[00:00] But I was spoiled by my younger brother. He really liked all those splatter-movies [00:00] and I was only 12 [00:00] (he was 10) [00:00] No wait, [00:00] I was 14, he 12 [00:00] I was scared shitless [00:00] heh [00:01] I remember when I was a kid, I was being babysat by my older cousin.. [00:01] but after a while I got used to it (altough I still cannot really like those films) [00:01] and he was like.. ok.. I've rented some movies I want to watch, they're scary as all hell.. you going to be able to watch them without getting upset? [00:01] so I got to see terminator and it I think.. [00:02] I'll admit to being a little freaked out :) [00:02] Hehe ... I have an idea [00:03] Take an innocent kid, let it watch T1, and then tell him, that the Terminator is now a Govenour in California [00:03] Should be really fun ;-) [00:03] lol.. [00:04] hmm ... I don't think, that it is cruel enough [00:05] anyway ... have you already fixed your little problem with your evolver [00:06] yeah.. [00:06] but I'm using the computrons for another project right now :) [00:06] so I haven't started it.. [00:06] soon tho [00:06] give the hill a chance to recover from my wrath. [00:07] How many machines usually work on nano? [00:07] I mean, how many do you usually use [00:08] 8 [00:09] hmm ... then I know, why it takes ages to get sth. useful on my computer [00:10] I'm evolving for '88, but still only have some simple clear and bombers [00:10] don't forget I also seed with chains and the like.. [00:10] that really shrinks the time.. [00:11] random code -> corewarrior is *PAINFULLY* slow. [00:11] I've taken every warrior from '88, made some statistics [00:11] I mean from Koenigstuhl/88 hill [00:11] remember it took me months to make MWR on all the machines I had [00:12] and will probably have to run it for even more month [00:12] how fast is your machine? [00:12] 2x500 [00:12] and how fast is your evolver? [00:12] 20% faster than pmars [00:13] everything running out of memory? or are you playing silly games with disks [00:13] no, everything runs in memory [00:13] it'll be a while.. [00:13] assuming everythings tuned well.. and you're lucky.. it'll be days at least.. [00:13] But I still have some things to do for optimizing the MARS [00:14] don't forgot to save the output.. and feed it back through.. [00:14] thats the biggest secret to evolving :) [00:14] In my current version, the best warriors "mate" and replace the rest AND the very best warriors survive [00:15] save what worked, feed it back through... because evolving isn't like human written warriors, they've got different needs, so taking your successful warriors and feeding them back in brings the stats back to earth a little. [00:15] do you replace the rest only if they score higher? [00:15] or just blindly replace? [00:16] I test for suicide and against a tiny benchmark before replacing [00:16] tiny = 3 warriors [00:16] how much score noise are you seeing with that? [00:17] I expect it would be awfully noisy with just 3.. [00:18] That is only the test for adding a new warrior. After that it fights 10 random warriors from the "soup" and 10 random warriors from the Koenigstuhl [00:18] each round the 10 random warriors from Koenigstuhl are the same for every warrior from the soup [00:19] Probably I do it all very WRONG [00:19] ;-) [00:19] hrm.. [00:19] thats interesting.. [00:19] how many rounds against your 3? [00:19] 150 [00:20] to low? [00:20] *too [00:20] meh.. [00:20] I'd say you might want to expand it a bit [00:20] 3 warriors with 150 rounds will be really noisy.. [00:21] so you'll fall into local minima really easily.. [00:21] so you'll end up with a bunch of crap warriors that are really effective at destroying your 3 test warriors.. [00:21] What should I take? [00:21] I'd say take 10-100 from K, and then vary the number of rounds against them.. [00:21] do say 20 for a quick test.. and 200ish for a slower test.. [00:22] the slower the test, the less noise, the fewer false positives you'll admit.. [00:23] because remember, if you're running several thousand battles per second, you're going to run across that one time that hardly ever happens when a warrior who should only win 1/10 wins 5/10 just by pure chance.. [00:23] instead of being a one time shot.. you'll start to see it more than you'd like.. and they pile up.. [00:23] so if you take a bad one say every 200 replacements.. you can push the good stuff out with lucky bad stuff pretty quickly [00:23] yes, sound logical to me [00:24] *sounds [00:24] What about the size of the soup? (at the moment 200 warriors, it already takes a lot of time for one round) [00:25] well.. [00:25] I've found that 