Open in new window / Try shogun cloud
--- Log opened Sat Apr 28 00:00:20 2012
--- Day changed Sat Apr 28 2012
@sonney2kgsomix, nite00:00
-!- karlnapf [] has joined #shogun00:16
-!- karlnapf [] has quit [Quit: Leaving.]01:27
-!- PhilTillet [] has left #shogun ["Leaving"]01:55
-!- pluskid [] has joined #shogun03:39
pluskida big tragedy installing 64bit system: I failed to backup my files, so lost many data, especially my various config files...04:45
pluskidspend a big amount of time to re-configure all those04:45
pluskidso a lightweighted SGVector and a heavy-weighted SGIVector?04:58
-!- av3ngr [] has joined #shogun06:31
pluskidI have some concerns of the new design, so sonney2k ping me when you come up06:59
pluskidor I can write to the mailing list later07:00
-!- av3ngr [] has quit [Quit: That's all folks!]08:36
-!- Marty28 [] has joined #shogun09:07
-!- Dark_Knight [~chatzilla@] has joined #shogun09:12
-!- Dark_Knight [~chatzilla@] has left #shogun []09:12
-!- Marty28 [] has quit [Quit: Colloquy for iPad -]09:17
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]09:53
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:53
-!- blackburn [~qdrgsm@] has joined #shogun11:03
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]11:49
-!- blackburn [~qdrgsm@] has joined #shogun11:51
-!- karlnapf [] has joined #shogun11:55
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]12:21
wikingtime for some stats12:45
blackburnwiking: benchmarks?12:47
-!- pluskid [] has quit [Ping timeout: 246 seconds]13:07
-!- pluskid [~pluskid@] has joined #shogun13:23
wikingfucking hell i hate dtrace now13:26
-!- karlnapf1 [] has joined #shogun13:31
-!- karlnapf [] has quit [Ping timeout: 276 seconds]13:32
-!- pausb [] has joined #shogun13:55
-!- karlnapf1 [] has quit [Quit: Leaving.]13:58
-!- pausb [] has left #shogun ["Leaving"]14:04
-!- PhilTillet [] has joined #shogun14:34
-!- emrecelikten [~Anubis@] has quit [Quit: Leaving.]15:05
pluskidthose registered by SG_ADD will automatically be saved in serialization, right?15:10
blackburnpluskid: exactly15:12
pluskidblackburn, thanks15:13
-!- PhilTillet [] has quit [Ping timeout: 256 seconds]15:18
CIA-64shogun: Sergey Lisitsyn master * rf2489bd / (17 files in 3 dirs): Extracted multitask things -
CIA-64shogun: Sergey Lisitsyn master * r801ce4d / (26 files in 6 dirs): Transfer learning restructuring -
CIA-64shogun: Sergey Lisitsyn master * r1927645 / (4 files in 2 dirs): Cleaned transfer learning stuff -
blackburnpluskid: looks ok to merge, should I?15:30
blackburnor anything you want to finish?15:30
pluskidblackburn: merge it at the moment please15:30
CIA-64shogun: Chiyuan Zhang master * rb4fbbdb / (4 files): remove get_strategy_type since we hope there would be no use of it. -
CIA-64shogun: Chiyuan Zhang master * r7830726 / (8 files in 2 dirs): Store number of classes in MulticlassStrategy -
CIA-64shogun: Chiyuan Zhang master * r022c18d / (11 files in 2 dirs): Use the remembered num_classes in a MulticlassStrategy -
CIA-64shogun: Chiyuan Zhang master * r1981a90 / src/shogun/multiclass/MulticlassStrategy.cpp : register m_num_classes of MulticlassStrategy. -
CIA-64shogun: Sergey Lisitsyn master * r3be395f / (13 files in 2 dirs): Merge branch 'multiclass-ecoc' of git:// -
CIA-64shogun: Sergey Lisitsyn master * rda6bddd / (3 files): Docu fixes -
shogun-buildbotbuild #790 of libshogun is complete: Failure [failed compile]  Build details are at  blamelist: blackburn91@gmail.com15:39
CIA-64shogun: Sergey Lisitsyn master * r2102799 / src/shogun/transfer/multitask/MultitaskKernelPlifNormalizer.h : Hopefully fixes compilation error of Plif multitask normalizer -
