Alaric 704 in CCRL 40/4

[Kirill Kryukov]

Alaric 704 recently played 1058 games in CCRL 40/4 time control. As a result it is now rated 2715 (+18 −18), which makes it #9 free single-CPU engine. It is always exciting when a new free engine gets into the top 10 so fast!

Free single-CPU rating list and cross-table (my version)
All results of Alaric 704.

Though, there is still uncertainty about ranking of engines surrounding it - several engines are very close to each other:

Code: Select all

Rank                 Engine                  ELO   +    -   Score  AvOp  Games
    7 Delfi 5.1                              2722  +16  -16  50.8%   -5.8  1329
    8 Pro Deo 1.2                            2717  +13  -13  44.7%  +38.6  2110
    9 Alaric 0704                            2715  +18  -18  49.3%   +4.8  1058
   10 Zappa 1.1 64-bit                       2714  +19  -19  47.6%  +17.2   961
   11 Slow Chess Blitz WV2.1                 2712  +10  -10  42.0%  +58.8  4181

Compared to Alaric 0702b, 704 is certain improvement with 30 points difference currently:
Comparison of Alaric versions.

LOS is only 96.2% because rating of Alaric 0702b is less certain due to smaller number of games. But look at the comparison table with common opponents!

Best,
Kirill

[Tony Thomas]

It is no surprise that Alaric got up there so quick. Peter is just doing the same thing that Daniel Shawul and Tord did in the past, that is to do a complete rewrite.

[Kirill Kryukov]

It is no surprise that Alaric got up there so quick. Peter is just doing the same thing that Daniel Shawul and Tord did in the past, that is to do a complete rewrite.

Well if it was as simple as rewriting your program to make top 10 then I guess we would see everyone doing it already. And top 10 is too little to include everyone anyway.

[Uri Blass]

It is no surprise that Alaric got up there so quick. Peter is just doing the same thing that Daniel Shawul and Tord did in the past, that is to do a complete rewrite.

Well if it was as simple as rewriting your program to make top 10 then I guess we would see everyone doing it already. And top 10 is too little to include everyone anyway.

Some comments:

1)I am not sure if alaric is top 10 engine

rating list of free engines is as follow:
1)Rybka 1.0 Beta 64-bit 2938
2)Toga II 1.3x4 HT80 2879
3)Spike 1.2 Turin 2828
4)Strelka 1.0b 32-bit 2826(not having enough games but clearly better than alaric)
5)Naum 2.0 32-bit 2800
6)Glaurung 1.2.1 32-bit 2757
7) Scorpio 1.9 (5-men-egbb) 2752 (1.91 has slightly better rating but not enough games)
8) Bright-0.1d 2741
9) Delfi 5.1 2722
10)Pro Deo 1.2 2717
11) Alaric 0704 2715
12) Zappa 1.1 64-bit 2714
13) Slow Chess Blitz WV2.1 2712

2)There are many commercial engines that could make top 10 by making their program free but chose not to do it.

1)Hiarcs
2)Loop
3)Fritz
4)Shredder
5)Zap
6)Junior
7)Tiger
8)Ktulu
9)Deep Sjeng
10)Chessmaster
11)Smarthink

3)in case of movei I suspect that I only need to implement hash correctly to get alaric level but when I tried to implement pruning based on hash it did not work for me and I dislike the subject of hash so I did not continue to try and prefered to improve movei by other means.

Movei does not use hash effectively and does not use it for pruning.
It has path dependent evaluation but it seems that this does not give it much(383 that is without path dependent evaluation has CCRL rating of 2631 that is less than 100 elo distance to the top 10)

The question is how much rating it is possible to achieve by effective use of hash relative to what I do.

