[01:26] MSG: Ping timeout: 240 seconds [01:26] Join: Caelian|w joined #corewars [07:58] Join: fiveop joined #corewars [08:43] Join: Caelian joined #corewars [09:31] MSG: Ping timeout: 240 seconds [09:32] Join: fiveop joined #corewars [14:22] Join: Ant_222 joined #corewars [14:22] Hello! Muzcu: got any new array splitters? [14:27] MSG: Ping timeout: 240 seconds [14:27] Join: Ant_222 joined #corewars [14:29] Seems, I got here too early... [14:29] MSG: Client Quit [18:37] ... [18:49] Join: Ant_222 joined #corewars [18:49] Ant_222: havent gotten mail the whole week [18:50] Oh.. You mean there's only one program — mine? [18:51] I got Roy's too [18:51] Roy's is modified mine? [18:52] no, older version, sent before you that [18:52] Roy said it doesn't work. Does it? [18:53] i have not tested, again [18:53] Ok... [18:53] have been mostly reading/thinking about data compression [18:53] In RedCode? [18:54] nope [18:54] though LZW would be quite easy to implement on redcode [18:54] Working on own compressor? [18:55] my C is horrible, so cant implement much [18:56] But you have invented something (in your mind)? [18:56] and i dont feel like watching other people's source iether [18:56] just ideas [18:56] I hate investigating other's sources too [18:57] Well, what ideas? Will I understand if you explain?" [18:59] well, you probably havent read my mumbling about C i had couple days ago [18:59] Where you were working a 2-dimentional array? [19:00] yes [19:00] I did [19:00] I ave [19:00] I have. [19:01] Don't like C? [19:01] such an array could be used to speed-up making prediction tables, or work as sort of weak predictive LZ [19:01] i just dont have enough experience with it [19:02] I get confused with all those {}<>*@||! and other symbols instead of begin/end, and, or, not... [19:03] Also this thing a+++++=++c [19:08] in the symbol's following symbols table (2-d array thingy) it would be slightly faster to find a order-1 perdiction, since you only have one string to analyze to get the symbols following [19:10] if array (a) contains only bbbbb, one can just put (ab) in directory of directory-based compressor [19:11] Yes, I get it. Order-1 - because you look only for one symbol after the current symbol. [19:13] and then one could take order-2 just by watching relative string of (b) [19:13] Yes. [19:14] Faster than scan the whole input text every time. [19:17] i dont think one needs to turn the file into into the table at one time either [19:18] ..the whole file into table.. [19:19] Yes, because it would be astronomiocally huge. So what to do? Divide into smaller parts? [19:20] read a part of file, tableize it, then analyze the table until you meet non-existant value of table [19:21] free the already read parts of table, assign more and get some more from file [19:22] So, if you get A->BBBBBBBB in this part that won't mean there are no other symbols after 'A' futher... [19:22] which of course is a problem if you are reading a symbol which doesnt exist until few gigs of file have been read [19:24] You can divide it into pieces big enough to allow for normal compression thereof separately, independantly of each other... [19:25] which is acceptable for "normal" compression, but you are making a model of context, then that can be a bad thing [19:27] .. [19:27] ... [19:28] well, one could cache blocks on HD [19:29] Yes, but... Quite space-consuming anyway... [19:29] but if one takes a cached block for one symbol, there will be a bubble in the main table [19:29] (a)bbbbccc...(content of 5 blocks)..d... [19:30] so d would need to be pushed forward when reading the 5 middle blocks [19:33] By a block you mean stats for one symbol? [19:34] by block i mean raw data before being added to the table [19:35] hmm, maybe that problem is not a problem after all [19:37] Raw data == part of the initial file? So cache it? — I don't get something... [19:37] So _why_ cache it? [19:38] so: to prevent loading whole file into memory, it is bypassed by dividing it into blocks [19:38] and a block in memory is tableized [19:38] Ok and.... [19:39] and the table is used to creating a model, until either block ends, or a symbol is required which is not in the table [19:39] i thought it as a problem, if the symbol required is not found in the next block [19:39] So if you wanna cache something, it must be tables, not the raw data... [19:40] but no matter how far you read blocks, the stuff between doesnt matter since you are searching for occurance of one symbol [19:40] so there is no need to cache blocks [19:40] Yes, but