Internet engine matches

[mtaktikos]

http://edgilbert.org/InternationalDraug ... ources.zip

New files and changed files (for Windows compile). Also add bitcount.cpp to the project.

Thank you Ed, but it didn't work. My compiler doesn't support in pthread std:mutex, here the error message:


Did you manage to get an exe file?

[Ed Gilbert]

Did you manage to get an exe file?

I added the Windows exe to the zip file. You may have to wait 15 minutes or so before getting it because the web server sometimes serves stale files from cache for a while. I also used static linking to libs so you don't need the VS2015 runtime dlls.

-- Ed

[mtaktikos]

Did you manage to get an exe file?

I added the Windows exe to the zip file. You may have to wait 15 minutes or so before getting it because the web server sometimes serves stale files from cache for a while. I also used static linking to libs so you don't need the VS2015 runtime dlls.

-- Ed

Thank you very much!
About every 30 minutes I make a download of the zip file, until now there is no exe packed herein,
but there is no haste, I will continue the downloads

Best regards,
Michael Taktikos

[mtaktikos]

Wonderful! It's there and it works on my old computer. Thx again

[Rein Halbersma]

Wonderful! It's there and it works on my old computer. Thx again

Did you try to install Visual Studio 2015 Community Edition? It's free and includes everything you need (including profile guided optimizations).

[mtaktikos]

Did you try to install Visual Studio 2015 Community Edition? It's free and includes everything you need (including profile guided optimizations).

No, I was not aware that Visual Studio is free, and thought, the MinGW compiler would be compatible with it. Until the prob with pthread_mutex occured, of course. Thanks for the suggestion, if there is in future again such a compatibility problem, then I will install Visual Studio to solve it

[Ed Gilbert]

I ran an engine match of 2-move start positions, kingsrow vs. scan, each using 6-piece dbs, 4 search threads, 120 moves in 3 minutes time controls. Results from kingsrow's perspective: +0, -33, =125, unknowns 0

Scan is really strong! Congrats to Fabien for creating a very powerful draughts program, and thanks for sharing the sources so the rest of us can try to understand what we need to do to move forward.

-- Ed

[MichelG]

I ran an engine match of 2-move start positions, kingsrow vs. scan, each using 6-piece dbs, 4 search threads, 120 moves in 3 minutes time controls. Results from kingsrow's perspective: +0, -33, =125, unknowns 0

Scan is really strong! Congrats to Fabien for creating a very powerful draughts program, and thanks for sharing the sources so the rest of us can try to understand what we need to do to move forward.

-- Ed

That is quite an incredible result! I had not expected such a big difference.

I just started a match against scan with similar settings. I'll have some results for dragon tomorrow.

Michel

[Rein Halbersma]

The number of transposition table entries is 2^tt-size. That means 2 raised to the power of tt-size. If tt-size is 6, the number of entries is 2^6 = 64. If tt-size is 24, then number of entries is 2^24 = 16.7 million. Each tt entry is 16 bytes, so 16.7 million entries is 256mb. I don't know how scan behaves with different tt sizes, but clearly 6 is much too small. I would guess a value of 23 would be good for matches of 3 to 5 minutes. It's probably not critical (but 6 is very bad).

-- Ed

tt-size = 30 would correspond to 64 times as much, about 16Gb of TT entries so no wonder that didn't work (unless the machine had 24 or 32 Gb or RAM).

[MichelG]

I have run a engine match between dragon v4.5 and scan with similar parameters as ed:
- 4 threads
- scan: 6 piece databases, 256 mb hash
- dragon: selected 8 piece databases, 6x6 eval patterns
- 3 minutes for 75 moves

I used 3 move ballots. The match is still running, but the first 158 games ended in
+0 -20 =138 (-43 elo)
from dragon's perspective. That confirms Ed's conclusion that scan is really strong Congratz to Fabien indeed.

I think it is amazing that scan did not loose a single game against dragon or kingsrow in 316 games!

Does anyone else have results against scan? And has someone played a scan vs scan match?

Michel

edit: games included

[Rein Halbersma]

I have run a engine match between dragon v4.5 and scan with similar parameters as ed:
- 4 threads
- scan: 6 piece databases, 256 mb hash
- dragon: selected 8 piece databases, 6x6 eval patterns
- 3 minutes for 75 moves

I used 3 move ballots. The match is still running, but the first 158 games ended in
+0 -20 =138 (-43 elo)
from dragon's perspective. That confirms Ed's conclusion that scan is really strong Congratz to Fabien indeed.

I think it is amazing that scan did not loose a single game against dragon or kingsrow in 316 games!

Does anyone else have results against scan? And has someone played a scan vs scan match?

Michel

Ed & Michel: could you both post the games to this list?

[Ed Gilbert]

Match games kingsrow vs scan.

[Fabien Letouzey]

I used 3 move ballots. The match is still running, but the first 158 games ended in
+0 -20 =138 (-43 elo)
from dragon's perspective. That confirms Ed's conclusion that scan is really strong Congratz to Fabien indeed.

Thanks Ed and Michel!

While I'm not complaining by any means, those results lead to a puzzle for me. Dragon Draughts already has the most accurate evaluation (through years of effort) and it's also gaining a lot from a breakthrough table that Scan doesn't have. I don't believe that my (very naive) ProbCut implementation can compensate for that if at all, so where are the missing Elo points coming from?

You don't have to answer right now BTW

Fabien.

[Rein Halbersma]

Thanks Ed and Michel!

While I'm not complaining by any means, those results lead to a puzzle for me. Dragon Draughts already has the most accurate evaluation (through years of effort) and it's also gaining a lot from a breakthrough table that Scan doesn't have. I don't believe that my (very naive) ProbCut implementation can compensate for that if at all, so where are the missing Elo points coming from?

You don't have to answer right now BTW

Fabien.

Wouldn't your bottom 4 patterns pick up potential breakthroughs? Btw, did you ever study particular pattern coefficients for plausibility?

Perhaps your qsearch is also sufficiently different from others to be a contributing factor?

[BertTuyt]

I don't believe that my (very naive) ProbCut implementation can compensate for that if at all, so where are the missing Elo points coming from?

First of all sorry that i didnt post Damage - Scan Match results, as my computers are still at home (according Rein I should activate remote access). But most likely I will visit Holland in some weeks from now, so i will take stronger hardware with me.

During the tournament, Fabien informed me that at one point Scan was searching deeper in comparison with Damage. This was remarkable, as Damage used 8 cores, at 4 GHz, and normally is searching much deeper as opponents (and sometimes throwing babies away with the pruning bathwater). Damage uses a combination of MCP, FHR and LMR.

So I guess that maybe also search depth is a major factor, maybe Ed and/or Michel observed the same.
Looking at the code, and I might have a better look, at least the probcut implementation looks interesting (although as Fabien explained the original statistiscs part is missing). Next to that I need to better understand the move sorting and LMR implementation. As Rein mentioned also the Q-search could be an explanation.
But for now, my bet would be the combination probcut, move sort and LMR.

Last but not least, I really thank Fabien, you have given our community a big boost by this result and by sharing all the internals, as can be observed in this forum...

Bert