Hotspots

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

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
lantonov
Posts: 216
Joined: Sun Apr 13, 2014 3:19 pm

Re: A plea to someone

Post by lantonov » Tue Apr 07, 2015 9:02 pm

Lyudmil Tsvetkov wrote:Sorry, SF guys, I just can not resist seeing Joerg pushing a piece values patch on the framework, and me not being able to do so... :(

I had the opportunity to suggest in the past a change for a single piece value, the pawn, but never for all pieces at the same time.

It is simply irresistible, I am sure my values are going to pass after 500 games. :)

Below are my suggested values:


PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 817, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2514, QueenValueEg = 2547,

MidgameLimit = 15581, EndgameLimit = 3998


Lyudmil, you possibly told you could push one other patch in the future, could I ask you for the last time that we try this one, just for a single time, maybe when there are fewer tests going?
I would be very grateful, it is simply irresistible.

Or maybe someone else could help me with pushing the above patch, I am sure Joerg will not mind to have one other version tested.
Sorry for being late, I just saw this thread. Sure, I'll try to push this one. I think the best moment will be after Joerg's test passes or fails. In both cases, your piece values will be a good test for the SPSA sanity.

BTW, I pushed the SPSA test for the rook psqt. The first time the values were too many (64+64), so I rescheduled it with only mg values and symmetric about vertical (32 values to tune).

User avatar
lantonov
Posts: 216
Joined: Sun Apr 13, 2014 3:19 pm

Re: A plea to someone

Post by lantonov » Tue Apr 07, 2015 9:15 pm

Lyudmil Tsvetkov wrote:Sorry, SF guys, I just can not resist seeing Joerg pushing a piece values patch on the framework, and me not being able to do so... :(

I had the opportunity to suggest in the past a change for a single piece value, the pawn, but never for all pieces at the same time.

It is simply irresistible, I am sure my values are going to pass after 500 games. :)

Below are my suggested values:


PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 817, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2514, QueenValueEg = 2547,

MidgameLimit = 15581, EndgameLimit = 3998

Most of the values in the SPSA now running converge not very far from yours:

PawnValueMg = 192, PawnValueEg = 262,
KnightValueMg = 821, KnightValueEg = 851,
BishopValueMg = 837, BishopValueEg = 864,
RookValueMg = 1267, RookValueEg = 1291,
QueenValueMg = 2514, QueenValueEg = 2546,

MidgameLimit = 15606, EndgameLimit = 3992

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: A plea to someone

Post by Lyudmil Tsvetkov » Tue Apr 07, 2015 10:24 pm

lantonov wrote:
Lyudmil Tsvetkov wrote:Sorry, SF guys, I just can not resist seeing Joerg pushing a piece values patch on the framework, and me not being able to do so... :(

I had the opportunity to suggest in the past a change for a single piece value, the pawn, but never for all pieces at the same time.

It is simply irresistible, I am sure my values are going to pass after 500 games. :)

Below are my suggested values:


PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 817, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2514, QueenValueEg = 2547,

MidgameLimit = 15581, EndgameLimit = 3998

Most of the values in the SPSA now running converge not very far from yours:

PawnValueMg = 192, PawnValueEg = 262,
KnightValueMg = 821, KnightValueEg = 851,
BishopValueMg = 837, BishopValueEg = 864,
RookValueMg = 1267, RookValueEg = 1291,
QueenValueMg = 2514, QueenValueEg = 2546,

MidgameLimit = 15606, EndgameLimit = 3992
Thanks Lyudmil!

I really hope you can push that one, maybe after Joerg's patch ends its run.

Close, but there are also significant differences: Joerg wants to give higher value in the eg to the knight, higher value to the pawn, but my version is quite the opposite.

You never know what happens with those imbalances - so many of them to consider.

Thanks again.

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: A plea to someone

Post by Lyudmil Tsvetkov » Wed Apr 08, 2015 5:18 am

Unfortunately, Joerg's patch scored just yellow.

Maybe we can push the alternative version now, until Joerg further improves on his values, I mean until Joerg's SPSA further improves...

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

10 king fianchettoe positions

Post by Lyudmil Tsvetkov » Wed Apr 08, 2015 6:02 am

For those who would like to check their engine's ability to choose appropriate kingside fianchettoes:

[d]rnbqkb1r/pppppppp/5n2/8/2PP4/8/PP2PPPP/RNBQKBNR b KQkq - 0 1
best move here? undoubtedly g6, followed by Bg7

I wonder how long would it take your engine to discard weaker moves like e6, b6, etc., and stick with g6 instead.

[d]rnbqk2r/ppp1ppbp/3p1np1/8/2PPP3/2N5/PP3PPP/R1BQKBNR b KQkq - 0 1
best move here?
well, undoubtedly g3, followed by Bg2, unless you want to play f3 and castle queenside
So, the best antidote to the black's KID is a white KID!, with a fianchettoed bishop. If you do not believe me, look at top GM games.

I wonder how long for your engine to pick up g3 above.