a search in a long file takes time proportional to its length! [19:41] which is a problem if the stuff_between is large [19:44] MSG: Ping timeout: 240 seconds [19:44] Join: Ant_222 joined #corewars [19:45] ... Sorry, bad connection. ...Yes, and you can estimate average search length in your data. Will it be high? [19:45] I am running out of money and they'll disconnect me from minute to minute ;((( [19:46] dont let me make you poor [19:46] I am downloading stuff. Don't worry. [19:49] maybe one could read every symbol in each block and reorder them to prevent big "hops" [19:49] reorder the block by referenced symbols, that is [19:50] referenced — those in one-dimentional array? [19:50] yes [19:51] Well, won't that lighten only one-order predictions? [19:53] it could lighten multi-order, if the data had non-uniform symbol spread [19:53] so wouldnt work [19:53] Why, is your symbol spread niform? [19:53] Normally it's not uniform. It's ofter normal. [19:54] let me make a picture [19:54] OK, but don't wonder if I suddenly disappear... [19:59] bah, i give up on that [20:00] OK, let me ask another dumb question: how can a required symbol be not in the table? [20:00] But idea would be to sort the blocks, that when reading first block there would be as many references only to block itself or next block, instead of some later blocks [20:02] hmm... [20:02] Unless it's the last symbol, and such that it's its the only occurence in the block. [20:03] i am a damn overengineer-er [20:04] thats completely correct [20:04] if a symbol is not in the table, then it is the next symbol in data [20:07] How is the algorithm you're trying to improve called? [20:07] LZ*? [20:08] currently just context-modeling [20:08] and a model can be used for LZP (lz-prediction) or PPM (prediction by partial matching) or if insane, DMC (dynamic markov coding) [20:09] I just accidentally found original articles about LZ77, LZ78 and LZW. Need links? [20:10] i have read a lot about them already [20:10] Original — by the inventors. [20:11] shouldnt be that hard to find anyway [20:11] i think wikipedia had a straight link to atleast one of them [20:24] (you dropped, or just reading a lot?) [20:26] Dropped what? [20:26] I am still here. [20:26] from net, but you are alive [20:26] Was reading [20:28] Strange, I have 0 rubles in my account and it's not getting below zero. Usually it did. [20:38] Ok, Good Luck with programming and bye! [20:39] bye [20:39] MSG: Quit: Ant_222 [21:53] Join: Roy joined #corewars [21:53] Mizcu: I created a range encoder not so long ago, also very fun to tweak :) [21:55] hR [21:58] hi! [21:59] I had an idea to send to sal-b to balance, but it socked.. [21:59] It happens.. I have ideas like that all the time, they all failed hehe [21:59] i need an ssd-clear that has everything required after the clear [22:00] well, i have the monowire's one, but still doesnt work [22:00] idea was -> 0.8c scan -> check if a bomb -> "leap" a ssd or d-clear accordingly [22:00] Look at the one that is in Bottomless Pit, thats a very good one [22:01] Not a really standard SSD but very effective [22:01] ugh, not forgotten scanner's clear [22:01] i hate it [22:02] hehe, or just a simple SD clear instead? [22:02] I really like this one: [22:02] ;redcode-94nop [22:02] ;name Breaker [22:02] ;author Roy van Rijn [22:02] ;assert 1 [22:02] cStep equ 35 [22:02] cOff equ 6893 [22:02] cAdd add.f cIncr, cScan [22:02] cScan sne.i cOff, cOff+17 [22:02] djn.f cAdd, @cScan [22:02] cIncr spl.f #-cStep, >-cStep [22:02] mov.i @cBomb, }cScan [22:02] djn.f -1, cBomb dat #8, {1 [22:02] spl #7, {-85 [22:02] end cScan [22:02] Its so small and simple [22:02] id send the source right now, but the better it at another computer [22:03] mono's clear was spl #0, mov @3, >gate mov @2, >gate djn.f -2, {2 | dat xxx, 1 | spl #blah [22:04] so it ignored the problems a stargate or rosegate has with some imps [22:05] hmm, its the same you gave, with > instead of } [22:05] whatever [22:05] For something completely different: Today me and my girlfriend got news from the real estate agent that we where picked so we can buy a house! Going to be build this winter [22:10] tell, do they adverstise ringtones/background for cellular phone in Neederlandese TV? [22:10] Not a fucking day i open TV without seeing one [22:12] Yes :( Jamba (!!?!!!) [22:12] J no longer here, thank god [22:12] And something called RingtoneKing [22:13] which is Jamba with a new name because Jamba had a 'bad reputation' [22:13] no shit [22:30] MSG: Quit: Lost terminal [23:29] MSG: