21st Amateur Series Division 5

[Graham Banks]

Thanks for the tournament

As for nanoSzachy, its author thinks that 3.7 was somehow overrated, and 3.8 is very similar (slightly faster + some endgame stuff).

He could well be right. More games are needed.

[Piotr Cichy]

Versions 3.7 and 3.8 of nano are almost identical and generate in 99.99% cases the same moves. There are only 2 differences between 3.7 and 3.8:

1. added recognition of some drawish pawnless endings
2. some speed optimizations

ad 1: these endings are too rare to have any influence on playing strength
ad 2: the speedup is about 3-4%, in practice hard to notice any difference

Normally I would not release version 3.8, because it is generally the same as 3.7, but version 3.8 took place in IOPCCC, so I decided to release it.

In CCRL we can see:

NanoSzachy 3.7 2600 +37 −37 54.9% −39.0 28.2% 255
NanoSzachy 3.8 2532 +30 −30 50.9% −8.1 28.6% 377

It is strange, that almost identical engines produce so different results, their ranges of ELO even don't overlap! What is the reason? Maybe not enough games or different set of opponents?

[Graham Banks]

Versions 3.7 and 3.8 of nano are almost identical and generate in 99.99% cases the same moves. There are only 2 differences between 3.7 and 3.8:

1. added recognition of some drawish pawnless endings
2. some speed optimizations

ad 1: these endings are too rare to have any influence on playing strength
ad 2: the speedup is about 3-4%, in practice hard to notice any difference

Normally I would not release version 3.8, because it is generally the same as 3.7, but version 3.8 took place in IOPCCC, so I decided to release it.

In CCRL we can see:

NanoSzachy 3.7 2600 +37 −37 54.9% −39.0 28.2% 255
NanoSzachy 3.8 2532 +30 −30 50.9% −8.1 28.6% 377

It is strange, that almost identical engines produce so different results, their ranges of ELO even don't overlap! What is the reason? Maybe not enough games or different set of opponents?

Taking into account what you've said, it could well be a combination of both. When I give NanoSzachy some more games, it will be with the intention of trying to get the average opponent rating closer to that of NanoSzachy 3.8.
Piotr - would you like the results combined under NanoSzachy 3.8? That would save me some work.

Cheers,
Graham.

[Graham Banks]

The playoff match between Pawny 0.2 and NanoSzachy 3.8 is in progress. Pawny currently leads by 5.0-3.0, but NanoSzachy could be winning Game 9.

[Piotr Cichy]

Piotr - would you like the results combined under NanoSzachy 3.8? That would save me some work.

Yes

[Graham Banks]

Piotr - would you like the results combined under NanoSzachy 3.8? That would save me some work.

Yes

Very good. This will be done for next weekend's update because this weekend's one has already been done.

Cheers,
Graham.

PS - I like NanoSzachy. Thanks for your efforts.

[Graham Banks]

The playoff match between Pawny 0.2 and NanoSzachy 3.8 is in progress. Pawny currently leads by 5.0-3.0, but NanoSzachy could be winning Game 9.

Pawny prevailed by 6.0-4.0, so will play in Division 4 of the next series.

[Mincho Georgiev]

Great! My intention is to delay the next release until sufficient amount of games gets collected (200+) to avoid the headache for you and others testers. Thanks Graham!

[Piotr Cichy]

The playoff match between Pawny 0.2 and NanoSzachy 3.8 is in progress. Pawny currently leads by 5.0-3.0, but NanoSzachy could be winning Game 9.

Pawny prevailed by 6.0-4.0, so will play in Division 4 of the next series.

A pity for nanoSzachy However, I don't plan to release new version in near future, so it probably will go down in ranking lists.

Graham, could you consider testing pikoSzachy instead of nanoSzachy? I think it is quite interesting engine because of it's size - only 10KB, and is not much weaker than nano.

[Graham Banks]

The playoff match between Pawny 0.2 and NanoSzachy 3.8 is in progress. Pawny currently leads by 5.0-3.0, but NanoSzachy could be winning Game 9.

Pawny prevailed by 6.0-4.0, so will play in Division 4 of the next series.

A pity for nanoSzachy However, I don't plan to release new version in near future, so it probably will go down in ranking lists.

Graham, could you consider testing pikoSzachy instead of nanoSzachy? I think it is quite interesting engine because of it's size - only 10KB, and is not much weaker than nano.

I'll ask the CCRL blitz testers if they could test pikoSzachy for you if they haven't already done so.
I'm having difficulty keeping up with all the new engine releases at present, but I'll try to test pikoSzachy at some stage too.

Cheers,
Graham.