Page 8 of 8

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 1:08 pm
by Ovyron
kramnik wrote:
Tue Oct 29, 2019 11:23 pm
But you should have given to it a more decent name at least XD... "Fat Fritz"... Whatever
Turns out Fat Fritz is a great name. I've been using chess engines since 2001. Since then all sort of engine names have appeared on my screen, Rebel, Ruffian, Rybka, Stockfish, Houdini... my mother has never been interested in any of them.

But what happened when I was reading these threads about Fat Fritz? She immediately saw the name and asked me what was it about. She wanted to know more about Fat Fritz :shock:

I think that it's a name designed to call people's attention, and it may get someone that hasn't been interested in them for 18 years to start talking about them.

So how good is a name that advertises itself?

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 2:12 pm
by Albert Silver
crem wrote:
Wed Oct 30, 2019 10:08 am
Albert Silver wrote:
Tue Oct 29, 2019 11:06 pm

I am also unsure about the argument of price always being proportionate to the work, but regardless will say that thousands of man hours have gone into this project from me alone
Few thousand of man hours is probably more than all human effort in LCZero project combined!
One full-time working year is typically 2000 hours.
Indeed, so now you have an idea too. Our skillsets are obviously different though. I use my knowledge of chess extensively, studying how it plays, changes in its behaviour, what is causing a regression or progression, etc. For every 5k steps (and remember this is for multiple runs) at least 1500-2500 test games will be played, with a good portion observed and studied by me. And yes, I am often up until crazy hours, weekends too, with this being the first thing I check when I wake.

To say that I know Leela and Fat Fritz would be a gross understatement. In the games that I have run tests, which use the same openings, I know almost every permutation they will take in the openings as well as what will likely happen when they start going in one direction or another, and I know which openings and structures they handle better or not, and details of which endings they do better or not in and which ones they are likely to fail in as well. In fact in the final of the TCEC Cup now when I saw the position in game 8 after move 30 I predicted Leela was going to lose in spite of both engines being comfortable that the position was completely equal.

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 2:20 pm
by Ovyron
Does Fat Fritz have anything to do with Deus X?

What's in a name

Posted: Wed Oct 30, 2019 2:37 pm
by Albert Silver
Ovyron wrote:
Wed Oct 30, 2019 2:20 pm
Does Fat Fritz have anything to do with Deus X?
Yes, everything or nothing really. That was the public name I gave for it when it was playing online for tournaments or just test games. But it has been a very relaxed one that has covered multiple neural networks to be completely honest. It has been used for no fewer than four completely different neural networks. Two small and two large ones of which Fat Fritz is the latest large one. However, internally not one has actually been called that believe it or not.

My first small one was dubbed MegaCorr, in reference to the source material, with my subsequent rebuild using some training refinements called MegaCorr2. When I decided for more ambitious ideas and a 20x256 it became BigKahuna, with multiple runs dubbed BigKahuna2, 3, etc. This is what played in Leiden. Fat Fritz internally was called AZK, with the one being published called AZK7. AZK meant AlphaZeroKiller with 7 being the 7th complete restart (I do not include partial rollbacks in my numbering scheme).

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 2:42 pm
by Ovyron
Thanks, so what was the inspiration for naming it "Fat Fritz"? The other names you mention give a feeling of "this is very strong", while Fat Fritz gives a vibe of "having fun", in my opinion.

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 2:50 pm
by Albert Silver
Ovyron wrote:
Wed Oct 30, 2019 2:42 pm
Thanks, so what was the inspiration for naming it "Fat Fritz"? The other names you mention give a feeling of "this is very strong", while Fat Fritz gives a vibe of "having fun", in my opinion.
Chessbase gets all credit for that. They feared people might balk at Deus X and claim it had religious connotations. For the record, the name came from Deus Ex Machina, 'God (the chess god) coming from the machine'. Deus Ex being a very famous game IP, I just removed the E.

They consulted a branding company, the same that came up with Fritz I believe, and came back with Fat Fritz. They told me to trust them, that the guys knew their stuff, and I said sure, no problem.

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 3:16 pm
by smatovic
Seems since the AlphaZero impact we have new roles in computer chess, beside
the classical programmer/advisor roles, we have additionally the nn-cpu/gpu low
level coder and nn-trainer/advisor, or alike.

--
Srdja

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 3:29 pm
by Ovyron
In that case I'll leave this here and run away :D

Image

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 3:50 pm
by crem
Albert Silver wrote:
Wed Oct 30, 2019 2:12 pm
crem wrote:
Wed Oct 30, 2019 10:08 am
Albert Silver wrote:
Tue Oct 29, 2019 11:06 pm

I am also unsure about the argument of price always being proportionate to the work, but regardless will say that thousands of man hours have gone into this project from me alone
Few thousand of man hours is probably more than all human effort in LCZero project combined!
One full-time working year is typically 2000 hours.
Indeed, so now you have an idea too. Our skillsets are obviously different though. I use my knowledge of chess extensively, studying how it plays, changes in its behaviour, what is causing a regression or progression, etc. For every 5k steps (and remember this is for multiple runs) at least 1500-2500 test games will be played, with a good portion observed and studied by me. And yes, I am often up until crazy hours, weekends too, with this being the first thing I check when I wake.
Just to clarify, the message I tried to convey in my response is that I don't consider that fully believable.
But I agree I may be wrong.

Re: "Fat fritz" stronger than Stockfish ?

Posted: Wed Oct 30, 2019 4:18 pm
by Albert Silver
crem wrote:
Wed Oct 30, 2019 3:50 pm
Albert Silver wrote:
Wed Oct 30, 2019 2:12 pm
crem wrote:
Wed Oct 30, 2019 10:08 am
Albert Silver wrote:
Tue Oct 29, 2019 11:06 pm

I am also unsure about the argument of price always being proportionate to the work, but regardless will say that thousands of man hours have gone into this project from me alone
Few thousand of man hours is probably more than all human effort in LCZero project combined!
One full-time working year is typically 2000 hours.
Indeed, so now you have an idea too. Our skillsets are obviously different though. I use my knowledge of chess extensively, studying how it plays, changes in its behaviour, what is causing a regression or progression, etc. For every 5k steps (and remember this is for multiple runs) at least 1500-2500 test games will be played, with a good portion observed and studied by me. And yes, I am often up until crazy hours, weekends too, with this being the first thing I check when I wake.
Just to clarify, the message I tried to convey in my response is that I don't consider that fully believable.
But I agree I may be wrong.
I had understood perfectly, which is why I elaborated in greater detail.