-!- emrecelikten [~emre@] has joined #shogun15:53
-!- gsomix [~gsomix@] has joined #shogun16:36
gsomixhome, sweet home :)16:36
shogun-buildbotbuild #792 of libshogun is complete: Success [build successful]  Build details are at
wikingxval is giving me a headache :)17:33
wikingERROR] assertion l failed in file kernel/Kernel.cpp line 10517:34
wikingException in thread "main" java.lang.RuntimeException: [ERROR] assertion l failed in file kernel/Kernel.cpp line 10517:34
wikingtime to debug then17:36
-!- n4nd0 [028900bc@gateway/web/freenode/ip.] has joined #shogun18:00
n4nd0I survived to my two exams during the last days, it feels good to be around here :)18:03
wikingn4nd0: congrats hope u have passed them as well ;)18:04
-!- pluskid [~pluskid@] has quit [Quit: Leaving]18:05
n4nd0wiking: we'll see, I am more or less confident of passing one of them but I am doubtful about the other :S18:05
blackburnn4nd0: how many to go?18:12
n4nd0blackburn: one on the 7th for the moment18:13
n4nd0yeah, these two last days were the tough ones18:13
n4nd0I am happy they are over18:13
blackburnyeah you should have accomplished mood :)18:14
-!- emrecelikten is now known as emre-away18:17
wikingblackburn: well yeah this crossval part is totally bogus...18:32
wikingcrashing like nothing18:32
blackburnwiking: huh18:34
blackburnwiking: you should contact heiko about it ;)18:34
wikingwell it either crashes my whole jvm or stops with a simple assertion error18:34
wikingbut there's no way i can make it work18:34
wikingso i'll use gdb now :)18:35
wiking"The Chair of cvpr-mcv has requested to publish Paper 23 Title: Multiple Kernel Learning Based Modality Classification for Medical Images to arXiv. Please log into CMT Author's Console to accept or decline this request."18:52
blackburnwiking: congrats18:52
wikingFUCKING HELL18:53
blackburnwiking: did you get good results with MKL?18:53
wikingpaper is rejected though18:54
wikingonly on arXiv18:54
wikingfuckin ghell18:56
wiking2 accepted the paper18:56
wikingone rejected18:56
blackburnwiking: editors?18:58
blackburnsame was for my jmlr mloss paper..18:58
blackburnyes sorry reviewers18:58
@sonney2kblackburn, do you know which concerns pluskid has?19:14
blackburnsonney2k: with?19:14
@sonney2kn4nd0, do you have any opinion on the new SGVector design19:14
@sonney2kblackburn, I guess SGVector...19:14
blackburnoh I planned to install opensuse right now19:14
blackburnok lets discuss it again19:15
blackburnsonney2k: I didn't ask him actually :(19:15
n4nd0sonney2k: probably I am not up to date with the very last discussion (I have not read logs from yesterday)19:15
blackburnsonney2k: do you use debian unstable with gnome 3?19:16
blackburnsonney2k: n4nd0: gsomix: lets discuss it right now?19:17
n4nd0sure! I am updating to last conversations right now19:18
* gsomix sits and listens to elders. :)19:19
n4nd0blackburn: so what is the main advantage of using SGVector and SGIVector?19:21
blackburnn4nd0: if I knew19:21
n4nd0is it to have something rather light-weighted for simple datatypes like int and double?19:21
blackburnn4nd0: it looks like we are probably unhappy with current memory handling19:22
n4nd0I like the idea of being able to make vectors of anything19:22
blackburnno we don't need vectors of anything19:22
-!- harshit_ [~harshit@] has joined #shogun19:22
n4nd0e.g. instead of CMachine* -> vector< CMachine >19:22
blackburnno no SGVector is only for numeric stuff19:23
n4nd0but the other one, SGIVector would be for the rest right?19:23
blackburnno, it is for vectors those are shared between different objects19:23
blackburnso it has refcounting19:23
blackburnwiking: you need somebody?19:24
