Of course a real playing-the-next-best-move engine is silly, it fouls up at the first trade. (Also, sometimes only one legal move is possible.) Thus, my idea: It sets a cutoff instead and plays the best moves with values < v. (This could also work for a neural net engine, since it has a winning expectancy which could be used).

Has this been tried for fun? With different values of v? (This could be interesting: at which v the position will get worse ultimately? -1 PU?)

Comparable but not exactly the same.

1 Answer 1


There's a "temperature" setting for Leela that does this. See the description on the Lc0 website.

Temperature: Tau value from softmax formula for the first move. If equal to 0, the engine picks the best move to make. Larger values increase randomness while making the move. Default value: 0.00 Minimum value: 0.00 Maximum value: 100.00

  • THX. So I recalled that correctly. Commented Jan 3 at 7:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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