Bounds
dav1312 opened this issue · 4 comments
It would be better if the alpha/beta bounds were the score difference to the current position, not to the starting position
I can't test right now but it seems to do what I wanted
Great. Once you have tested it, feel free to close the issue. :)
Now that I was able to test it it doesn't seem to be doing what I wanted.
Basically, due to chessdb's eval decay, every few ply there is a "random" 1cp change in the position eval that can go up or down.
For example:
| Move | Eval |
|---|---|
| g4 | -1.20 |
| d5 | -1.20 |
| e3 | -1.20 |
| Nc6 | -1.20 |
| d4 | -1.20 |
| e5 | -1.21 |
| Nc3 | -1.21 |
| Be6 | -1.22 |
In my idea, a bound of +-1 would not stop at the end, because the eval of the previous move was within those bounds.
In the current implementation, a 1cp bound will never reach the last move.
Note that this is not just about the eval decay since my idea extends to inaccurate moves too for example.
If I ask for 10cp bounds in a 0cp starting position, the current implementation limits the amount of moves to only the ones that are +-10cp, but my idea would allow a series of innacuracies as long as they are within those bounds like 0 -> 10 -> 20 -> 30 -> 40 -> 30