Idea I have for improving engine's eval.

[voyagerOne]

When doing brute force testing of 10s of thousands of games...the lost games should be broken down into categories. Such as:

Open Position
Closed
Queens off board
Mid-Game
End-Game
Double Bishops
Was pawn isolated
Was double pawn
Material imbalances
etc...etc...etc...

So for instance if the majority of lost games was because of closed positions...that's where the attention should be to improve the eval.

I don't know...just thinking out loud... what you think?

[Rein Halbersma]

nihil sub sole novum http://chessprogramming.wikispaces.com/Evaluation

[voyagerOne]

@ Rein... the idea is that instead of getting number of w/l/d. One can see which categories (such as the one I listed) that had the majority of losses.

This may help one know what weaknesses of the engine that they need to improve upon.

[Kempelen]

When doing brute force testing of 10s of thousands of games...the lost games should be broken down into categories. Such as:

Open Position
Closed
Queens off board
Mid-Game
End-Game
Double Bishops
Was pawn isolated
Was double pawn
Material imbalances
etc...etc...etc...

So for instance if the majority of lost games was because of closed positions...that's where the attention should be to improve the eval.

I don't know...just thinking out loud... what you think?

I think many people do this in one way or another. I do this for material imbalances. I had a feature in my engine that catalogues rodin loses and put the in a csv file, so I can see whare better tunning is needed.