Game length between random players... Kai Laskos

[Zenmastur]

Could Kai or anyone that has a random player handy run a match between two random player and give me some very basic statistics (mean, and Standard deviation is all I need) about game length between them. It doesn't have to be huge number of games 100 to a 1,000 is plenty I would think. And if possible a random player against a recent version of stock fish at 10 plies or if that's not possible a 10Sec + 0.1sec increment.

Thanks in advance and regards,

Zenmaster

[pferd]

I did play 10,000 games between RuyRandom and itself and 10,000 games between RuyRandom and latest Stockfish at depth 10, but I am unable to generate the statistics. No opening book was used. Games can be grabbed here:

https://mega.nz/#!P40AxSLY!ZX1w0zmatn4A ... 3kPfxjuKps

[Ovyron]

I don't know how relevant this is, but this guy did extensive testing between random and many chess algorithms he came up with, including entities that played Stockfish a % of the time, and random the rest of the moves:

https://www.youtube.com/watch?v=DpXy041BIlA

Papers:

http://tom7.org/chess/

[Zenmastur]

I did play 10,000 games between RuyRandom and itself and 10,000 games between RuyRandom and latest Stockfish at depth 10, but I am unable to generate the statistics. No opening book was used. Games can be grabbed here:

https://mega.nz/#!P40AxSLY!ZX1w0zmatn4A ... 3kPfxjuKps

Thank you! I'm sure I can calculate the stats myself, I just didn't have a random player to use. Maybe they should put a pseudo-random player in every engine since they are quite handy for certain theoretical things.

Regards,

Zenmastur

[Guenther]

Could Kai or anyone that has a random player handy run a match between two random player and give me some very basic statistics (mean, and Standard deviation is all I need) about game length between them. It doesn't have to be huge number of games 100 to a 1,000 is plenty I would think. And if possible a random player against a recent version of stock fish at 10 plies or if that's not possible a 10Sec + 0.1sec increment.

Thanks in advance and regards,

Zenmaster

CCRL did play a lot of games with random player entities (for scaling the lowest end of the rankings), but probably not much games between themselves and also they most likely played some buggy opponents too with similar problems as random movers.

Note that your average move length depends a lot on adjudication rules (win and draw adjudication) then, because random movers ofc cannot win/draw easily (if at all), no matter how big or how drawish the eval appears.

Code: Select all

CCRL 40/4 Rating List — All engines

...

635     NEG 0.3d                549     +555    -550    47.6%   +32.5   39.6%   2789
636     Ram 2.0                 510     +593    -588    40.7%   +76.7   58.1%   2486
        Acqua 20160918          508     +596    -591    54.7%   -21.7   31.7%   3281
637     CPP1                    459     +645    -639    46.7%   +39.6   46.8%   1810
        Alouette 0.0.3 64-bit   430     +674    -669    42.0%   +76.0   51.0%   11512
638     POS 1.20                424     +680    -674    57.1%   -30.2   45.2%   3106
639     LaMoSca 0.10            368     +736    -730    47.6%   +33.0   86.5%   2477
640     ACE 0.1                 361     +743    -738    46.0%   +62.4   57.4%   3337
        Sapeli 1.02 RND 64-bit  283     +821    -816    30.8%   +116.7  59.7%   6248
        Brutus RND              259     +845    -839    30.4%   +150.8  56.7%   2218

[Branko Radovanovic]

One billion random games by Steven Edwards: http://www.talkchess.com/forum3/viewtopic.php?t=40193

[Guenther]

This might be interesting for you too: (graph with game length distribution per game termination type for rnd movers)
http://www.talkchess.com/forum3/viewtop ... er#p797409

[Laskos]

Could Kai or anyone that has a random player handy run a match between two random player and give me some very basic statistics (mean, and Standard deviation is all I need) about game length between them. It doesn't have to be huge number of games 100 to a 1,000 is plenty I would think. And if possible a random player against a recent version of stock fish at 10 plies or if that's not possible a 10Sec + 0.1sec increment.

Thanks in advance and regards,

Zenmaster

Just arrived from a trip. I used the PGN of random mover games supplied in this thread, I saw that there are no adjudications and everything seems fine. Here are some stats on the length:

Mean = 170.3 moves
Median = 176 moves
Standard Deviation = 55.6 moves

Here is the histogram on length of the games:

Random_01.jpg

[Zenmastur]

I did play 10,000 games between RuyRandom and itself and 10,000 games between RuyRandom and latest Stockfish at depth 10, but I am unable to generate the statistics. No opening book was used. Games can be grabbed here:

https://mega.nz/#!P40AxSLY!ZX1w0zmatn4A ... 3kPfxjuKps

One question: how many threads was SF running and what hardware it was running one.

I also found a copy of RuyRandom so now I can do it myself. I had no clue so many utils for random play are out there. I should have guessed since MCTS is so popular.

Thanks again,

Zenmastur

[Zenmastur]

One billion random games by Steven Edwards: http://www.talkchess.com/forum3/viewtopic.php?t=40193

A very nice thread!

Thank you!

Regards,

Zenmastur