Kai's 2moves_80_100, try it... You'll like it.

[Dann Corbit]

I have been analyzing Kai's collection of positions designed to get engines out of their comfort zone.
They are obviously computer generated rather than collected from games or something like that.
I am analyzing to 36 plies all the positions from that list I did not already have analyzed to at least that depth.
I must say, it has been enormously entertaining watching the analysis.
Some of the positions are new, interesting ideas that I think might beneficially come into practice.
And some of them are downright hilarious.
I laughed out loud when I saw this one:
[d]rnbq1bnr/pppkpppp/3p4/8/7P/7R/PPPPPPP1/RNBQKBN1 w Q -
I imagined two five year olds playing chess.
The first one says, "Better watch out, I'm bringing out my canon!"
The second one says, "Fear not, Milady, I shall protect you!"
It's all there, from the sublime to the ridiculous, and free for the taking.
Give it a spin, and watch them go by as they are analyzed.
Hours of chess analysis pleasure.
You'll either laugh or cry, depending upon your chess sensibilities.

[Dann Corbit]

I have been analyzing Kai's collection of positions designed to get engines out of their comfort zone.
They are obviously computer generated rather than collected from games or something like that.
I am analyzing to 36 plies all the positions from that list I did not already have analyzed to at least that depth.
I must say, it has been enormously entertaining watching the analysis.
Some of the positions are new, interesting ideas that I think might beneficially come into practice.
And some of them are downright hilarious.
I laughed out loud when I saw this one:
[d]rnbq1bnr/pppkpppp/3p4/8/7P/7R/PPPPPPP1/RNBQKBN1 w Q -
I imagined two five year olds playing chess.
The first one says, "Better watch out, I'm bringing out my canon!"
The second one says, "Fear not, Milady, I shall protect you!"
It's all there, from the sublime to the ridiculous, and free for the taking.
Give it a spin, and watch them go by as they are analyzed.
Hours of chess analysis pleasure.
You'll either laugh or cry, depending upon your chess sensibilities.

The black king also threw himself in front of the queen when he could have simply captured the rook.
Just precious.

[lkaufman]

I have been analyzing Kai's collection of positions designed to get engines out of their comfort zone.
They are obviously computer generated rather than collected from games or something like that.
I am analyzing to 36 plies all the positions from that list I did not already have analyzed to at least that depth.
I must say, it has been enormously entertaining watching the analysis.
Some of the positions are new, interesting ideas that I think might beneficially come into practice.
And some of them are downright hilarious.
I laughed out loud when I saw this one:
[d]rnbq1bnr/pppkpppp/3p4/8/7P/7R/PPPPPPP1/RNBQKBN1 w Q -
I imagined two five year olds playing chess.
The first one says, "Better watch out, I'm bringing out my canon!"
The second one says, "Fear not, Milady, I shall protect you!"
It's all there, from the sublime to the ridiculous, and free for the taking.
Give it a spin, and watch them go by as they are analyzed.
Hours of chess analysis pleasure.
You'll either laugh or cry, depending upon your chess sensibilities.

I had a similar but in some sense opposite idea, how to force players (at least humans) to think from the start after a short random sequence, but instead of going for silly looking lines, require that all moves are better than passing at least, no clearly silly moves that a human would refuse to play. Basic idea is like chess 960, FRC, but instead of starting with illegal positions and new rules, all positions are legal and sensible. Here are the rules: The first 3 plies are chosen randomly (presumably by computer, though it could be done without special equipment) with these constraints: All moves must be either knight to bishop's three (old notation) or a single-square pawn move excluding the "f" pawn. So nine legal moves for plies 1 and 2, then 7 or 8 for ply 3. This should give about 600 legal move sequences, about 300 legal positions due to transposition. I'm reasonably sure that none of them would give either side anywhere close to a winning advantage, and there would not yet be interaction that leads to immediate forced variations. But I must admit, this won't reduce draws in engine play dramatically, as most of the positions will still be fairly even. Still, it avoids the need for an opening book for testing, and feels more like normal chess than positions with an early ...Kd7 for example as well as being more normal than FRC. If 300 or so positions isn't enough, the fourth ply can be included but with some restrictions as otherwise some positions will probably be losing for Black. It's probably more suitable for human chess than for engine testing, but ok for both. I'd call it "chess 300" if 300 was the exact count, but I don't know the precise number to use for the name.

