New Engine: Infrared 2

[sfbea]

Hello, I'd like to have my engine rated on CCRL if possible,

I've recently finished developing a new chess engine. It's not anything special; I estimate it's about 2400 elo on rapid CCRL (based off of a limited amount of local games, e.g. 4-2-2 wins/losses/draws, against Fornax - unfortunately I don't have the compute power for statistically satisfactory testing).

https://github.com/SFBdragon/Infrared2/ ... tag/v2.0.0

There's a 64-bit Windows executable (and Linux) through the link above, which is relatively untested on Windows, apologies if there are any issues, I'll do my best to address them. The Linux build should be in good order though.

Information:
Engine name: Infrared 2
Developer name and nationality: Shaun Beautement from South Africa
Licensing and Availability: Open Source
Estimated Rating: 2400
Interface: UCI

Of note, the engine is by default configured to be CCRL noncompliant (for friends' sake) but for CCRL, under UCI options for the engine: disable 'OwnBook' (or delete the opening book), set 'Hash' accordingly (in MiB, internally it rounds down to a power of two), set Syzygy Men to 0 to disable, or to 6 if online EGTB access is allowed (queries http://tablebase.lichess.ovh/), and set 'Thread Count' to 4 depending on your setup.

It *should* handle all time controls, but doesn't allow for search controls at the moment.

Thank you, especially for the effort and work that's gone into CCRL, it's fantastic.

[bastiball]

Hello, I'd like to have my engine rated on CCRL if possible,

I've recently finished developing a new chess engine. It's not anything special; I estimate it's about 2400 elo on rapid CCRL (based off of a limited amount of local games, e.g. 4-2-2 wins/losses/draws, against Fornax - unfortunately I don't have the compute power for statistically satisfactory testing).

https://github.com/SFBdragon/Infrared2/ ... tag/v2.0.0

There's a 64-bit Windows executable (and Linux) through the link above, which is relatively untested on Windows, apologies if there are any issues, I'll do my best to address them. The Linux build should be in good order though.

Information:
Engine name: Infrared 2
Developer name and nationality: Shaun Beautement from South Africa
Licensing and Availability: Open Source
Estimated Rating: 2400
Interface: UCI

Of note, the engine is by default configured to be CCRL noncompliant (for friends' sake) but for CCRL, under UCI options for the engine: disable 'OwnBook' (or delete the opening book), set 'Hash' accordingly (in MiB, internally it rounds down to a power of two), set Syzygy Men to 0 to disable, or to 6 if online EGTB access is allowed (queries http://tablebase.lichess.ovh/), and set 'Thread Count' to 4 depending on your setup.

It *should* handle all time controls, but doesn't allow for search controls at the moment.

Thank you, especially for the effort and work that's gone into CCRL, it's fantastic.

I think if Gabor done testing his gauntlets, he'll test it BTW Congrats to you and to your engine!

[Gabor Szots]

Hello Shaun,

Does '2' belong to the name? More clearly, is the name and version Infrared2 2.0.0 or Infrared 2 2.0.0 or Infrared 2.0.0? UCI id only displays Infrared 2.

[Gabor Szots]

I think if Gabor done testing his gauntlets, he'll test it BTW Congrats to you and to your engine!

You know Basti is interested only in engines above 3400.

[sfbea]

(I'm not sure if my previous post went through. Apologies for the duplicate if so!)

Does '2' belong to the name? More clearly, is the name and version Infrared2 2.0.0 or Infrared 2 2.0.0 or Infrared 2.0.0? UCI id only displays Infrared 2.

Good question, I think "Infrared 2.0.0" is most appropriate, name:Infrared and version:2.0.0.

Thank you very much!

[bastiball]

I think if Gabor done testing his gauntlets, he'll test it BTW Congrats to you and to your engine!

You know Basti is interested only in engines above 3400.

I'm testing in FRC but you always put it on queue and doing it faster than mine

[Gabor Szots]

There is a problem. When I set threads to 1 the engine still uses 50 % CPU on my 4-core machine.

[sfbea]

Oops, thanks for letting me know! I've fixed the bug; new version is available here: https://github.com/SFBdragon/Infrared2/ ... tag/v2.0.1
Apologies for the inconvenience.

[Gabor Szots]

It's OK now. However, I still observe an odd thing, while it is thinking it shows 133 MB memory usage but when the opponent is on move that decreases to 5 MB.

[sfbea]

I assume this is with a 256MB hash table? I've just realised my allocation code has a bug where it'll half the table size of powers of two inputs. Setting the UCI Hash option to 257 should fix this (it'll round down correctly I think), but I'll fix the bug and release a new version too.

The hash table is not persistent, and is allocated and deallocated upon each search. This was done due to persistent hashes causing serious issues with draws by repetition, especially when trying to close out a long-distance mate. This might cause issues with very fast time controls where initializing the memory span may incur a rather crucial performance hit, but it hasn't been a problem in my testing (it should compile down to a memset to zero? I'm unsure.), although if it seems to be causing issues, please let me know.

[Gabor Szots]

I assume this is with a 256MB hash table? I've just realised my allocation code has a bug where it'll half the table size of powers of two inputs. Setting the UCI Hash option to 257 should fix this (it'll round down correctly I think), but I'll fix the bug and release a new version too.

The hash table is not persistent, and is allocated and deallocated upon each search. This was done due to persistent hashes causing serious issues with draws by repetition, especially when trying to close out a long-distance mate. This might cause issues with very fast time controls where initializing the memory span may incur a rather crucial performance hit, but it hasn't been a problem in my testing (it should compile down to a memset to zero? I'm unsure.), although if it seems to be causing issues, please let me know.

Yes, I use a common 256 MB hash setting under Arena. Not a serious issue, once I did a test and have found that doubling the hash size increases strength by about 5 Elo points only.
What you do with the hash table is quite unusual as far as I know. I wonder if you could not obtain useful advice about persistent hash at the Talkchess programming subforum. http://talkchess.com/forum3/viewforum.php?f=7

[Graham Banks]

viewtopic.php?f=7&t=14641

[sfbea]

What you do with the hash table is quite unusual as far as I know. I wonder if you could not obtain useful advice about persistent hash at the Talkchess programming subforum.

I'll try that if (or more likely, when!) I get around to improving the engine further. The Chess Programming wiki also makes it clear its the best place for this kind of discussion, although it's unfortunate that I live in a country/IP block that is blocked by TalkChess - at least as far as I can tell. This isn't difficult to bypass, but it is quite a nuisance. "Forbidden\n\nYou don't have permission to access /forum3/viewtopic.php on this server." I don't suppose there's anyone here who knows the website administrators, in case it's something unintentional.

Apologies though, it seems I significantly overestimated Infrared's ability (according to the tournament results)! I wonder if architecture/hardware differences are truly enough to have gotten the difference in results I was getting, or if there were more likely other issues in my engine or judgement. I have a Ryzen 3 chip, if that's of any interest.

Nonetheless, thanks again, appreciate the time and effort!