW db. Joost, I'm sorry that I didn't respond earlier when you were asking for DTW counts. I had ...

not 100% sure, can you verify:

Format: configuration all_win all_draw all_lose, noncapture_win noncapture_draw noncapture_lose

1221 678795283 1395118040 7606227 86768707 568453887 260



Bert, thanks! I will compare these numbers with what I've got here.

Edit:

The numbers compare nicely ...

The per-thread egdb cache is an interesting idea. I didn't realize you could read a small block from an SSD in 5 us. The last time I benchmarked this stuff was quite a while ago. I have a single egdb cache and use locks to protect critical regions.



I devised a benchmark using 40 million ...

Ed, I heard from Bert that he doesn’t include positions with a 'non-side' capture in his database either. However, I’ve decided to include them because Ares frequently encounters positions where a 'non-side' capture needs to be evaluated. On a modern PC, it matters very little whether a database ...

Ed Gilbert wrote: Mon Aug 31, 2026 13:57 Joost, I have this list of WLD counts for 2 through 7 pieces in my db. IIRC the counts exclude direct captures.

Ed, thanks!

This helps me to complete the comparison.

When compressing I exclude direct captures too, because of this the size of the 7P database has shrunk to 30 GB.

Joost

I noticed that on the Dragon Database Statistics website the counts for the sub-database 1221 are missing, are these counts (capture/non capture) by any means known (including the depth-counts for DTW)?

In just one day of work with Codex the GUI is already capable of playing a match between two HUB engines. Ares does not support HUB, probably it will never support HUB, but I will add it with a proprietary protocol.


Why would you add a proprietary protocol instead of HUB when the latter is ...

I decided to take the easy route too. Although I already programmed a large part of a GUI with MFC it has two drawbacks, it looks a bit old-fashioned, and it's impossible to port it to Linux. So I decided to use QT as the primary toolbox (another option would have been wxWidgets).

In just one day ...

Out of curiosity I let Codex design a neural network implementation with exactly the same parameters that the Ares network uses.

First of all, the code generated by Codex contains a lot of unnecessary crap that will only hurt performance. Secondly, it was not able to generate code that comes close ...