[Andrew]

Have you got a link for these positions, sounds interesting/curious! I expect the typical response above after
Kd7 would be a4. If one rook is a threat, two rooks is a nightmare!

Andrew

[Laskos]

Have you got a link for these positions, sounds interesting/curious! I expect the typical response above after
Kd7 would be a4. If one rook is a threat, two rooks is a nightmare!

Andrew

The positions can be found here:
http://talkchess.com/forum3/viewtopic.p ... 5&start=38

They are generated randomly and obey the condition to fall in the interval [0.80, 1.00] of SF evaluation at short TC. That hopefully ensures that opening position is somewhat unbalanced on the border White win / Draw, and the matches from ultra-fast to LTC have about 40-60% draw rates, never going to 80+% draw rates to LTC. The openings are not drawish at all and exaggerate Elo differences, giving usually larger Elo differences between engines than regular opening suites. With pentanomial variance, the error margins can be compressed and the efficiency defined as (Elo difference) divided by (Elo error margins) is better for this suite than from regular opening suites, sometimes by a factor of 4 or so.

[Dann Corbit]

I have analyzed all of them to 36 plies or more.
The top few hundred are simply wins for the side to play (more than a pawn advantage).
This is the worst offender, a two and a half pawn advantage:
[d]rnbqkbnr/ppppp1p1/8/5p1p/7N/8/PPPPPPPP/RNBQKB1R w KQkq - acd 36; bm e4; ce 252; pm e4; pv e4 e6 Ng6 Rh7 d4 d5 exf5 exf5 Nc3 Nf6 Nxf8 Kxf8 Bg5 Qe7+ Be2 Nc6 Qd2 Be6 O-O-O Qd6 Rhe1 a6 h4 Ne4 Nxe4 fxe4 f3 Bf5 fxe4 Bxe4 Bc4 Re8 Rxe4 Rxe4 Bd3 g6 Bxe4 dxe4 d5 Ne5 Qf4+ Nf7 Rf1 Qxf4+ Rxf4 b5 Rf6 Ke7;
The bottom few hundred are simply draws (near zero score).
For example:
[d]rnbqkbnr/1ppppppp/8/8/p6P/2P5/PP1PPPP1/RNBQKBNR w KQkq - acd 36; bm d4; ce 0; pm d4; pv d4 Nf6 c4 e6 a3 d5 Nc3 Nbd7 cxd5 exd5 Bf4 c6 Qc2 Nb6 e3 Be7 Nf3 O-O Bd3 h6 O-O Be6 Rfb1 Nfd7 b3 axb3 Rxb3 Bg4 Nh2 Be6;
So I did a sql query to get a band that favors the side to move but not simply won and not simply drawn.
There are 1009 positions, so I called it Kai prime. (Pun intentional)
This link contains a 7-zip file with all of the positions along with analysis to ply 36 ( kai-2moves.epd ).
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime.epd ) which is sorted by score high to low.
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime-sor.epd ) which is sorted by Epd low to high.
This is the link to the file:


I would recommend playing the kai-prime-sor.epd file because you won't know the size of the advantage so there is more mystery.
And by all means, watch the games. It's more fun than a barrel of monkeys.

