KMT Chess 1.2.1 released. Now with analysis support!

[P. Villanueva]

Hello!

I've released KMT Chess 1.2.1. Main changes:

- Analysis support.
- Can run .epd suites if the interface supports them (tested in Arena).
- Fixed a bug that allows reductions in passed pawn advances.
- Fixed a bug that caused (very rarely) search erronously return -infinite, and KMT resigns even being in advantage.
- Fixed a bug that caused big nps drop while pondering.
- Fixed a bug that caused using an illegal move as pondermove.
- Now KMT uses the transposition table to avoid null-moves in positions where there is no chance of getting a cut off.
- Now null-move reduction is proportional to depth. Branching factor reduced.

In a test tournament I got these results:

Code: Select all

Ponder off.
Used Nunn positions as starting positions.
Level: Blitz 2/1

Engine                      Score     S-B
1: Rodin 2.3a               55,5/80   1978,5
2: KMTChess 1.2.1           41,5/80   1528,0
3: ATAK 6.5                 41,0/80   1606,2
4: Feuerstein 0.4.6.1       32,0/80   1228,7
5: KMTChess 1.2             30,0/80   1251,0

Best regards

[swami]

Thank you for adding in analysis support.

[beachknight]

Thanks, Paul.

Will be included in next tournament in a couple of hours.

Best,

[swami]

I found one bug:

The engine keeps restarting itself everytime the new position is supplied. As in Arena Automatic analysis mode.

Therefore it takes a lot of time to test it since restarting itself takes about 3 seconds. 10 seconds per position might look like 13 seconds per position because the 3 seconds is a lag.

[P. Villanueva]

I found one bug:

The engine keeps restarting itself everytime the new position is supplied. As in Arena Automatic analysis mode.

Therefore it takes a lot of time to test it since restarting itself takes about 3 seconds. 10 seconds per position might look like 13 seconds per position because the 3 seconds is a lag.

Ouch!

That is because the winboard feature "reuse" is set to 0. Then every time a new game begins, the interface close the engine and launch it again.

I´ve just uploaded the files with that feature set to 1. That solves the problem.

[swami]

I´ve just uploaded the files with that feature set to 1. That solves the problem.

Thanks for the quick fix! Will download it and test it now

[swami]

KMTChess 1.2.1
by Paul Villanueva, Spain

Strategic Test Suite Conditions:

Arena GUI
10 seconds per position
800 positions
Core2Quad 32 bits, Q6600, 2 GB RAM, 2.4GHZ
Engine uses 99 Mb Hash.

Subject-wise Scores:

STS (v1.0) - Undermining:
63/100, Grade: B

STS (v2.1) - Open Files and Diagonals:
57/100, Grade: C+

STS (v3.0) - Knight Outposts/Centralization/Repositioning:
42/100, Grade: E

STS (v4.1) - Square Vacancy:
49/100, Grade: D

STS (v5.0) - Bishop vs Knight:
66/100, Grade: B+

STS (v6.0) - Re-Capturing:
51/100, Grade: C

STS (v7.0) - Offer of Simplification:
50/100, Grade: C

STS (v8.1) - Advancement of f/g/h Pawns:
32/100, Grade: ?

Overall Performance:

Total Score: 410/800

Overall Average: 51.25 %

Grade: C

Regards,
Swami

[P. Villanueva]

STS (v8.1) - Advancement of f/g/h Pawns:
32/100, Grade: ?

I suspeted that my king safety code is atrocious but don´t know how to make it better. I can´t imagine how to improve KMT results in that test.

[swami]

STS (v8.1) - Advancement of f/g/h Pawns:
32/100, Grade: ?

I suspeted that my king safety code is atrocious but don´t know how to make it better. I can´t imagine how to improve KMT results in that test.

That test is the toughest of all the tests, for all engines. Most engines value king side pawn cover more. Aggressive engines will see it differently.

I think KMT needs to improve "Knight Outposts/Centralization/Repositioning" more - It's one of the easiest for other engines.

KMT has very good knowledge of "Bishop vs Knight", the score on this test is on par with score of many 2600+ engines.

[P. Villanueva]

Now there is a 64bit version by Dann Corbit. Thanks Dan!

http://sourceforge.net/projects/kmtches ... p/download