IID (skipped when in check and Score greater than alpha)

[tomitank]

Hi Colin!

Currently not available. I have to tune my search algorithm.

I'll let you know

[op12no2]

Hi Colin!

Currently not available. I have to tune my search algorithm.

I'll let you know

OK, cool.

[tomitank]

tomitankChess - PV node, not in check
tomitankChessV4 - without IID

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 tomitankChess       6   14   14   400   52%    -6   41% 
   2 tomitankChessV4    -6   14   14   400   48%     6   41% 

[AlvaroBegue]

tomitankChess - PV node, not in check
tomitankChessV4 - without IID

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 tomitankChess       6   14   14   400   52%    -6   41% 
   2 tomitankChessV4    -6   14   14   400   48%     6   41% 

That's not a statistically significant result. You need to play many more games if you want to be reasonably sure that using IID is an improvement.

[hgm]

tomitankChess - PV node, not in check
tomitankChessV4 - without IID

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 tomitankChess       6   14   14   400   52%    -6   41% 
   2 tomitankChessV4    -6   14   14   400   48%     6   41% 

That's not a statistically significant result. You need to play many more games if you want to be reasonably sure that using IID is an improvement.

But if the improvement is so small, why bother? Just use it for enhanced stability. There are properties that you cannot judge from the Elo measurement, which is an average. With the same average you can have a small difference between the worst case and a best case, or a very large one. The latter would be much more desirable.

[AlvaroBegue]

[...]

That's not a statistically significant result. You need to play many more games if you want to be reasonably sure that using IID is an improvement.

But if the improvement is so small, why bother?

My thoughts exactly. That's why I removed IID from my code. Granted, perhaps I did a shoddy job and a better implementation could work. But I need to see some Elo gain before I put it back.

[hgm]

Funny, because it is exactly the reason why I would keep it in.

Was your search really more robust without IID?

[AlvaroBegue]

Funny, because it is exactly the reason why I would keep it in.

Was your search really more robust without IID?

My search is simpler without IID, and I do value simplicity.

I don't really know what you mean by robustness. How do you define it? And how do you measure it? (To be clear, these are not rhetorical questions: I genuinely want to know your answers.)

[tomitank]

Just use it for enhanced stability

Absolutely agree.

Most engines use it. The loss or win cannot be measured for me. (or just with over ~ 10,000 games). I think with IDD the move ordering is more stability.

I will measure the most interesting results (with different time controls):
(PV,! InCheck) vs. (PV, Score > alpha)

..if I have time for it.

[flok]

Code: Select all

Rank Name       Elo    +    - games score oppo. draws
   1 dorpsgek    31    2    3 232726   55%    -8   15%
   2 iid2        -6    3    4 58152   45%    31   15%
   3 iid3        -8    4    3 58066   45%    31   15%
   4 trunk       -8    3    3 58287   45%    31   15%
   5 iid1        -9    3    3 58221   45%    31   15%

trunk: if no hash

iid1: if no hash && !isCheck

iid2: if no hash && score > alpha

iid3: if no hash && !isCheck && score > alpha