I'm looking for metrics that quantify how similar/different two positions are.

I have found a paper that used pawn structures to search for similar games in a database. This one tries to use reachable squares, as well as attack and defense structures do achieve this.

Metrics can be arbitrary, but I wonder if other approaches exist.

  • 1
    Similar positions will have similar game trees from that position, with similar branches in fact explored by chess engines. If a somewhat crude similarity measure can be found, perhaps it could be refined by using that similarity measure on a large number of positions reachable from the given position, in much the same way that chess engines can take a static evaluation function and apply it to a tree of reachable positions. I have no idea how to flesh this out, but there are various metrics which have been used in discrete math to measure tree similarity. Jan 21, 2021 at 19:58
  • 2
    A tricky task - optical similarity can be highly misleading, the slightest tactical nuance might make the difference. Jan 24, 2021 at 22:21
  • @2080 it would be best to define what you mean by similarity - optical similarity (i.e. humans at a glance with no understanding of chess rules would find the positions similar-looking) or gametree similarity (i.e. the positions are similar from the perspective of considerations pertaining to what should be played in each). Sep 11, 2021 at 0:54

1 Answer 1


Ultimately, what is considered 'similar' depends on what one is trying to achieve by measuring similarity. I suspect, a common chess engine use-case would be to try predicting if the two positions are going to return same evaluation result.

Here are a few brainstorm-style ideas:

  • Compare lists of legal moves and measure the degree to which they overlap.
  • Base the measure on the number of pieces whose positions are mismatched
  • Base the measure on the minimum number of moves needed to get from position A to position B.

Some food for thought: https://en.wikipedia.org/wiki/Edit_distance

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.