00:08:31: <blob_> right now I'm thinking that a table-driven approach could simplify things a lot. appropriate actions are index by op code, modes and modifier.
00:09:47: <blob_> of course tables are not really easier to read than spaghetti code, but the code would at least be much shorter
00:14:26: <blob_> and ots less error-prone
00:27:38: -!- blob_ has quit (Quit: leaving).
03:45:49: -!- yoR has quit (Quit: Page closed).
12:00:16: -!- Payhn has quit (Quit: ZNC - http://znc.in).
12:02:27: -!- Payhn has joined #corewars.
17:44:44: <bvowk_> ...
17:58:56: <impomatic> Hi Bvowk_ :-)
18:21:12: <bvowk_> how's it going Mr Imp?
18:53:30: <john_metcalf> I'm about half way to finishing the mirror on Jens' site. It's taking a while because all of the hyperlinks need changing
19:15:11: -!- Doxin has joined #corewars.
20:06:28: <impomatic> bvowk_: any chance you've got a copy of dissc_client.zip ?
20:18:34: <bvowk_> hrm..
20:23:54: <bvowk_> Sorry john, it looks like I didn't keep that.
20:24:05: <bvowk_> or the tar.gz version.
20:32:28: <impomatic> Not to worry
20:36:07: * Doxin worries intently
20:43:43: * bvowk_ medicates Doxin.
20:44:30: <Doxin> ._.
20:45:02: <bvowk_> So, I fixed a couple bugs in my evolver, and stole a couple interesting ideas for maintaining diversity from uGP and a couple other sources...
20:45:24: <bvowk_> and now instead of slowly making clears... I make a lot of slightly different clears really fast.
20:48:50: <bvowk_> I'm thinking I might need to tweak my selection algorithm to account for the scoring dimensions as well as the absolute score to get things that don't evolve as quickly as clears to survive.
20:49:19: <Doxin> bvowk_: you'll want to adjust fitness score based on how unique a warrior is
20:49:33: <bvowk_> just basing it on scoring dimensions is a bit too far afield tho.
20:49:36: <Doxin> bvowk_: so after 100s of generations of only clears those start scoring worse.
20:50:36: <bvowk_> the problem is that crappy pseudo clears become clears much easier than crappy papers become viable.
20:50:56: <bvowk_> hrm.
20:51:01: <bvowk_> Thanks Doxin :)
20:51:38: <Doxin> bvowk_: then way at the end of your run you compare the past best warriors without the uniqueness metric
20:51:44: <bvowk_> I think I just rubber ducked you.
20:53:38: <Doxin> you might have.
20:53:53: <bvowk_> I've got a bit more of a big data approach to evolving tho, there's not really an end point.. I just keep collecting forever.
20:54:32: <bvowk_> I throw away the really crappy things, but the rest are saved and scored against a benchmark and can be resurrected.
20:54:49: <Doxin> right
20:55:09: <Doxin> bvowk_: that's pretty much the required approach when doing diminishing-returns metrics like uniqueness.
20:55:21: <bvowk_> There's ~50M "good" warriors in my collection now, which means they score >~120 on a somewhat noisy benchamrk from valhalla
20:55:34: <Doxin> fancy
20:55:40: <impomatic> You could always reward warriors which score differently to the other warriors http://corewar.co.uk/astolfi/diversity.htm
20:56:05: <bvowk_> imp: yeah, I actually tried using that for fitness.
20:56:25: <bvowk_> and then ranking based on absolute score adjusted by diversity.
20:57:21: <bvowk_> But I think fitting that into the selection part might be the ticket.
20:57:57: <bvowk_> Prune based on fitness, select based on uniqueness and diversity, so the oddballs get more chances to reproduce.
20:58:51: <Doxin> that'd work
20:59:08: <bvowk_> We'll find out )
20:59:09: <bvowk_> :)
21:08:45: <bvowk_> Maybe the oddballs will just fail more.
22:25:00: -!- gb_master has joined #corewars.
23:26:29: -!- impomatic has quit (Ping timeout: 276 seconds).