20-100 is about optimal for a group... [00:25] you won't have that much variation in them anyways.. [00:26] but if you take a big group and divide it up into a bunch of smaller species.. (that don't or *VERY* rarely interbreed) you can up your chances of getting good stuff [00:26] so my evolver is usually 1000-5000 warriors in a pool, with species between 10-100 [00:26] should the different species fight against each other? [00:27] I don't [00:27] but most people do.. [00:27] or only against their own spcies [00:27] I only test mine against the benchmark.. [00:27] and new warriors can only replace warriors of the same species.. [00:27] so they don't really directly compete.. [00:28] what you're trying to do is keep some diversity.. [00:28] they tend to breed eachother out really quickly.. as a tiny mutation can have a huge score difference. [00:29] we should write an evolvers faq :) [00:30] good idea, altough I've found several interesting papers about evolving for cw [00:30] Unfortunately most of them were written a long time ago [00:30] ergo: small groups, few rounds [00:31] yeah.. [00:31] evolvers usually need to think bigger than they're used to in order to see gains.. [00:31] so it really helps to automate much of it.. [00:32] ... and have a whole bunch of fast computer [00:32] yeah.. [00:32] s [00:32] that helps too [00:32] but the whole process of running the evolver, saving the good results, reseeding, starting again.. saving.. [00:32] its tiresome :) [00:33] I only start my script (which either creates a new soup or reads in an already existing), evolves and saves the results each round until killed [00:34] and some info about average score, best, worst score is written to stdout [00:36] I'm talking too much [00:36] First I should publish PyCorewar [00:36] Then the evolver [00:36] heh [00:39] OK, now I really should sleep a little bit [00:40] Metcalf: Don't waste your time watching Narnia [00:40] * Fluffy waves [00:40] Part: Fluffy left #corewars [01:34] MSG: Read error: Operation timed out [01:38] Join: pkhuong joined #corewars [02:38] Join: Mizcu joined #corewars [03:32] MSG: [03:52] Join: Mizcu joined #corewars [08:23] Join: fiveop joined #corewars [10:21] Join: Roy joined #corewars [10:28] h R [10:32] Hi Miz [10:50] MSG: Ping timeout: 252 seconds [10:53] Join: Mizcu joined #corewars [10:54] goddamnit [10:55] I forgot that the version of clonecd i have has buffer overflow or something similiar, when reading .bin it eats all available memory and dies.. [10:56] Well, its uninstalled now. Never trust a warez-version. [10:59] Well, you have to trust it sometimes :) else you'll have to spend lots of money [11:03] http://www.koti.phnet.fi/redman/pl_boys.txt [11:07] arr :) [13:41] Join: asw joined #corewars [14:59] Join: Fluffy joined #corewars [14:59] :-) [14:59] fluffer.. [14:59] * Fluffy kicks Roy! [14:59] Hi Roy, nice to meet you [14:59] * Roy evades [14:59] Hi fluffer, how are you? [15:00] * Fluffy kicks Roy again and does not miss! [15:00] I'm frustrated [15:00] How come..? [15:00] I've written a little program to benchmark all the different MARS, that I could find [15:00] Second last? [15:01] I've taken some '88 warriors from Koenigstuhl and averaged the results: [15:01] pMARS 4.06 MIPS [15:01] exhaust 5.32 MIPS [15:01] exMARS 6.544 MIPS [15:01] PyMars 1.03 MIPS [15:01] exhaust-ma 7.054 MIPS [15:01] fmars (restricted to the instructions of the warriors) 10.445 MIPS [15:02] PyCorewar 5.077 MIPS :-( [15:02] Well, its getting better... [15:02] You've beat pMars, thats a nice feat [15:02] Most of the optimizations I've stolen from exhaust [15:02] and exhaust-ma [15:03] But at least there are still some left unimplemented [15:03] Well, how much is in Python atm and how much in C++? [15:04] The MARS is written completely in C + a thin access layer for Python, the parser completely in Python [15:04] I don't want to use no C++-crap, no thank you ;-) [15:05] Unfortunately I couldn't compile qMARS [15:06] hey fluffy.. [15:06] eoy [15:06] :) [15:11] er.. hey roy.. not eoy [15:11] Is there a difference? [15:12] my wife took me to see narnia last night fluffy.. its not as bad as you made it out :P [15:12] yeah.. eoy guy is an asshole.. [15:12] roys alright tho. [15:12] hehe [15:13] bvowk: Probably depends on what you expect from the movie. But at least you have to admit, that most actors are somewhat stiff and pale [15:13] they were supposed to be english during the war! of course they were pale :) [15:14] not bad for a kids movie [15:14] That is the problem ... me being no longer a kid [15:14] ;-) [15:14] my wife really liked it.. [15:14] although.. with all those weapons.. you really think