[01:52] MSG: Quit: humhum [07:56] hey mizcu: your pocket called me this morning.. don't forget to lock the keypad on your cellphone ;) [07:59] yeah.. [07:59] i have program for automatic keylock, but not one to autostart it with cell [08:01] Join: Nenad joined #corewars [08:01] hmm.. [08:01] hmm... [08:01] * Nenad says hi [08:01] "did your pocket call me or were you just happy to hear me?" [08:01] * Nenad feels he entered in a middle of a conversation [08:02] heh [08:07] offs [08:07] 3 different software already found, all for series 3 [08:10] never liked Nokia's webpage, and it just keeps getting worse [08:13] 8 euros for a simple prog, no thanks [08:15] not worth it, the only downside is a periodic teasing from people who your pocket called. [08:17] * Nenad doesn't let his pockets call anyone [08:17] and with prices nowadays, i guess my pocket would have to call.. about 200 times [08:17] since i will most likely remove your number once you leave Fin anyway [08:19] I'm only here for another 52 days [08:26] right [08:44] so whats everyone up to today? [09:01] work :P [09:27] yeah? [09:44] * bvowk too. [09:49] need to make it faster.. [09:50] and sadly.. I can't do it by throwing more hardware [09:50] need a data wrangler to pre-process this into tiny pre-chewed chunks I think.. [09:51] a single layer of polygons that tile nicely. [09:51] So.. evolving.. [09:51] you got any ideas for keeping more diversity in a group of warriors competing in a hill style evolver nenad? [10:14] well, theres a paper somewhere about 'uniqueness hill' [10:17] is there? [10:19] either google rgc for 'uniqueness', or check Metcalfs page, theres probably copy there [10:19] hmmm... diversity issue. [10:20] well, no ideal solution. I am afraid I never looked at the details of corewar evolvers [10:20] so I don't know what has been tried [10:20] in the past [10:21] well... [10:21] hill-style is probably motr tricky, but you could make the randomness and reinsertion of old code be dependent on the variance in the scores of the current population [10:22] and also evolve mutation vectors alongside with warriors, include it in the algorithm [10:22] as for the benchmark-based evolvers [10:22] I'd suggest an approach similar to DM AdaBoost algorithm for classification improvement [10:22] to add weights to the benchmark warriors [10:23] i.e. dynamically edit the score formula on the bench [10:23] and by changing the weights [10:23] you'll enable different types to emerge [10:23] 'cause good papers would sooner or later give in to very high weights of the clr types in the bench etc [10:24] so you wouldn't force the population into convergence too soon [10:24] and you could also always in those moments when the variance drops reinsert and mutate a bit more [10:24] 'til those warriors emerge that can handle new changes in the bench weights well [10:25] 'cause it'd also be a sign of change in the strategic structure of the population [10:25] and things like that... [10:25] dunno if it makes sense to you [10:26] well, I've got a randomly changing bias that adjusts the scores based on warriors in the big hill [10:26] (ie, the big hill that is filled with human coded warriors for evaluation purposes) [10:26] that makes a big difference [10:31] Join: fiveop joined #corewars [10:32] MSG: Client Quit [10:32] hrm. [10:36] well, I just suggested a few things, that's all :) I know how GA work, and did a bit of it for my grad thesis, but not that much practical experience :P so you ought to know better, probably :) you've evolved more corewar warriors than there are human beings on earth, I think (if all the populations are considered, not just the ones you've saved in memory) [10:36] :) [10:38] you could also divide the population, bench each population subgroup against different (strategically) warrior set - one just against scanners, other just against paper, third only against stones etc [10:38] that way, different strategies *must* emerge in different groups, since their ecological niches differ greatly [10:39] and mingle them a bit with some very small probability [10:39] of course :) [10:39] and add all the aforementioned dynamic tweaks [10:39] at least that's how I'd try making an evolver if I were to have the time to do it [10:41] diversity hill might be just the thing I'm looking for.. [10:47] :P [10:49] no, it really is :) [10:49] because it would limit the number of similar hills [10:49] er.. warriros on the hill.. [10:49] and the better of two similar warriors pushes the slightly less good version off. [11:06] jeez, i got another of these 'funny' anecdotes [11:06] How do you know depressed American? He drinks too much. [11:06] How do you know depressed Finn? He doesnt drink anymore [11:06] (How do you know depressed Russian? You dont) [12:42] :) [12:42] do you drink? [13:06] * bvowk does [13:06] not like the finns do tho.. [13:20] hmm... [13:20] poor people [13:21] don't know what they're missing :P lol [13:21] though I rarely get to drink, maybe once a week or so, and almost never get really drunk [13:36] * bvowk isn't a fan of being drunk [13:36] I like good beer and wine however.. [13:36] and periodically.. tequilla. [13:43] never drank tequilla [13:43] but, wine... mmmmmmmm... [13:48] * Nenad is still killing ants at work and wonders if he's being payed for coding or bug-smashing [13:48] * Nenad prefers the former [14:14] i thought they are both kinda the same thing? [14:25] :P [14:25] most of the time, as far as I can see [14:26] luckily, I usually don't end up with too many bugs. I prefer to code 20% longer if necessary, make the code readable enough, be careful and then save hours of smashing my head against the desk [14:44] Join: fiveop joined #corewars [14:49] hello fiveop [14:50] hi [14:50] how's it going? [14:59] good [14:59] you? [15:05] pretty good to [15:17] any corewar plans? [15:17] :P [15:25] hmm, ginko icetea, interesting stuff they sell at nearby market these days [15:26] (tastes and looks kinda like local lakewater..) [15:37] hmmm... they must've been watching fools&horses while they were younger [15:41] dont know it that tells more about the tea or the lake [15:59] :P [15:59] well, neither does your statement. who knows?! maybe you lake is quite tasty [16:00] that is also a reasonable assumption :) [16:00] lol [16:00] anyway, I'm off now, so - see you later [16:00] * Nenad waves [16:00] MSG: [21:29] MSG: Remote host closed the connection [21:29] Join: alexander joined #corewars