New(?) method to find clones

[Jouni]

Is this already tried: run test suite with 100-400 positions with many engines. And simply count correlation coefficient between solution TIMES.
May be this is better way than ponder hits analysis method? Or may be bad idea?

Jouni

[Dann Corbit]

Is this already tried: run test suite with 100-400 positions with many engines. And simply count correlation coefficient between solution TIMES.
May be this is better way than ponder hits analysis method? Or may be bad idea?

Jouni

It seems to me that the easy positions will solve quickly and the hard ones will solve slowly.

I guess that it will make most engines look like clones.

[Christopher Conkie]

It seems to me that the easy positions will solve quickly and the hard ones will solve slowly.

Full of insight.....

The actual idea suggested is not unreasonable, it has been used and still is. The thing we do is to have engine specific suites...say a Crafty suite for example.

Christopher

[Mike S.]

Yes... but with enough positions sorted by difficulty, those positions could give you hints where two engines were much faster than average in the same difficult positions, and/or had unusual trouble in the same easy positions. Although I have seen from regular engines, that a version X can be quite difficult in that sense, to version X-1 of the same engine from a year ago although they certainly were very similar still, in total.

Maybe it is suitable to detect 'copy&paste' clones quickly, at least.