[Laskos]

I have analyzed all of them to 36 plies or more.
The top few hundred are simply wins for the side to play (more than a pawn advantage).
This is the worst offender, a two and a half pawn advantage:
[d]rnbqkbnr/ppppp1p1/8/5p1p/7N/8/PPPPPPPP/RNBQKB1R w KQkq - acd 36; bm e4; ce 252; pm e4; pv e4 e6 Ng6 Rh7 d4 d5 exf5 exf5 Nc3 Nf6 Nxf8 Kxf8 Bg5 Qe7+ Be2 Nc6 Qd2 Be6 O-O-O Qd6 Rhe1 a6 h4 Ne4 Nxe4 fxe4 f3 Bf5 fxe4 Bxe4 Bc4 Re8 Rxe4 Rxe4 Bd3 g6 Bxe4 dxe4 d5 Ne5 Qf4+ Nf7 Rf1 Qxf4+ Rxf4 b5 Rf6 Ke7;
The bottom few hundred are simply draws (near zero score).
For example:
[d]rnbqkbnr/1ppppppp/8/8/p6P/2P5/PP1PPPP1/RNBQKBNR w KQkq - acd 36; bm d4; ce 0; pm d4; pv d4 Nf6 c4 e6 a3 d5 Nc3 Nbd7 cxd5 exd5 Bf4 c6 Qc2 Nb6 e3 Be7 Nf3 O-O Bd3 h6 O-O Be6 Rfb1 Nfd7 b3 axb3 Rxb3 Bg4 Nh2 Be6;
So I did a sql query to get a band that favors the side to move but not simply won and not simply drawn.
There are 1009 positions, so I called it Kai prime. (Pun intentional)
This link contains a 7-zip file with all of the positions along with analysis to ply 36 ( kai-2moves.epd ).
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime.epd ) which is sorted by score high to low.
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime-sor.epd ) which is sorted by Epd low to high.
This is the link to the file:


I would recommend playing the kai-prime-sor.epd file because you won't know the size of the advantage so there is more mystery.
And by all means, watch the games. It's more fun than a barrel of monkeys.

Great Dann, very useful files, I need them all for some further experiments. Thanks!

[Laskos]

I have analyzed all of them to 36 plies or more.
The top few hundred are simply wins for the side to play (more than a pawn advantage).
This is the worst offender, a two and a half pawn advantage:
[d]rnbqkbnr/ppppp1p1/8/5p1p/7N/8/PPPPPPPP/RNBQKB1R w KQkq - acd 36; bm e4; ce 252; pm e4; pv e4 e6 Ng6 Rh7 d4 d5 exf5 exf5 Nc3 Nf6 Nxf8 Kxf8 Bg5 Qe7+ Be2 Nc6 Qd2 Be6 O-O-O Qd6 Rhe1 a6 h4 Ne4 Nxe4 fxe4 f3 Bf5 fxe4 Bxe4 Bc4 Re8 Rxe4 Rxe4 Bd3 g6 Bxe4 dxe4 d5 Ne5 Qf4+ Nf7 Rf1 Qxf4+ Rxf4 b5 Rf6 Ke7;
The bottom few hundred are simply draws (near zero score).
For example:
[d]rnbqkbnr/1ppppppp/8/8/p6P/2P5/PP1PPPP1/RNBQKBNR w KQkq - acd 36; bm d4; ce 0; pm d4; pv d4 Nf6 c4 e6 a3 d5 Nc3 Nbd7 cxd5 exd5 Bf4 c6 Qc2 Nb6 e3 Be7 Nf3 O-O Bd3 h6 O-O Be6 Rfb1 Nfd7 b3 axb3 Rxb3 Bg4 Nh2 Be6;
So I did a sql query to get a band that favors the side to move but not simply won and not simply drawn.
There are 1009 positions, so I called it Kai prime. (Pun intentional)
This link contains a 7-zip file with all of the positions along with analysis to ply 36 ( kai-2moves.epd ).
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime.epd ) which is sorted by score high to low.
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime-sor.epd ) which is sorted by Epd low to high.
This is the link to the file:


I would recommend playing the kai-prime-sor.epd file because you won't know the size of the advantage so there is more mystery.
And by all means, watch the games. It's more fun than a barrel of monkeys.

