About the LCTII from the past

[Daniel Anulliero]

Hi all
Do you think the LCTII
https://chessprogramming.wikispaces.com/LCT+II
is allways a good test set or it's obsolete ?
Louguet created that 20 years ago , I know

[Dann Corbit]

Hi all
Do you think the LCTII
https://chessprogramming.wikispaces.com/LCT+II
is allways a good test set or it's obsolete ?
Louguet created that 20 years ago , I know

A really great test set.
Considering how long ago it was written, it is incredible how well it has withstood the test of time.

Only one problem has a serious alternative:
[d]r3k2r/3bbp1p/p1nppp2/5P2/1p1NP3/5NP1/PPPK3P/3R1B1R b kq - acd 36; ce 48; acs 443; pv Ne5 Kc1 Bf8 Be2 Ke7 Nd2 Bh6 Kb1 Be3 Nc4 Bxd4 Rxd4 Bb5 Nxe5 fxe5 Rd2 Bxe2 Rxe2 exf5 exf5 Kf6 Rd1 Kxf5 Rxd6 Rhd8 Rf2+ Kg5 Rxd8 Rxd8 a3 bxa3 Rxf7 Rd1+ Ka2 axb2 Kxb2 Rd2 Rg7+ Kf5 Rxh7 e4 Rh5+ Ke6 ; bm Ne5;

And the suggest best move still has a high score.

[Dann Corbit]

I should mention that it is not nearly so hard to solve as when it was originally written, so the time controls do not make sense for today, for the most part.

[Daniel Anulliero]

Thx Dann !
So the test just give an "idea " of engine strenght (error elo +-100)?
My engine got 235 pts for 2135 elo .
First time I have a good result with LCTII

[Jouni]

Some of the positional ones are at least dubious (tactical and endings probably correct). E.g. pos7 pos11 and pos12. In pos11 bxc5 is winning transposition.

[Dann Corbit]

I will have to try those in multi-pv mode.

I have 7 separate analysis passes for each of those positions, and every pass chose the same key move (identical to the one listed in the set).

I don't record the engine name (a bit tragic there, I will admit) so I don't know which engines produced the result sets (other than the fact that at least one pass with Stockfish is always made).

The other analysis passes might all be Stockfish or they could be other engines like Houdini and Komodo.

[jdart]

I am still running this, although now only as a sanity check before release:

see https://github.com/jdart1/arasan-chess/ ... ts/RESULTS (last line).

--Jon

[Dann Corbit]

Thx Dann !
So the test just give an "idea " of engine strenght (error elo +-100)?
My engine got 235 pts for 2135 elo .
First time I have a good result with LCTII

Epd sets don't really give a good indication of engine strength (except perhaps in a very fuzzy sense).

What they do tell you is how well your engine performs tactically, for the most part.

Most of the moves in a real chess game are quiet moves. Furthermore, the tactical moves chosen in test sets often feature sacrifice. While sacrificial moves are pretty, they are usually poor choices {in general, though not - perhaps - for the move in question}, so that null move reduces those searches. Hence, engines that solve tactical sets quickly might be tuned to have poor null move pruning.

The biggest value of test sets for engine testing is to find holes in your evaluation.

If some particular engine has a very tough time with a particularly themed test set, it is an indication that the evaluation might need to be examined for how that feature is calculated.

On the other hand, I think that they are great fun, from a human standpoint.

[Daniel Anulliero]

I am still running this, although now only as a sanity check before release:

see https://github.com/jdart1/arasan-chess/ ... ts/RESULTS (last line).

--Jon

Thx Jon!
I see I have some work to catch up Arasan

[Jouni]

What's the definition for "bm" actually? Is 0,01 better evaluation enough? I think good test position must have one clearly best move with around +2 over next moves!