someone might have lost an eye or something [15:15] probably more [15:26] Sorry, afk for a bit (work) [15:26] eoy was bugging me [15:27] Only C... what is faster, C or C++? does that speed matter? [15:28] Don't know, but if C or C++ is faster, the overhead of Python is probably more to worry about [15:29] Ah.., well, thats probably unavoidable if you want a Python oriented mars ;) [15:29] You could try to make a complete C version (just for the heck of it) [15:29] No, exhaust-ma and fmars are already out there, if you want to use C [15:30] But then you won't know how good it (can) be [15:31] Yes, but I only want a good MARS with a nice Python interface [15:31] not the fastest MARS [15:31] If I where you I would love to know how fast my mars was [15:31] Well, you already have that now [15:32] only for '88 [15:33] Yes, 94 is a bit harder.. [15:34] hehe, now I'm at 5.18 MIPS :-) [15:34] I've started to unroll everything [15:35] So far I had on the most common instructions unrolled [15:35] At least now it pays off to have testsuite [15:37] Unrolling does save time yeah.. [15:38] But after a while it is quite hard to stay focused to decide, what can be left out in the current special case and what not [15:39] Ah.. [15:55] fluffy: fmars has a python interface.. [15:55] its blindingly fast [15:55] I know, but I don't like, that you have to recompile it. [15:56] meh.. [15:56] its really really fast. [15:57] * Fluffy sometimes thinks, that bvowk sounds like a sheep (hrm, meh, meh, ... ;-) [15:57] It is fast, about 125% faster than PyCorewar [15:58] MSG: [16:01] bvowk: Let's simply assume, that I want to create yet another MARS with a Python interface :-) [16:07] bvowk: Still alive? [16:07] yes [16:08] I already thought, that mentioning sheep was to much for you ;-) [16:11] heh [16:20] Time to go [16:20] * Fluffy waves [16:20] Part: Fluffy left #corewars [17:32] Join: Metcalf joined #corewars [17:32] Hi [17:32] heya met! [17:33] Anything new today? [17:35] not really.. [17:35] pk was bragging about a new scanner technique that he picked up from something roy did [17:37] :-) [17:37] Nice, I wonder what it is... [17:37] * Metcalf must look very carefully at the code Roy posted. [17:42] Incredible, what makes someone watch the same movie 200 times! [17:44] that was yesterday [17:45] Hi PVK [17:45] hi [17:46] 4/5 line scanners are very successful on the nano hill ;-) [17:46] lol [17:46] 4 lines isn't worth the drop in speed over 5, though [17:46] (and the shitty spl -2 bomb) [17:48] carpetting with mov.i #1, -1 is much better [17:48] Join: Roy joined #corewars [17:48] howdy [17:48] Or just a single mov.i #1, -1 followed by dat carpetting [17:48] Hi Roy [17:49] So, you want to see the Narnia movie? :-) [17:49] Not anymore, discouraged now :) [17:50] Join: Neogryzor joined #corewars [17:50] Maybe if you come to Holland ;-) [17:50] Come around the end of January, then Fizmo is in Amsterdam, we could both visit him :-P [17:50] hello [17:50] Hi Neo [17:57] I decided to call my future company "Redcode" :-D [17:57] :-) [17:58] are you going to offer corewars consulting? :) [17:58] I guess corewars is going pro! [17:58] Consulting heh, going to do that as a side thing [17:59] bvowk: MME, right? ;) [18:00] I want to have a redcode cup! [18:00] I'd like a t-shirt :-) [18:01] I never did find a good price last time I was looking [18:02] http://tinyurl.com/89ho3 [18:05] The company logo :-) [18:07] A redcode cup wpuld be really cool [18:10] A real cup, and a tourney every year, the winner gets the cup [18:11] Yeah, that's a good idea. [18:11] It could be a cup like the ashes. [18:11] What could we burn? :-) [18:13] Old issues of the Corewarrior [18:15] :-( [18:22] 4 line scanner just entered the nano hill :-) [18:23] :p [18:23] Nice [18:23] add/jmz/mov/djn? [18:23] Nice too that it pushed off bvowk's oldest warrior ;-) [18:24] Yes, some combination of add/djn/jmz/mov :-) [18:30] Join: Nenad joined #corewars [18:30] Hi [18:31] Hi Nenad [18:32] Hi Nenad [18:32] hi Nenad [18:36] What have you been up to lately? [18:37] well, mostly busy learning for upcoming exams [18:37] :-/ [18:37] not much time for redcode this month [18:38] I plan to post issue 3 of CE this year, though [18:38] Roy: a nice scanner idea :) [18:38] you seem to have become a scanner expert lately [18:39] Heh expert > only making crappy non-working things, non where succesfull [18:39] how much did you manage to get against nopfsh with this one? [18:39] Why is it better than a quick fixed length attack? [18:39] well, fixed length attack is worse against stones, right? [18:40] you waste more time on carpeting empty space [18:40] But you carpet it quicker ;-) [18:40] :) [18:40] well, look at the hill. do you see any agony-style warriors there? [18:41] only once did I manage to place such scanner on 94nop - and it died quickly [18:41] The pro is the speed :) when you jump back to zooom-mode you scan .66 instead of .5 [18:41] Hmmm... that sounds like a challenge [18:42] JM: well, you can always try aging the hill a little more :) [18:42] although P. Kline is doing a good aging job with his tests [18:43] Roy: you haven't answered yet - how much against fsh? [18:43] Metcalf: you only stop carpetting when it's useless or when you found something else. [18:44] Not sure that, say, a 10 insn carpet is better than 3x3 spl (on average) [18:45] The something else you find might not be something you want to attack either. [18:45] yeah, but on average... [18:45] Why is it better than a variable length HSA style attack? [18:45] because HSA have major weaknesses [18:45] Which is quicker, and continues attacking until there is no longer a point. [18:45] they can lose up to 90% [18:45] HSA, or HSA attack? :-) [18:46] :) [18:46] well, against special stones, they lose dramatically [18:46] Or possibly, a restricted length HSA attack [18:46] try the stone from snake pit or something similar and you'll see what I was talking about [18:47] Nenad: (sorry I'm slow today) I haven't made a working version yet, no constants [18:47] oh. so that's why there were XXXX [18:47] * Neogryzor can't understand why nano-zoomishes work so badly [18:48] heh indeed :) [18:48] I thought that you simply didn't want to reveal your current set of values [18:48] didn't you try the usual failure technique - bruteforcing some? [18:48] Neo: did you see my attempt at a nano Zooom? [18:48] Fixed length HSA (Recon 2/Speed of light) do well [18:49] why are those called HSA? [18:49] Recon 2? [18:49] No Nenad, didn't want to spend time brute forcing it, seemed hopeless > its one line too big, I just know its possible to do it with one less, and pk conformed this [18:50] Metcalf I think i did, but don't recall how was it now. What was the name? [18:50] Because uhm they stop scanning when they attack, no idea, I don't think its a HSA afterall, HSA changed its bomb as quick attack [18:51] I never reffered to any fixed length scanner as HSA-type [18:51] HSA scans nearly every location and switches to dat attack, isn't it? [18:51] Neo: it was called Nano Zooom... :-P [18:51] Oh.. heh [18:52] http://corewar.co.uk/infinano/NZOOOM.RED [18:52] let me try it on the hill [18:52] Roy: how much time do you have in the following week? [18:54] would you like to help with CE3? it would be finished sooner if both of us were writing some articles. [18:57] Next week? Disney! :) [18:57] Disney? [18:57] Yeah, tonight I'm leaving for Euro Disney for a couple of days, and then Christmas, maybe after Christmas I'll have more time [18:57] MSG: Read error: Operation timed out [19:01] ... I've got to go now. I have to continue reading some probability theory for tomorrows test. I just came by to see what's new and to say hi. :) I hope I'll have some more time for redcode soon. bye! [19:01] * Nenad waves [19:01] Hope to see more of you soon [19:01] * Neogryzor waves [19:01] good luck on your test [19:02] MSG: [19:02] I'm going too, going to my family-in-law... aww [19:02] * Roy waves [19:02] :-) [19:02] * Metcalf waves [19:02] * Neogryzor waves [19:02] MSG: [19:08] my last try [19:09] Hmmm... I've had an idea what Kline might have come up with [19:10] I must go now [19:10] * Neogryzor waves [19:10] MSG: Quit: Physical Wreck IRC webscript: http://www.pswclan.com/cgi-bin/irc/irc.cgi [EOF] [19:12] * Metcalf waves [19:15] MSG: [19:28] Time to go for me too [19:28] * Metcalf waves [19:28] MSG: Quit: mov.i #1,1 [19:52] Join: Mizcu joined #corewars [20:13] Join: Fluffy joined #corewars [20:13] * Fluffy is back again [20:13] :) [20:19] hi Fluffy, hows the fur doing? [20:19] Not much fur left ... I'm getting old [20:21] And you? A new warrior? [20:22] My ears are steaming as im trying to crack a certain egg [20:23] ? [20:23] nothing worth spilling to this channel [20:24] Join: asw joined #corewars [20:24] Hi asw! [20:31] Now I know, why so many people commit suicide during xmas. All those flashy lights drive me crazy! [20:32] hi asw [20:32] Can't they restrict themselves to one little candle/flat [20:32] sorry for the lack of answer re my inquiry. [20:32] I changed program, from DMD to maths-cs ;) [20:33] * asw waves to Fluffy and pkhuong [20:36] anything new in the quantum coreworld? [20:37] MSG: [20:49] Part: Fluffy left #corewars [20:49] Join: Core29 joined #corewars [22:05] MSG: Quit: Trillian (http://www.ceruleanstudios.com