[01:03] I'm using 'He Bombs Alone' because it was the highest one on the 94 hill with bomb in the name [01:19] neat, there's actually a bit of correlation [01:36] Score = 35.1 + 108 * Length [01:39] and R-sq = .492, which is acceptable but not great [05:58] Join: impomatic joined #corewars [07:08] ... but I expected a negative correlation. long scanners should lose to short bombers [07:08] perhaps the longer programs are just on average written well; so there is no causation from size [07:30] that's better. now I'm just using the top 100 from 94nop [07:30] Score = 252 - 253 * Length [07:30] against quicksilver [07:31] and R-sq improved to .691 [07:40] nope. that was wrong (calculated given instead of score). most relevant result is loss percentage vs. length [07:41] high negative correlation, so long warriors lose much less often. [07:41] theories? [08:59] Join: fiveop joined #corewars [09:45] Join: evenant joined #corewars [09:45] MSG: Read error: Connection reset by peer [11:15] Mannerisk: you should multiply the lenght by the effective ratio of bombing [11:47] MSG: Read error: Connection reset by peer [11:47] Join: nescience joined #corewars [12:50] MSG: Quit: humhum [12:51] Join: alexander joined #corewars [15:18] Mizcu: what do you mean by the effective ratio of bombing? [15:20] the amount of bomb-throwing instructions per instructions in total loop [15:20] random chaff will also play a part, but that is harder to judge [15:22] 4-instruction and 6-instruction stones are not directly comparable because they tend to veer towards different purposes [15:24] perhaps I should just use a 4-instruction bomber for the comparison [15:25] right now I'm just comparing against a single warrior [15:30] what's a good, optimized one that could work? [15:31] 4-instruction? thats a challenge.. [15:31] try baseline plus, if it had a 4-liner [17:48] Join: Fizmo joined #corewars [17:48] hi hi [18:20] Hi :-) [19:00] Hi John [19:02] today I started writing my first evolver [19:02] ;-) [19:08] Oh? :-) What language? Any special features? [19:09] Hmmm... :-( I reached a dead end again following up references. As usual I hit the ACM / IEEE brick wall :-( [19:09] it's a corewar evolver for the tiny hill [19:09] well, I don't know much about the other evolvers [19:10] Just for one hill? [19:10] We need a "Which Evolver?" article like Will Varfar's "Which MARS?" [19:11] it will be sometimes ready for all settings [19:11] I just straight forward write it code [19:11] want make ASAP the first runs [19:12] finetuning will come later [19:12] I wrote an evolver which manages to enter the nano hill :-) [19:13] cool :-) [19:14] It's written in GWBasic though! [19:15] I'm pretty optimistic, that my concept will evolve a high ratio of competetive warriors [19:15] I'm using freebasic [19:16] so its gotta be a basic evolver [19:16] You'll have to send something to Terry Newton's new hill :-) Good luck. [19:16] Maybe even a basic basic evolver :-P [19:17] Oh, I shouldn't have a problem following up my reference after all :-) [19:17] yes, it'S a basic evolver [19:17] The paper is written by someone Dutch, so yoR probably knows him ;-) [19:18] but I wrote some snippets [19:19] which I hope will speed up the evolving significantly [19:19] maybe in some days I proved me wrong :-P [19:27] ok, the great moment, I start the evolve the first time now [19:31] Evil programming challenge: [19:31] a: dat xxx [19:31] b: dat yyy [19:31] exchange the values at a and b. You are only allowed to use the DJN instruction :-P [19:32] how many? [19:33] Less than 20 :-) [19:34] you can probably do it with less than 10.. if you are allowed to screw with the second fields of a & b [19:36] mental re-write cycle 2: might take a couple more [19:37] Just DJN.B $ .... , $ .... [19:38] and a couple more.. [19:40] It's pretty slow ;-) [19:41] Anything interesting yet Fizmo? [19:43] im playing here on the side, my logical lobe is .. .. .. left with lack of .. limbering livisciousness? [19:43] lot of errors so far [19:43] just started bugfixing [19:52] MSG: Ping timeout: 240 seconds [19:56] Join: impomatic joined #corewars [20:13] bugfixing done [20:15] evolver running [20:15] :-) [20:18] but still some more tweaking to do [20:18] it's not evolve in the way I wanted it [20:20] Hmmm... maybe it wants to evolve something different to what you expect? [20:36] no, just some more hidden bugs [20:36] :-( [20:36] e.g. no adressing modes on the b-field [20:37] just need to look through the evolved warriors [20:40] also the distribution between the opcodes isn't what I expected [20:41] seems another hidden bug [20:43] anyway it can already evolve :-P [21:08] the amount of wins slowly increasing [21:08] at the beginning mainly ties [21:08] That's good news :-) [21:09] Are you evolving vs a benchmark? [21:09] no [21:09] actually it just evolve aginst itself [21:10] ;redcode-tiny [21:10] ;name breed153 [21:10] ;author Christian Schmidt [21:10] ;assert 1 [21:10] ;evolved with -=< Maezumo - Fizmo's Corewar Evolver >=- [21:10] sne.i < -397, { -150 [21:10] spl } 0, < 296 [21:10] spl # 0, < 78 [21:10] mov.i > 1, { -21 [21:10] add { 241, > -6 [21:10] sne.i @ 321, @ -400 [21:10] spl { 0, } -370 [21:10] spl } 0, # -22 [21:10] djn.f < -2, < -264 [21:10] spl < 0, < -247 [21:10] mov.i { 1, < -6 [21:10] spl @ 0, < 151 [21:10] djn.f # -3, < 384 [21:10] dat { 87, } -250 [21:10] dat # 146, < 109 [21:10] end [21:11] anyway, there is still a bug. The a-field of the spl shouldn't be only 0 [21:11] but I'm to tired now to fix it [21:12] it triples your speed of evolving for a while as is [21:12] it not interfere with the evolving [21:12] yeah [21:13] I wanted that 0 is prefered to other values for a-field of spl [21:13] ;-) [21:13] but not to be the only value [21:23] I transfered to my second pc and let it run over night [21:23] let's see what I will get out already [21:44] time to go to bed [21:44] * Fizmo waves [21:44] byebye [21:44] MSG: Quit: ChatZilla 0.9.84 [Firefox 3.0.8/2009032609] [21:44] * impomatic waves [22:02] There appears to be a company called DynaHill