## On-line engine blitz tourney July

Discussion of chess software programming and technical issues.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
chrisw
Posts: 3352
Joined: Tue Apr 03, 2012 2:28 pm

### Re: On-line engine blitz tourney July

Rebel wrote:
Sat Jul 18, 2020 9:23 am
Seems Chris is working on the long awaited Chess System Tal III
Do you have a stats tool for determining (well, in quite restricted set up, find an engine with which yours is roughly equal at 50%) the average length of game wins? Or even the statistical spread?

Rebel
Posts: 5349
Joined: Thu Aug 18, 2011 10:04 am

### Re: On-line engine blitz tourney July

chrisw wrote:
Sat Jul 18, 2020 6:26 pm
Rebel wrote:
Sat Jul 18, 2020 9:23 am
Seems Chris is working on the long awaited Chess System Tal III
Do you have a stats tool for determining (well, in quite restricted set up, find an engine with which yours is roughly equal at 50%) the average length of game wins? Or even the statistical spread?
I looked, but don't have.
90% of coding is debugging, the other 10% is writing bugs.

chrisw
Posts: 3352
Joined: Tue Apr 03, 2012 2:28 pm

### Re: On-line engine blitz tourney July

Rebel wrote:
Sun Jul 19, 2020 6:42 am
chrisw wrote:
Sat Jul 18, 2020 6:26 pm
Rebel wrote:
Sat Jul 18, 2020 9:23 am
Seems Chris is working on the long awaited Chess System Tal III
Do you have a stats tool for determining (well, in quite restricted set up, find an engine with which yours is roughly equal at 50%) the average length of game wins? Or even the statistical spread?
I looked, but don't have.
Useful? Get a bunch of engines, say around 3000, that are mostly equal in matches against each other.
Play tournament, let’s say 10000 games each at bullet.
You get the usual roughly 30/30/30 split WDL. Same random book for all, same end of game decision rules.
Take the wins for each engine and compute the mean length of game. On the basis that short wins are probably connected to mating attacks, you should be able to list the engines on an exciting/boring scale.

chrisw
Posts: 3352
Joined: Tue Apr 03, 2012 2:28 pm

### Re: On-line engine blitz tourney July

Rebel wrote:
Sun Jul 19, 2020 6:42 am
chrisw wrote:
Sat Jul 18, 2020 6:26 pm
Rebel wrote:
Sat Jul 18, 2020 9:23 am
Seems Chris is working on the long awaited Chess System Tal III
Do you have a stats tool for determining (well, in quite restricted set up, find an engine with which yours is roughly equal at 50%) the average length of game wins? Or even the statistical spread?
I looked, but don't have.

Well, it kind of works. I grabbed two recent 10000 game test runs against a couple of equal-ish sparring partners, I think around 3000 CCRL Elo, won't say who thjey, because that is not the point, but one is in the Stockfish paradigm, and the other I don't know, save that Ed suggested it as an opponent because it plays interesting chess. Did the histograms of winning game length for mine paired with each of the other two.
Results are, (what I expected)
Mine, mean game length to win vs engine1: 97.6 ply
Engine1, mean game length to win vs mine: 111.1 ply

Mine, mean game length to win vs engine2: 100.0 ply
Engine2, mean game length to win vs mine: 111.8 ply

With a bit of refinement his might be a possible metric for play style. Fireworks engines should win their wins quickly. They'll probably go down in flames quickly too, of course.

Edit: median is useful. 50% of the wins are shorter than median value.
vs engine1, median is 91, opponent 104
vs engine2, 91, 108 (everything is half moves, btw)

Rebel
Posts: 5349
Joined: Thu Aug 18, 2011 10:04 am

### Re: On-line engine blitz tourney July

Nice way to measure playing style. Is the tool available? Else I must write it myself and I am so busy.

Correction, that should read as lazy.

90% of coding is debugging, the other 10% is writing bugs.

chrisw
Posts: 3352
Joined: Tue Apr 03, 2012 2:28 pm

### Re: On-line engine blitz tourney July

Rebel wrote:
Sun Jul 19, 2020 3:00 pm
Nice way to measure playing style. Is the tool available? Else I must write it myself and I am so busy.
Sure, what do you want it to do? Ideally, give it paired PGN batches for a bunch of similar engines. Each PGN batch off same opening book, same ply depth, same termination rule, same time control (fast).
Then process the EPDs, report median and mean winning game lengths and save some pretty histograms? If you want to prepare masses of paired games, I'll construct the results analyser for you.

Correction, that should read as lazy.