Dann, can you do the same, but to much shallower depth of say 25 or 27 for this very large unbalanced 5-plier (about 44k positions)?
http://s000.tinyupload.com/?file_id=582 ... 8615010287
I guess it can result in a very large and good unbalanced opening suite adequate even for Stockfish testing framework in case many draws will appear too often (which is already the case, many tests have above 80% draw rate from 2moves_v1). They do use pentanomial, which is necessary to use to calculate error margins with unbalanced opening suites.

[Laskos]

I have analyzed all of them to 36 plies or more.
The top few hundred are simply wins for the side to play (more than a pawn advantage).
This is the worst offender, a two and a half pawn advantage:
[d]rnbqkbnr/ppppp1p1/8/5p1p/7N/8/PPPPPPPP/RNBQKB1R w KQkq - acd 36; bm e4; ce 252; pm e4; pv e4 e6 Ng6 Rh7 d4 d5 exf5 exf5 Nc3 Nf6 Nxf8 Kxf8 Bg5 Qe7+ Be2 Nc6 Qd2 Be6 O-O-O Qd6 Rhe1 a6 h4 Ne4 Nxe4 fxe4 f3 Bf5 fxe4 Bxe4 Bc4 Re8 Rxe4 Rxe4 Bd3 g6 Bxe4 dxe4 d5 Ne5 Qf4+ Nf7 Rf1 Qxf4+ Rxf4 b5 Rf6 Ke7;
The bottom few hundred are simply draws (near zero score).
For example:
[d]rnbqkbnr/1ppppppp/8/8/p6P/2P5/PP1PPPP1/RNBQKBNR w KQkq - acd 36; bm d4; ce 0; pm d4; pv d4 Nf6 c4 e6 a3 d5 Nc3 Nbd7 cxd5 exd5 Bf4 c6 Qc2 Nb6 e3 Be7 Nf3 O-O Bd3 h6 O-O Be6 Rfb1 Nfd7 b3 axb3 Rxb3 Bg4 Nh2 Be6;
So I did a sql query to get a band that favors the side to move but not simply won and not simply drawn.
There are 1009 positions, so I called it Kai prime. (Pun intentional)
This link contains a 7-zip file with all of the positions along with analysis to ply 36 ( kai-2moves.epd ).
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime.epd ) which is sorted by score high to low.
It also contains a filtered version with the easy wins and likely draws removed ( kai-prime-sor.epd ) which is sorted by Epd low to high.
This is the link to the file:


I would recommend playing the kai-prime-sor.epd file because you won't know the size of the advantage so there is more mystery.
And by all means, watch the games. It's more fun than a barrel of monkeys.

Dann, can you do the same, but to much shallower depth of say 25 or 27 for this very large unbalanced 5-plier (about 44k positions)?
http://s000.tinyupload.com/?file_id=582 ... 8615010287
I guess it can result in a very large and good unbalanced opening suite adequate even for Stockfish testing framework in case many draws will appear too often (which is already the case, many tests have above 80% draw rate from 2moves_v1). They do use pentanomial, which is necessary to use to calculate error margins with unbalanced opening suites.

In fact I found a larger 2 mover unbalanced suite in my old files, about 8500 positions, I guess about a half of them are shown as truly unbalanced after an analysis. It seems one doesn't need to go to 5 plies to have large unbalanced opening suite. I am attaching it, can you analyze it to say depth 30 and build those files as you did for 2moves_80_100.epd?
Here is the file:
http://s000.tinyupload.com/?file_id=085 ... 9733442899

[Dann Corbit]

My monster machine is scheduled pretty heavily, but I think I will have time in about a day.
I have some other really fast machines, but they are shut down for the summer (my house can't take 15KW of heat burden).