Smoothing SF progress of nextchessmove

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Vinvin
Posts: 4418
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Smoothing SF progress of nextchessmove

Post by Vinvin » Thu Jan 16, 2020 9:32 am

I like to look at https://nextchessmove.com/dev-builds where you can find 20000 test games of the latest SF builds.
But the curve still irregular, so I made a smoothing in Excel : for each version I took the average of the 10 previous ratings. (+ a linear regression inside Excel on the dash line).
The curve start with Stockfish 10 ( 20181129-1445).

Image

Nice to view !

Jouni
Posts: 2055
Joined: Wed Mar 08, 2006 7:15 pm

Re: Smoothing SF progress of nextchessmove

Post by Jouni » Thu Jan 16, 2020 12:57 pm

Definitely average or more games is needed there! It's always funny when "Update Readme.md" gaves +5 ELO :D !
Jouni

Alayan
Posts: 47
Joined: Tue Nov 19, 2019 7:48 pm
Full name: Alayan Feh

Re: Smoothing SF progress of nextchessmove

Post by Alayan » Thu Jan 16, 2020 2:49 pm

The 100 elo regression patch in March is marring the curve. Otherwise, the averaging makes it much nicer. :)

Vinvin
Posts: 4418
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Smoothing SF progress of nextchessmove

Post by Vinvin » Tue Jan 21, 2020 9:29 am

Updated :
From SF 10 (20181129-1445) to SF 11 (20200118-0044)
Buggy version "20190405-1837" is removed to make a clean graph.

Image

Dann Corbit
Posts: 10242
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Smoothing SF progress of nextchessmove

Post by Dann Corbit » Tue Jan 21, 2020 5:17 pm

I guess that even with the bad version the correlation coefficient is very close to 1, which means the progression is very nearly linear.
Since Elo is exponential, this is excellent evidence of constantly having exponential progress in the strength of Stockfish.
That is really pretty amazing.
Too bad our salaries don't do that.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Raphexon
Posts: 138
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: Smoothing SF progress of nextchessmove

Post by Raphexon » Tue Jan 21, 2020 5:19 pm

Dann Corbit wrote:
Tue Jan 21, 2020 5:17 pm
I guess that even with the bad version the correlation coefficient is very close to 1, which means the progression is very nearly linear.
Since Elo is exponential, this is excellent evidence of constantly exponential progress in the strength of Stockfish.
That is really pretty amazing.
Too bad our salaries don't do that.
Year by year few % growth is exponential.
So your salary probably does show exponential growth.
But prices also go up exponentially.

Dann Corbit
Posts: 10242
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Smoothing SF progress of nextchessmove

Post by Dann Corbit » Tue Jan 21, 2020 5:29 pm

I'm paid pretty well, but a graph of my salary over time won't look like the SF Elo graph.
Mine would look more like a square root function, where it rapidly rose after college for a few years then rose more and more slowly.
I do agree that even 0.05% interest growth per year would be exponential. But the constant in the exponent not nearly so desirable as the constant in the SF graph
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
Ovyron
Posts: 3065
Joined: Tue Jul 03, 2007 2:30 am

Re: Smoothing SF progress of nextchessmove

Post by Ovyron » Wed Jan 22, 2020 3:04 am

Raphexon wrote:
Tue Jan 21, 2020 5:19 pm
But prices also go up exponentially.
This is by design, if everyone's salary goes up, there's more demand, and not enough supply, so to fix that you increase the price of the supply, and we're back to how we started. The problem is the accumulation of wealth by people that already had wealth, and since their wealth increases faster than people's salary, prices seem to go up faster than salaries.

Post Reply