wiking"As of now, your paper is a weak rejection (the only paper with conflict reviews). However, if you think you can address the negative issues being raised by reviewer. I can give you a rebuttal chance."19:25
wikingI NEED HEEELP!19:25
blackburnnot just anybody?19:25
gsomixblackburn, too beatle.19:25
blackburnwiking: I'd be happy to help to improve it19:25
@sonney2kwiking, sounds good19:26
wikingsonney2k: HEEELP!!!19:26
blackburnheh my help is probably not needed :D19:26
@sonney2kwiking, ?19:27
@sonney2kn4nd0, no19:27
wikingblackburn: anybody's good19:27
wikingbut sonney2k is experienced in this matter afaik19:27
blackburnwiking: sure19:27
@sonney2kn4nd0, it is all only for numerical stuff ... let me repeat what I've said somewhen yesterday19:28
@sonney2kn4nd0, SGVector would be just the double* ptr and int for length19:28
@sonney2knothing more -> no overhead19:28
@sonney2kwill only be used if the vector is just to be used in this function and no more19:29
@sonney2knot in member variables not for interface exchange stuff19:30
gsomixsonney2k, what is strategy to ref counting in SGIVector?19:30
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]19:30
blackburnokay I am totally lost why do we need it19:31
@sonney2kn4nd0, SGIVector wold be the one with ref counting19:31
@sonney2kthat's it19:31
@sonney2kgsomix, like with SGVector19:31
@sonney2kgsomix, err SGObject19:31
gsomixsonney2k, mmm, and I'm asking again. do we plan to work with pointers to SGIVector?19:32
@sonney2kgsomix, no19:33
gsomixsonney2k, copies?19:33
wikingblackburn: sent19:34
blackburnwiking: got desperated? :D19:34
wikingblackburn: well i'm appreciating anybody's help!19:35
@sonney2kyes - but I just realize that this then would need some different solution for refcounts19:36
@sonney2k(due to copies)19:36
blackburnsonney2k: yes count should be stored somewhere19:36
gsomixsonney2k, like pluskid solution?19:37
@sonney2kgsomix, but w/o the hack19:46
-!- PhilTillet [] has joined #shogun19:47
@sonney2kgsomix, so e.g. allocate some memory for the refcount (so have a int32_t* in SGIVector pointing to the counter - this ptr is shared between SGIVectors and memory freed if *ptr == 0)19:49
gsomixsonney2k, aha. It's clear for me.19:50
n4nd0what would be the relation between SGVector and SGIVector?19:51
n4nd0I see there should be inheritance there but I think I am still missing something19:51
@sonney2kn4nd0, just inheritance to not write the code that is in SGVector twice19:52
n4nd0ok so SGVector <--- SGIVector19:53
@sonney2kor other way round :)19:53
n4nd0how can it be that it does not matter?19:53
n4nd0the idea is to have the T* and the int for the length in SGVector19:56
n4nd0in SGIVector add the ref counting stuff19:56
n4nd0SGIVector inheiths fom SGVector19:56
n4nd0is that right?19:56
@sonney2kn4nd0, exactly like that19:57
@sonney2kn4nd0, so any thoughts / comments?20:00
@sonney2kwiking, I still don't understand what you need? don't you have a couple of senior authors?20:01
wikingsonney2k: hahahah i wish i would20:02
@sonney2kwiking, sole author or what?20:02
wikingwell when shit happens...20:02
@sonney2kare you crazy?20:02
wikingit's not me who is crazy20:02
wikingor? :)20:03
wikingi mean yes it is fucking crazy that i'm doing this alone20:03
wikingbut this is the truth20:03
n4nd0sonney2k: yeah, so that we said is the how to do it, but why do we want to do it? I just come up to reduce some memory usage. SGVector would occupy less than SGIVector20:04
@sonney2kI think I don't know any phd student who publishes w/o a senior author20:05
wikingsonney2k: yep, i know but since my supervisors are basically not existing... what can i do?20:06