Uri

[Kirill Kryukov]

Some comments:

1)I am not sure if alaric is top 10 engine

rating list of free engines is as follow:
1)Rybka 1.0 Beta 64-bit 2938
2)Toga II 1.3x4 HT80 2879
3)Spike 1.2 Turin 2828
4)Strelka 1.0b 32-bit 2826(not having enough games but clearly better than alaric)
5)Naum 2.0 32-bit 2800
6)Glaurung 1.2.1 32-bit 2757
7) Scorpio 1.9 (5-men-egbb) 2752 (1.91 has slightly better rating but not enough games)
8) Bright-0.1d 2741
9) Delfi 5.1 2722
10)Pro Deo 1.2 2717
11) Alaric 0704 2715
12) Zappa 1.1 64-bit 2714
13) Slow Chess Blitz WV2.1 2712

I included only stable versions released to public, that's why Toga II 1.3x4 and Bright are out of my list. Unless they are already released and available.

About Scorpio versions I honestly have no idea which one is stronger so I just test and list 1.91 because this way I can at least provide it with large enough number of games.

Also I still did not make up my mind about Strelka so let me skip it completely for now and get back when I have more data.

2)There are many commercial engines that could make top 10 by making their program free but chose not to do it.

This is true of course. But this is the whole point of constructing a free list - commercial engines can't be listed there. Yes there are many engines which could be in the list if they were free, but they are not free so they are not in the list. Simple. Also constructing separate free list should theoretically provide some more fun to the free engine authors like yourself because the competition is just with other hobby developers.

3)in case of movei I suspect that I only need to implement hash correctly to get alaric level but when I tried to implement pruning based on hash it did not work for me and I dislike the subject of hash so I did not continue to try and prefered to improve movei by other means.

Movei does not use hash effectively and does not use it for pruning.
It has path dependent evaluation but it seems that this does not give it much(383 that is without path dependent evaluation has CCRL rating of 2631 that is less than 100 elo distance to the top 10)

The question is how much rating it is possible to achieve by effective use of hash relative to what I do.

Uri

I wish you luck improving Movei. I enjoy your engine and I will enjoy it becoming of Alaric level or higher. I still use Movei 00.8.403 because as far as I know it is the latest release. When you will release the next version I will be happy to include it in some of my gauntlets (with default parameters only).

Best,
Kirill

[Peter Fendrich]

Thanks for testing Alaric!
I noticed that the difference beteween programs at this level is quite low. It is "only" 40 points up to Glaurung/Scorpio. Let's make a try...
I hope to release a new version before summer.

BTW. In case of WB, I hope that you disabled the resign option. It is severly buggy.

/Peter

[Tony Thomas]

It is no surprise that Alaric got up there so quick. Peter is just doing the same thing that Daniel Shawul and Tord did in the past, that is to do a complete rewrite.

Well if it was as simple as rewriting your program to make top 10 then I guess we would see everyone doing it already. And top 10 is too little to include everyone anyway.

Rewrite isnt a simple thing to do, most programmers dont have enough time to do a complete rewrite. Romichess has already reached the max level it can reach (current beta) using the algorithms it currently do, may be a more experienced programmer can squeeze out few more points that's all. Now Michael knows much more than he did when he first started Romichess, recently he got inspired by Tord and is planning on doing a complete rewrite.

[Kirill Kryukov]

Thanks for testing Alaric!
I noticed that the difference beteween programs at this level is quite low. It is "only" 40 points up to Glaurung/Scorpio. Let's make a try...
I hope to release a new version before summer.

I enjoyed testing it, as it was stronger than I expected. Good work! It is somewhat unstable though, it crashed about once every 100 games in my tests. (Although this is more stable than 0702b was).

I would not say that 40 points is a small difference, especially at that level, but I too hope you can do it. Given what you already achieved it should not be impossible.

BTW. In case of WB, I hope that you disabled the resign option. It is severly buggy.

/Peter

I tested it via UCI. Also, actually I always try to disable the "resign" options for all engines that I test. (Because some engines have mating bugs which I don't want to ignore).

Best,
Kirill