WAC Testing

[Martin Brown]

I have been tuning my feeble engine to get most of the tactical motifs of the WAC test set correct and there remains a stubborn rump of positions where it doesn't get a solution. I decided to take a careful look at the last 21 awkward squad positions and they are mostly things where classical chess engines do badly. Blocked positions and endgames.

The tricky ones were
2, 71, 74, 80, 87, 100, 141, 146, 163, 196, 213, 222, 229, 230, 235, 247, 250, 262, 264, 265, 269, 274

And I decided to inspect these more closely with Shredder10 and a bit of manual guidance. I found four positions where there is a close or marginally better move than the official WAC best move and would be grateful if someone could verify or refute my analysis. These are:

WAC146 Bc8 (is flashy) but Bd3 or Bh3 gets the job done
WAC100 b6+ but Be3 is just as good @ 28 ply search
WAC235 Rxf7+ Qxf7+ but Qe4 is better (Q grabs p b7 later)
WAC264 Rab8 but Qe5 looks better

Have I missed something or do I have an old copy of WAC?

[Edmund]

You can download a new version at http://cap.connx.com/EPD/

The file is: wacnew.epd.bz2


All the errors you mention have already been dealt with.

[Martin Brown]

Thanks!

[bob]

I have been tuning my feeble engine to get most of the tactical motifs of the WAC test set correct and there remains a stubborn rump of positions where it doesn't get a solution. I decided to take a careful look at the last 21 awkward squad positions and they are mostly things where classical chess engines do badly. Blocked positions and endgames.

The tricky ones were
2, 71, 74, 80, 87, 100, 141, 146, 163, 196, 213, 222, 229, 230, 235, 247, 250, 262, 264, 265, 269, 274

And I decided to inspect these more closely with Shredder10 and a bit of manual guidance. I found four positions where there is a close or marginally better move than the official WAC best move and would be grateful if someone could verify or refute my analysis. These are:

WAC146 Bc8 (is flashy) but Bd3 or Bh3 gets the job done
WAC100 b6+ but Be3 is just as good @ 28 ply search
WAC235 Rxf7+ Qxf7+ but Qe4 is better (Q grabs p b7 later)
WAC264 Rab8 but Qe5 looks better

For 146, I have both Bc8 and Bd3. Do not have Bh3. All 3 appear to be pretty equal, however.
For 100, Be3 works as well, so long as b6+ is the second move.
for 235, I have not seen Qe4 as better. I tested this myself, and found Qf7 was significantly better. You may have run it a lot longer perhaps, which might lead to a different conclusion. The two moves score +7.5 and +4.5 here. 7.5 is better.
for 264, I have only Rab8 as an accepted solution.

Have I missed something or do I have an old copy of WAC?

[JVMerlino]

You can download a new version at http://cap.connx.com/EPD/

The file is: wacnew.epd.bz2

All the errors you mention have already been dealt with.

I just downloaded this, and unfortunately it makes GradualTest crash. Apparently it requires that the position IDs have quotation marks around them.

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id "WAC.002";

...instead of...

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id WAC.002;

Does anybody know of a new version of GradualTest that resolves this?

jm

[Dann Corbit]

You can download a new version at http://cap.connx.com/EPD/

The file is: wacnew.epd.bz2

All the errors you mention have already been dealt with.

I just downloaded this, and unfortunately it makes GradualTest crash. Apparently it requires that the position IDs have quotation marks around them.

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id "WAC.002";

...instead of...

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id WAC.002;

Does anybody know of a new version of GradualTest that resolves this?

jm

It comes with source:
http://home.online.no/~malin/sjakk/down ... st_src.zip

I may look into it tomorrow

[JVMerlino]

You can download a new version at http://cap.connx.com/EPD/

The file is: wacnew.epd.bz2

All the errors you mention have already been dealt with.

I just downloaded this, and unfortunately it makes GradualTest crash. Apparently it requires that the position IDs have quotation marks around them.

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id "WAC.002";

...instead of...

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id WAC.002;

Does anybody know of a new version of GradualTest that resolves this?

jm

It comes with source:
http://home.online.no/~malin/sjakk/down ... st_src.zip

I may look into it tomorrow

It's possible that that source is old. It says version 2.0, but the version I downloaded (can't remember where from) says 2.4.

jm

[jwes]

You can download a new version at http://cap.connx.com/EPD/

The file is: wacnew.epd.bz2

All the errors you mention have already been dealt with.

I just downloaded this, and unfortunately it makes GradualTest crash. Apparently it requires that the position IDs have quotation marks around them.

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id "WAC.002";

...instead of...

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id WAC.002;

Does anybody know of a new version of GradualTest that resolves this?

jm

The PGN standard for EPD requires the quotes. It might be simpler to fix the epd file rather than the program. I've done it with an editor macro.

[Martin Brown]

A marginally better move than the official WAC best move and would be grateful if someone could verify or refute my analysis. These are:

WAC146 Bc8 (is flashy) but Bd3 or Bh3 gets the job done
WAC100 b6+ but Be3 is just as good @ 28 ply search
WAC235 Rxf7+ Qxf7+ but Qe4 is better (Q grabs p b7 later)
WAC264 Rab8 but Qe5 looks better

For 146, I have both Bc8 and Bd3. Do not have Bh3. All 3 appear to be pretty equal, however.
For 100, Be3 works as well, so long as b6+ is the second move.
for 235, I have not seen Qe4 as better. I tested this myself, and found Qf7 was significantly better. You may have run it a lot longer perhaps, which might lead to a different conclusion. The two moves score +7.5 and +4.5 here. 7.5 is better.
for 264, I have only Rab8 as an accepted solution.

[/quote]

146 Rybka strongly prefers the direct obvious move here.

PV 17.06 Bd3 Ke3 c5 Kf2 cxd6 Ke3 d7 Kd2 d8Q

But it is a really difficult win with 2 queens ~ #50 after Bd3 Ke3 c5 Kxd3. Or Bd3 Ke3 Bf1 Kf2 c5 either way heading for

3Q4/8/2K5/3P4/8/8/4k3/5q2 w - - 0 7

Bh3 & Bc8 hover around 9.3. Bc8 is a much cleaner win.

235 appears to be a peculiarity of Shredder. It seems to prefer the two passed pawns. All other engines agree with you. And it is obvious playing out the lines that time to Q is much shorter your way.

264 again may be a quirk of Shredder10 (queen mobility perhaps?). Although HIARCS gives Qe5 good runner up status after ply 13. This position sometimes crashes my copy of Rybka2.32 but it does agree with Shredder that Qe5 is worthwhile after ply14 and beyond. PV at ply16

-3.05 Qe5 Bc2 Bxc2 Qxc2 Rxd4 g3 Rad8 Rxd4 Rxd4 Rb1 Rd2 Qb3 Qxc5
vs
-2.68 Rab8 Rxe7 Rxb2 Rf1 Rxd4 g3 Bb8 Re4 Rbd2 Re7 Rxc4 Rfe1 Kf8

(subject to typos - I still can't copy/paste from the CB analysis window)

Thanks for your insights.

[Martin Brown]

A couple more iffy ones I looked at this lunchtime from the previous ply.

WAC41 is inside 6 men egtb
book answer Ka5#12 Kc5#14 but Ka6 Rf8+ are also #14

And it seems to me that WAC297 is too generous in scoring Bxh2+ OK
Bxg2 is far more devastating and engines should find it.