blackburnsonney2k: I'll do it solely too probably if I do my phd here20:06
blackburnwiking: do you have only virtual supervisor?20:07
wikingblackburn: actually 2 of them!20:07
blackburn2 virtuals?20:07
wikingneither of them does anything20:07
n4nd0blackburn: you should go outside for your PhD and discover new world!20:09
@sonney2kn4nd0, exactly that is the reason - SGVector has a lot less overhead, so for example when using 2d byte vectors you will not have the overhead memory and copy wise20:10
@sonney2kn4nd0, for example if you compute kernel matrix and get ptr's to feature vectors you don't need the ref/unref20:10
@sonney2kwhich would make everything twice as costly20:11
@sonney2kwiking, I always had virtual and real supervisors20:11
@sonney2kthe real one is the GUN part of shogun :)20:11
n4nd0sonney2k: I must check how this ref counting is done in SGObject to understand better that there is a real overhead20:12
wikingsonney2k: in my case they are having virtual String review (String paper) =0;20:12
@sonney2kn4nd0, you need the refcount and you need to in/decref it and do a check whether to free the vector20:13
@sonney2kwiking, I guess none of the papers I wrote when I started would have been accepted w/o help from 'senior' authors20:14
wikingsonney2k: i have 2 accepted now 2 months ago20:15
wikingboth of them written like this one :>20:15
@sonney2kwhich conf?20:15
wikingand the other is a minor one20:15
@sonney2kpakdd == kdd?20:18
wikingnono it's the pacific-asia kdd20:18
@sonney2kif not I don't know this one...20:18
@sonney2kahh ok20:18
@sonney2kwiking, it gets exponentially harder with confs like kdd/icml/nips20:19
wikingso it's not as big as kdd but it's ok20:19
wikingsonney2k: i know20:19
wikingsonney2k: this one where i've got 2 vs 1 was cvpr20:19
@sonney2kcvpr is top tier or?20:20
wikingbut it's just a workshop20:20
wikingnot the conference20:20
wikingbut then again it gets indexed the same way as cvpr :PP20:21
wikingbut yes as u said i'm very afraid that i won't be able to get any paper accepted like this on a better conference... and then i'm basically fucked with my whole phd ;P20:23
wikingsonney2k: but now that u've told me this it's just reassured me that it won't work out :D yeeeey \o/20:29
-!- Priyans [~Priyans@] has joined #shogun20:31
-!- n4nd0 [028900bc@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:35
@sonney2kwiking, don't get me wrong but there are always virtual supervisors but you need to find real ones to have fun / a chance to compete20:53
wikingsonney2k: yep i know that's why i messaged u the other day  ;)21:00
blackburnhah sonney2k congrats with becoming a supervisor ;)21:01
-!- PhilTillet [] has quit [Ping timeout: 245 seconds]21:02
blackburnsonney2k: with addition to enfant terrible like me :D21:02
blackburnwiking: what is the deadline to resubmit?21:04
wiking36 hours :)21:04
blackburnwiking: oh that's nice21:06
-!- shogun-buildbot [] has quit [Ping timeout: 276 seconds]21:07
wikingheheh yeah i have now a nice marathon :>21:07
wikinghe left us21:08
blackburnI meant the same21:08
blackburnwiking: was my comments any useful?21:10
wikingyeah of course!21:10
wikingi'm now collecting comments and thinking of how it could be done better21:10
wikingand i'll give it a go21:10
-!- shogun-buildbot [] has joined #shogun21:10
wikingworst case they reject it for sure :)21:10
blackburnagain, I think you should stress the fact you work with features of different nature21:11
wikingi guess i have to lament about it more :)21:12
blackburnwiking: I really had the same problem with my paper about road signs recognition21:12
