A Draw oriented opening position epd file

[Norm Pollock]

I made a draw-oriented opening position file (epd) for use in Arena. I believe this type of position file is exceptionally well suited for engine testing, assuming you like to test engines from an opening epd file using colors reversed.

All positions occurred after 8 full moves. The positions occurred multiple times in a file of 40K master level (2425+ elo) drawn (1/2-1/2) games played since 2006 at long time controls, with 25+ full moves, and without blunders in the first 25 moves.

If you find any positions that are not fair, ie, favor one side, please let me know.

https://dl.dropboxusercontent.com/u/662 ... ineTest.7z

[Mike S.]

1,056 positions! That should be sufficient. Thanks.

[Mike S.]

P.S. For interfaces which do not support EPD starting positions directly, I have converted the file into PGN format:

http://remixshare.com/dl/hq2ug/NP-engineTestPGN.zip
(15 seconds waiting time only)

[Ferdy]

I guess this position favors white after the move Bc1f4.
[d]r1b1kb1r/ppq1pppp/1np2n2/8/2NP4/2N2Q2/PPP2PPP/R1B1KB1R w KQkq -

[Norm Pollock]

I guess this position favors white after the move Bc1f4.
[d]r1b1kb1r/ppq1pppp/1np2n2/8/2NP4/2N2Q2/PPP2PPP/R1B1KB1R w KQkq -

This position has a centipawn score of +0.32 after 60 seconds of analysis by Houdini 1.5 on my 2cpu intel machine. I would like classify that as a slight advantage that a very strong engine like Houdini could turn into a win against most opponents.

So what I am doing is using Houdini 1.5 to do an automatic analysis in Arena for each position for 15 seconds, and any position with a centipawn score over +0.25 or less than -0.25, will be extracted and tested again for 60 seconds. Then the outliers will be removed and an update released.

[Norm Pollock]

Updated engineTest.epd by removing some outlier positions that might not be neutral, based on analysis by Houdini 1.5.

File now contains 996 neutral draw oriented positions.

https://dl.dropboxusercontent.com/u/662 ... ineTest.7z

[Norm Pollock]

engineTest.epd has been updated to 1570 neutral opening positions (after 16 plies) suitable for fair engine testing.

The download link has been added to my chess homepage (see WWW below).

Fair and neutral opening positions are needed to ensure that no engine wins a game (or draws a game) based largely on a favorable opening position. Likewise, no engine draws (or loses a game) based largely on an unfavorable opening position. Testing engines should be just that, and biased out-of-book opening positions should be eliminated when testing.

[carldaman]

engineTest.epd has been updated to 1570 neutral opening positions (after 16 plies) suitable for fair engine testing.

The download link has been added to my chess homepage (see WWW below).

Fair and neutral opening positions are needed to ensure that no engine wins a game (or draws a game) based largely on a favorable opening position. Likewise, no engine draws (or loses a game) based largely on an unfavorable opening position. Testing engines should be just that, and biased out-of-book opening positions should be eliminated when testing.

Hi Norm,
What would you consider a good threshold for defining 'biased' in this context? It may be somewhat subjective, although I would agree that out-of-book positions should not overly favor/disfavor either player. Also, an out-of-book position can actually be overly drawish, leading to a likely draw, and thus being unfavorable to a stronger player.

It could perhaps be argued that having too many draws tends to defeat the purpose of testing.

Regards,
CL

[Norm Pollock]

engineTest.epd has been updated to 1570 neutral opening positions (after 16 plies) suitable for fair engine testing.

The download link has been added to my chess homepage (see WWW below).

Fair and neutral opening positions are needed to ensure that no engine wins a game (or draws a game) based largely on a favorable opening position. Likewise, no engine draws (or loses a game) based largely on an unfavorable opening position. Testing engines should be just that, and biased out-of-book opening positions should be eliminated when testing.

Hi Norm,
What would you consider a good threshold for defining 'biased' in this context? It may be somewhat subjective, although I would agree that out-of-book positions should not overly favor/disfavor either player. Also, an out-of-book position can actually be overly drawish, leading to a likely draw, and thus being unfavorable to a stronger player.

It could perhaps be argued that having too many draws tends to defeat the purpose of testing.

Regards,
CL

"Biased" is a subjective call. I define "biased" as being significantly different in centipawn valuation from the standard opening position. I consider the standard opening position to have a centipawn valuation of +0.15. I consider any out-of-book centipawn valuation outside the range -0.30 to +0.30 to be "biased". But this is subjective, and of course, the evaluation of the centipawn valuation is not an accurate measurement in the first place.

Theoretically out-of-book opening positions that are extremely drawish should result in draws if the 2 engines are close in ability. But I do not see how I can determine a position to be overly drawish after only 8 full moves. Certainly I can determine if a position is overly winnable, but overly drawish is very subjective. In any case, the stronger engine has the better chances.