- Created Saturday, December 13th 2014 @ 00:07:06
Hi new here, concept looks very interesting. I'm a bit confused though how we test our bots. If I make a change I'd like to run games locally to ensure its better before uploading, otherwise I have to wait a long time to see enough games to know if there was a difference (how often do they run after upload anyway?).
Is there a way to download the server code to run my own test games, or do I need to build that all myself based on the rules page?
- Created Saturday, December 13th 2014 @ 00:56:38
Engine for previous warlight AI contest was available on github https://github.com/theaigames/conquest-engine. For warligh ai 2 seem like not available yet.
- Created Saturday, December 13th 2014 @ 11:09:39
When this competition is off of Beta, the engine source will be released.
- Created Saturday, December 13th 2014 @ 23:15:05
In the first Warlight AI competition, I found that just testing on the server was fine. It played a test match that let me watch my bot run against a dummy (or basic) bot, and that was enough to give me an idea of how my bot ran.
For real strategy testing, you need to go up against real (and different) opposing bots anyways, and the server is the best way to do that. I believe I got several games per hour last April.
- Created Saturday, December 13th 2014 @ 23:46:10
OK thanks. Several games per hour won't cut it for small incremental improvements :) But yes finding varying opponents is a must, until (if) people open source their stuff it will be against older versions of my own bot. I've done chess programming in the past where I wouldn't release a change until at least 10K games had shown an overall improvement.
- Created Sunday, December 14th 2014 @ 00:20:36
Wow. I wander what time control for engines was per game during testing? 10K games is ALOT.. Did you use some cluster for testing?
- Created Sunday, December 14th 2014 @ 02:49:35
Yeah most of it was really fast. I wrote the LittleBlitzer (http://kimiensoftware.com/software/chess/littleblitzer) automation tool just for that purpose :) I would usually do 10K insanely fast games in an hour or so for very fast validation of a change, and then do longer games for more accuracy.
- Created Sunday, December 14th 2014 @ 13:28:21
Expect more games to be played in the coming week, I'm going to let the original Warlight AI Challenge play less games so there is more room for the new competition
- Created Monday, December 29th 2014 @ 22:44:17
"I'm going to let the original Warlight AI Challenge play less games": Did you already changed it?
- Created Tuesday, December 30th 2014 @ 00:42:42
yes, half of what it was