*** zapper has left the channel2009-11-21 21:16:36
*** King_Kendo_Nagasaki has joined the channel2009-11-22 02:55:41
*** King_Kendo_Nagasaki has left the channel2009-11-22 03:37:27
*** Shelwien has left the channel2009-11-22 07:55:13
*** Desmonde_Wolf has joined the channel2009-11-22 16:55:38
<Desmonde_Wolf> does anyone actuallt use this???2009-11-22 17:23:41
 gues not...2009-11-22 17:23:54
*** Desmonde_Wolf has left the channel2009-11-22 17:24:01
*** Shelwien has joined the channel2009-11-22 21:02:06
*** pinc has joined the channel2009-11-23 07:21:55
*** rr has joined the channel2009-11-23 08:49:42
*** rr has left the channel2009-11-23 08:50:55
*** Shelwien has left the channel2009-11-23 14:29:52
*** pinc has left the channel2009-11-23 17:07:16
*** Shelwien has joined the channel2009-11-23 18:01:35
*** mike has joined the channel2009-11-23 18:06:20
<mike___> hi2009-11-23 18:17:15
<Shelwien> hi2009-11-23 18:17:23
<mike___> not much activity here, hm?2009-11-23 18:18:40
<Shelwien> there's some, see http://ps16893.dreamhost.com/log2009-11-23 18:19:03
<mike___> ah, nice2009-11-23 18:26:07
 i'll just add it to my autojoinlist, maybe i can read some discussions now and then :)2009-11-23 18:29:12
<Shelwien> that's a good idea ;)2009-11-23 18:29:50
 unfortunately most of people on the forum don't understand IRC2009-11-23 18:30:09
 so they appear, wait for a minute, see that nothing happens, and quit ;)2009-11-23 18:30:30
<mike___> hehe. i just discovered that thread in the forum. and since the forum is a quite nice source of information i thought i'll try this channel...2009-11-23 18:35:05
<Shelwien> well, its more of a "talk to Shelwien" channel for now ;)2009-11-23 18:35:30
<mike___> i've read your entry about mixing with negative weights. have you investigated that any further?2009-11-23 18:37:30
<Shelwien> i don't really remember, that's something fairly old probably ;)2009-11-23 18:38:25
 but no, and I don't think thats a good idea2009-11-23 18:38:39
 its basically an example of overtuning, when it improves compression2009-11-23 18:39:22
 maybe inverted counters have a similar effect though2009-11-23 18:41:18
<mike___> inverted counters?2009-11-23 18:41:31
<Shelwien> i mean, ones where probability of 0 increases when 1 encountered ;)2009-11-23 18:41:40
<mike___> ah. like "false" models2009-11-23 18:41:55
<Shelwien> my linear counters can turn like that with appropriate parameters2009-11-23 18:42:09
 and automatic optimizers sets them like that sometimes2009-11-23 18:42:28
 like in the BWT model2009-11-23 18:42:37
 ...2009-11-23 18:48:32
 btw, there's a decompiled ccm source in the topic, with a parallel rangecoder attached to it for speed improvement2009-11-23 18:49:06
<mike___> i haven't looked that much into all those code posted at the forums. i'm quite new to the field of data compression. i've been testing different kinds of mixing strategies lately and now i'm thinking of mixing with negative weights, but so far i haven't come up with a reasonable mixing scheme :/2009-11-23 18:52:07
<Shelwien> well, the main problem with that is that there's no physical sense in it2009-11-23 18:52:55
 wtf is negative probability?2009-11-23 18:53:07
 but if you want it just for fun, its not so hard to implement2009-11-23 18:54:01
<mike___> well of course i'm trying to mix it somehow so that the mixed probability is positive and in (0,1) 2009-11-23 18:54:57
<Shelwien> easiest would be to make a static linear ( p=p1*(1-w)+p2*w ) or logistic ( s=s1*(1-w)+s2*w ) mixer2009-11-23 18:55:02
<mike___> but i didn't succeed yet :)2009-11-23 18:55:27
<Shelwien> and try to find some cases where it would provide better results with w out of [0..1]2009-11-23 18:55:30
 of course, the output probability always has to be in the range - there's no way to use it for encoding otherwise2009-11-23 18:56:22
 but that linear mixing has a physical interpretation2009-11-23 18:56:58
 it means that we have a submodel #2 being applicable with probability w, and #1 with probability (1-w)2009-11-23 18:58:02
 in which case we have to merge the code space intervals corresponding to bits with different submodels2009-11-23 18:59:34
<mike___> sure, you can't use that interpretation with negative weights. but maybe there is another sound interpretation...2009-11-23 19:00:07
<Shelwien> well, its somewhat similar to inverted counters as i said2009-11-23 19:01:11
 consider a mix like p1*(1-2)+0.5*22009-11-23 19:01:30
*** mike___ has left the channel2009-11-23 19:26:08
*** mike___ has joined the channel2009-11-23 19:29:22
*** pinc has joined the channel2009-11-23 20:03:21
*** pinc has left the channel2009-11-23 20:37:30
*** pinc has joined the channel2009-11-23 20:37:48
*** pinc has left the channel2009-11-23 21:01:31
*** mike___ has left the channel2009-11-23 22:39:10
*** Spectator has joined the channel2009-11-23 23:54:16
*** Spectator has left the channel2009-11-23 23:55:24
*** no_one_here_again has joined the channel2009-11-24 02:53:23
*** no_one_here_again has left the channel2009-11-24 02:53:50
*** pinc has joined the channel2009-11-24 09:00:02
 http://status.dreamhost.com/2009-11-24 09:08:34
*** Shelwien has left the channel2009-11-24 09:26:58
*** Guest9968193 has joined the channel2009-11-24 09:27:02
*** Shelwien has left the channel2009-11-24 10:16:27
*** Guest9968193 has joined the channel2009-11-24 10:16:27
*** toffer has joined the channel2009-11-24 10:34:49
<toffer> hi2009-11-24 10:35:04
<Shelwien> hi2009-11-24 10:35:09
<toffer> today i finished my thesis with that presentation - now i'm finally officially an engineer :D2009-11-24 10:37:26
<Shelwien> congratulations ;)2009-11-24 10:37:26
<toffer> thanks alot2009-11-24 10:37:26
 pretty glad now2009-11-24 10:37:26
 and i got the full score2009-11-24 10:37:26
<Shelwien> cool2009-11-24 10:37:26
 i always was too lazy for that ;)2009-11-24 10:37:26
 and what's the specialty btw?2009-11-24 10:37:42
<toffer> i had pretty much effort there2009-11-24 10:37:42
 specialiy?2009-11-24 10:37:48
 speciality2009-11-24 10:37:52