Bumblebee bug fix release

[zd3nik]

After taking a break for a while I came back to discover a couple bugs in bumblebee's evaluation routines. So I've created a point release:

https://github.com/zd3nik/BumblebeeRele ... .0.36898e1

The linux binaries in this release were built in a much more up-to-date environment than the last release. Built on linux mint 17.3 with gcc 4.8.4, libc version 2.19. The first release was built on fedora 22.

I've only run a couple short round robins to verify the patches. So I can't say exactly how much of an ELO increase this version has over the prior. But the increase is "noticeable" in the small tournaments I've run so far.

Coming soon (I hope): version 1.1

[Graham Banks]

After taking a break for a while I came back to discover a couple bugs in bumblebee's evaluation routines. So I've created a point release:

https://github.com/zd3nik/BumblebeeRele ... .0.36898e1

The linux binaries in this release were built in a much more up-to-date environment than the last release. Built on linux mint 17.3 with gcc 4.8.4, libc version 2.19. The first release was built on fedora 22.

I've only run a couple short round robins to verify the patches. So I can't say exactly how much of an ELO increase this version has over the prior. But the increase is "noticeable" in the small tournaments I've run so far.

Coming soon (I hope): version 1.1

If the strength gain is noticeable, could I suggest naming this release as Bumblebee 1.01?

[zd3nik]

If the strength gain is noticeable, could I suggest naming this release as Bumblebee 1.01?

Not a fan of the git revision hash to distinguish versions? Can't say I blame you.

I'll consider changing the version scheme, but don't hold your breath. I would like to finish running it through the gauntlet first, to verify there's a significant change - I've been fooled by the first few games of testing before (almost every single time in fact). The round robin I'm currently running is set at a quick 1 min with half sec increment and only 3 participants, yet it'll take another day to finish. After that I'd like to run one with a longer time control.

[Graham Banks]

If the strength gain is noticeable, could I suggest naming this release as Bumblebee 1.01?

Not a fan of the git revision hash to distinguish versions? Can't say I blame you.

I'll consider changing the version scheme, but don't hold your breath. I would like to finish running it through the gauntlet first, to verify there's a significant change - I've been fooled by the first few games of testing before (almost every single time in fact). The round robin I'm currently running is set at a quick 1 min with half sec increment and only 3 participants, yet it'll take another day to finish. After that I'd like to run one with a longer time control.

Sounds good.

[zd3nik]

If the strength gain is noticeable, could I suggest naming this release as Bumblebee 1.01?

Not a fan of the git revision hash to distinguish versions? Can't say I blame you.

I'll consider changing the version scheme, but don't hold your breath. I would like to finish running it through the gauntlet first, to verify there's a significant change - I've been fooled by the first few games of testing before (almost every single time in fact). The round robin I'm currently running is set at a quick 1 min with half sec increment and only 3 participants, yet it'll take another day to finish. After that I'd like to run one with a longer time control.

Sounds good.

I added a 4th contestant to the 1/0.5 bullet RR so it took a bit longer to finish. The result: the "buggy" version of bumblebee did better than the "fixed" version, suggesting 2 things:

1. my evaluation routine is so bad that it gets better by adding randomization to it.
2. the first few games of testing are always, ALWAYS, misleading.

I should have stuck with the original name I intended to use: bunglebee

I've started a 40 round tournament with the same 4 engines and 5 min +1 sec increment. I don't expect the result to be much different, if at all, but we'll see.

The good news (to me) is that one of the engines in this match up is bumblebee 1.1 (still in development) and it did better than both bumblebee 1.0 versions.

[zd3nik]

Turns out I had screwed up the engine setup in Arena in the tournaments referenced above. I created the bug-fix engine config by duplicating one of my (failed) experimental versions of bumblebee and I only updated the name of the config - forgot to update which engine it actually loaded!

After correcting that error and re-running the bullet tournament the results are exactly what I would have expected. The bug fix version of bumblebee scores just a couple points higher than the buggy version. The blitz tournament will take some time to re-run, but I expect similar results.

[Jamal Bubker]

Thanks Shawn for releasing this new Bumblebee version !

[zd3nik]

Thanks Shawn for releasing this new Bumblebee version !

My pleasure. I hope the engine serves you well, whatever you're using it for.