[d]r1bq1rk1/pppp1ppp/2n2n2/4p3/1bP1P3/2NP1N2/PP3PPP/R1BQKB1R b KQ - 0 1
best move for white above?

g3, followed by Bg2

[d]rnbqkbnr/ppp1pppp/8/8/2Pp4/5N2/PP1PPPPP/RNBQKB1R b KQkq - 0 1
in the Reti, again g3 and Bg2 seems the most appropriate move

I hope there are not any engines around that would instead play e3 and exchange on d4 first.

[d]rnbqkb1r/pp2pppp/2p2n2/3p4/5B2/2PP1N2/PP2PPPP/RN1QKB1R b KQkq - 0 1
best move for black above?

again, g6, followed by Bg7

[d]r1bqkbnr/pp1ppppp/2n5/1Bp5/4P3/5N2/PPPP1PPP/RNBQK2R b KQkq - 0 1
best move in the Rossolimo variation of the Sicilian Defence?

again, g6, followed by Bg7 (but that is true also of many other Sicilian lines)

[d]r1bqkbnr/pp1ppppp/2n5/2p5/2P5/2N5/PP1PPPPP/R1BQKBNR b KQkq - 0 1
English Symmetrical

g3 is the best move for white, followed by Bg2; same for black, g7, followed by Bg7

[d]r1bqkb1r/ppp1pppp/2n2n2/3p4/3P4/2N2N2/PPP1PPPP/R1BQKB1R b KQkq - 0 1
4 Knights

I doubt there is a better move than g3, followed by Bg2

[d]rnbqkb1r/ppp1pppp/5n2/3p4/3P4/2N2N2/PPP1PPPP/R1BQKB1R b KQkq - 0 1
Vienna Game

best attempt by black to exploit the awkward first move by white is to play g6, followed by Bg7

[d]rnbqkb1r/ppppp1pp/5n2/5p2/3P4/5N2/PPP1PPPP/RNBQKB1R b KQkq - 0 1

Dutch

possible best move here is g3, followed by Bg2

Etc., etc.

So you can play kingside fianchettoes in almost all openings and practically every second or third game.

I wonder in how many of the above positions your engine would like to fianchettoe its bishop kingside?
What is the percentage of games in your engine database, where the engine plays g3, Bg2 in the opening?

If lower than 30%, you should think very seriously what is wrong with your engine. (of course, this concerns only aspiring engines)

Probably, taking the above diagrams and other similar early opening positions, you might decide what you should do, what eval and search terms to change in order that your engine starts choosing kingside fianchettoes more often.

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: 10 king fianchettoe positions

Post by Lyudmil Tsvetkov » Wed Apr 08, 2015 6:12 am

I wonder if your engine would need more time/deeper search, in order to recognise the strength of such moves.

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: 10 king fianchettoe positions

Post by Lyudmil Tsvetkov » Wed Apr 08, 2015 6:35 am

[d]rnbqkb1r/pppppppp/5n2/8/2PP4/8/PP2PPPP/RNBQKBNR b KQkq - 0 1

OK, give this one a longer think.

Why is e6 considered by most engines, including SF and supposedly Komodo, the better move, when the Nimzo-Indian in most of its lines is actually quite dubious for black?

Why is g7 almost never considered by even top engines, when a king with a fianchettoed bishop beside it is much safer than otherwise, and this bishop has very nice control of the center, attacking enemy pieces along the long diagonal at the same time?

You do not need to go any further than that: a bishop on d6 or c5 just has somewhat better psqt and mobility, that are temporary factors, while attacking the enemy king shelter at the same time, while a bishop on g7 improves the black king safety almost permanently and controls the center and attacks enemy pieces along the long diagonal almost permanently.

So, Bd6/Bc5/Bb4 provide short-term advantages while Bg7 provides long-term advantages. I think, if anything else does not work, a nice king safety bonus for Bg2/Bg7 should be the right approach.

User avatar
lantonov
Posts: 216
Joined: Sun Apr 13, 2014 3:19 pm

Re: A plea to someone

Post by lantonov » Wed Apr 08, 2015 6:41 am

Lyudmil Tsvetkov wrote:Unfortunately, Joerg's patch scored just yellow.

Maybe we can push the alternative version now, until Joerg further improves on his values, I mean until Joerg's SPSA further improves...
I thought that values that failed are from the previous (erroneous) SPSA, but I see now that these are very close to the SPSA now running (the correct one). So I'll push this patch.

User avatar
lantonov
Posts: 216
Joined: Sun Apr 13, 2014 3:19 pm

Re: A plea to someone

Post by lantonov » Wed Apr 08, 2015 7:22 am

Patch pushed xxxx fingers

Lyudmil Tsvetkov
Posts: 6037
Joined: Tue Jun 12, 2012 10:41 am

Re: A plea to someone

Post by Lyudmil Tsvetkov » Wed Apr 08, 2015 7:55 am

lantonov wrote:Patch pushed xxxx fingers
Many thanks, Lyudmil.

My only hopes are it will not fail before 1000 games are through.

In any case, it is very difficult to resist, when one fimrly believes that magically tuning piece values could bring another 15 elo... :)

I never understood those imbalances, they simply seem not to obey any reasonable adjustment rule.

Post Reply