blackburnthey are pretty similar actually21:12
blackburnhomogay map again21:12
blackburnand no strong evidence of novelty :(21:13
wikingweeee SUCK21:13
blackburnhowever my paper goes to very low level *russian* journal21:13
blackburnits goal is to strength up my bachelors work and nothing more :D21:14
wikingheheh yeah that's another story21:14
wikingi'm in a different position :S21:14
blackburnwiking: how long have you been as phd students?21:14
blackburnI see21:14
blackburnwere you at industry before?21:15
blackburnwhy did you leave it btw?21:15
wikingmmmm was very boring21:16
blackburnI'm leaving my kind of industry job in may21:16
wikingi've tried several places21:16
wikingthe problem is that it is great... for the first couple of months21:16
wikingand then it's becoming more and more dull21:16
wikingso i felt like becoming brain damaged... and there was a too much of a politics involved...21:17
wikingpolitics = management ...21:17
blackburnwiking: yeah I had one example of overmanagement recently21:18
blackburnit took a week to fix one line21:19
blackburnexactly one line21:19
wikingwith phd it's great that you are more independent... but the shit part is that is my case when u are completely independent21:19
blackburnfindById instead of findByPrimaryKey21:19
wikingheheh yeah i know how that feels21:19
wikingi mean when i started to write code generator for my tasks at the work21:19
blackburnwiking: yeah it shouldn't be like that..21:19
wikingi've realized that i have to quite21:20
wikingi mean there was not much brain involved in the stories21:20
wikingmore like hard core coding21:20
blackburnmy problem is that I do not code at all21:20
blackburnI have to fix some sql or some code but not to code something new21:21
blackburnthere is nothing to do left21:21
-!- PhilTillet [] has joined #shogun21:23
wikingfixing other people's code can really suck21:25
blackburnwiking: I will use libqp too21:30
-!- PhilTillet [] has quit [Remote host closed the connection]21:36
blackburnkernel mean matching namely21:36
-!- wiking_ [] has joined #shogun21:39
-!- wiking_ [] has quit [Changing host]21:39
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun21:39
wiking_asdf :> stack overflow21:42
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 276 seconds]21:43
-!- wiking_ is now known as wiking21:43
CIA-64shogun: Sergey Lisitsyn master * reab67ba / (4 files in 2 dirs): Moved pr_loqo -
CIA-64shogun: Sergey Lisitsyn master * r3208b54 / (5 files in 2 dirs): Moved ssl -
-!- gsomix [~gsomix@] has quit [Ping timeout: 246 seconds]21:56
-!- harshit_ [~harshit@] has joined #shogun22:04
CIA-64shogun: Sergey Lisitsyn master * r00f3b79 / (13 files in 2 dirs): Moved gpdt,gpm,gptdsolve -
-!- PhilTillet [~Philippe@2001:660:3203:402:84f7:1e57:be45:2237] has joined #shogun22:59
-!- Priyans [~Priyans@] has quit [Quit: Leaving]23:04
-!- shogun-buildbot [] has quit [Ping timeout: 250 seconds]23:05
-!- shogun-buildbot [] has joined #shogun23:06
-!- shogun-buildbot [] has quit [Ping timeout: 272 seconds]23:11
-!- shogun-buildbot [] has joined #shogun23:15
CIA-64shogun: Sergey Lisitsyn master * r667fcd5 / src/shogun/features/RejectionStrategy.h : Added experimental dixon Q test rejection strategy -
CIA-64shogun: Sergey Lisitsyn master * r686b412 / .gitignore : Updated git ignores -
-!- emre-away [~emre@] has quit [Ping timeout: 246 seconds]23:36
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]23:42
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]23:42
-!- PhilTillet [~Philippe@2001:660:3203:402:84f7:1e57:be45:2237] has quit [Ping timeout: 260 seconds]23:54
--- Log closed Sun Apr 29 00:00:21 2012