I'm actually working on modifying Toga II to be a "sparring partner".
To do that I'm making some tests now but I think I'm near a correct elo decreasing formula without decrease the calculated nodes or slow down the engine.
As an example, The 2500 plays against 2200,2300,2400,2600,2700,2800 (+/- 300 elo).
Here are partial results calculated with bayeselo :
After a little bit more games, it seems that the formula is ok...
But the test continues .
The idea of this formula is to do like a human. Usually good players don't see deeper moves than not so good players ... They just make less misevaluations and give less often a gift. A "not so good" player can sometimes play an entire game without a mistake.
It's the same here.
I had a bit randomness in making blunders or evaluating a position.
De Noose Daniel wrote: ↑Tue Feb 18, 2020 1:48 pm
After a little bit more games, it seems that the formula is ok...
But the test continues .
The idea of this formula is to do like a human. Usually good players don't see deeper moves than not so good players ... They just make less misevaluations and give less often a gift. A "not so good" player can sometimes play an entire game without a mistake.
It's the same here.
I had a bit randomness in making blunders or evaluating a position.
Sounds good. I'm definitely into such things. Please let us know when a download is available to try out.
