Aspiration Windows: Rubbish!

[Harald]

I did not look at everything but this I saw immediately:

Why is there an if (false) ?

The for loop is gone and replaced by the do while (true) loop.
Therefore the depth++ is missing.

The depth-- in the fail high/low cases in my code were only there to repeat the
search with new bounds but the same depth. In your code now this does not work.

In case of no fail low/high the values alpha_delta and beta_delta are not
restored in your code.

And there should be some code for mate scores that I did not write in my
pseudo code. Or at limit for growing deltas after that you try -/+infinity.

Please try again.

Harald

[ZirconiumX]

I did not look at everything but this I saw immediately:

1. Why is there an if (false) ?

2. The for loop is gone and replaced by the do while (true) loop.
Therefore the depth++ is missing.

3. The depth-- in the fail high/low cases in my code were only there to repeat the
search with new bounds but the same depth. In your code now this does not work.

4. In case of no fail low/high the values alpha_delta and beta_delta are not
restored in your code.

5. And there should be some code for mate scores that I did not write in my
pseudo code. Or at limit for growing deltas after that you try -/+infinity.

Please try again.

Harald

1. Standard Fruit practice; I have found them everywhere, and I would like my code to fit in as neatly as possible.

2. This happens at the function that calls this, but in a newer version I have code which raises depth to fit the original depth once we have a stable, useable window.

3. Sorry, I could not tell this from your code.

4. Again, this is fixed at the very top of the search by calling a new instance of this function.

5. I can accept that; Michael's code showed this.

Matthew:out

[yl84]

Aspiration windows never worked for me since my search is unstable (due to different heavy pruning). For some positions it is a win, for some others it is not interesting. So i don't use it until my search will become stable
Yves

[Richard Allbert]

Hi Matthew,

Seems you've made progress with this

One point I think I already made - fruit 1.0 has aspiration search implemented, I think this may a good reference. It's not in 2.1, but the way fruit is written, it should be easy to implemented as in 1.0.

Richard

[ZirconiumX]

I may need your 'facilities' again, Richard. (I'm sorry)

Could you please tell me the UCI error that you get with cutechess-cli? I cannot run cutechess-cli as my computer does not support Qt.

I didn't mean to upset you w.r.t Smash.

EDIT: Oooh! All this stuff! I never knew Fruit came with its own Tablebases!

FruitFly 1.2 will be interesting!

Matthew:out

[ZirconiumX]

Granted, the tablebases don't work, but they are there.

I have looked at the code, and it seems a mix between the Ippo* and Classical approaches.

Maybe Ippo took it from Fruit?

Matthew:out

[Richard Allbert]

I'll look at the UCI option tomorrow for you.

Regarding EGTB for Fruit, in the ReadMe Fabian states that the code is there for (what were) planned selfmade Bitbases.

Richard