Open in new window / Try shogun cloud
--- Log opened Mon Oct 29 00:00:00 2012
--- Day changed Mon Oct 29 2012
blackburnyes00:00
wiking:P00:00
heikowiking, btw I just read the word unit test, whats the state of that?00:00
blackburnit have longed a month already00:00
blackburnhas longed :D00:00
blackburnI can haz00:00
@sonney2kblackburn, uhh ranking algorithms00:00
@sonney2kI've talked to the guy who did that at yahoo00:00
@sonney2kresearch00:00
blackburnsonney2k: I don't care what to do for that money00:01
blackburn:D00:01
wikingheiko: lol good grep. well the thing is that it's tricky. i have to add extra code into the makefile so that gmock and gtest are compiled with the same flags as shogun. and u need to provide the src of gmock/gtest to shogun to be able to do that00:01
@sonney2kblackburn, learning to rank is certainly interesting00:01
wikingheiko: lot of mindless ./configure and Makefile hacking00:01
blackburnif being serious it is researchy00:01
blackburnmy current job is like a factory worker job00:01
blackburnand twice less money00:01
@sonney2ktwice less == half ?00:02
wikingblackburn: clearcase!!!!00:02
blackburnhaha yes00:02
heikowiking, so one downloads gmock/gtest source, and provides path to shogun configure and then everything is done automatically?00:02
wikingblackburn: ain't no factory without clearcase :D00:02
@sonney2kwiking, what is clearcase?00:02
blackburnwiking: ain't no sunshine when clearcase00:02
heikowiking, maybe one could even add a script that downloads the google libs via wget or so?00:03
blackburnknow that song?00:03
blackburn:D00:03
blackburnsonney2k: cvs00:03
wikingheiko: then u compile those sources with the same flags as shogun and then u have your static libs that u link it to our unittesting objects00:03
wikingheiko: and then u are done :)00:03
wikingheiko: well actually standard distribs provide them00:03
heikook, wiking, does ubuntu?00:04
heikothere were these problems I remember00:04
wikingheiko: gtest yes00:04
wikingheiko: gmock no00:04
wiking:)00:04
heikok00:04
wikinggmock is still a precompiled library :P00:04
blackburnsonney2k: when one commits in git other guy in clearcase creates activity, check outs, ...... , commits, check ins, closes activity00:04
wikingbut in case of gtest it puts the source under /usr/src..00:04
heikoMaybe when you are done, you could write a small howto and put it in the test directory00:04
wikingsonney2k: that is clearcase ^ :D00:04
blackburnand during asks for proper rights, tries to grasp its interface and attempts to kill himself00:05
wikingheiko: aintnotime :D00:05
wikingheiko: will do after next week maybe00:05
heikowiking, no worries :) I know the problem00:05
heikowiking, what are you up to currenlty?P00:05
wikingheiko: paper00:05
wiking:>00:05
wikingand trying to get within the deadline :D00:05
@sonney2kheiko, the new doc should appear if tonights nightly_default builds00:06
heikosonney2k, cool thing! whats the link?00:06
@sonney2khttp://shogun-toolbox.org/doc/en/current/shogun_tutorial.pdf00:06
@sonney2kactually latest00:06
@sonney2knot current00:06
@sonney2kcurrent doesn't have that00:06
@sonney2kheiko, feel free to update the website!00:07
@sonney2kwith pointers...00:07
blackburncan't open sonney2k00:07
heikosonney2k, ok Ill have a try00:07
wikingmm china will dist-upgrade next month :)00:07
@sonney2kblackburn, heiko well *after* the build tonight00:07
blackburnahh00:08
heikosonney2k, sure00:08
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Quit: buildmaster reconfigured: bot disconnecting]00:08
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun00:08
heikobtw I just wrote a bit more on the MMD stuff00:08
heikowe should keep on doing this - very nice thing to have00:08
heikoblackburn, I also added a part that explains SHOGUNs structures, like Features, Kernels, etc00:08
blackburnheiko: nice00:09
heikoAnd I want to add a part about the model selection soon00:09
@sonney2kheiko, I can basically copy all kernels form my phd thesis00:09
@sonney2kso kernel section should be easy00:09
@sonney2kcoffin from some other paper00:09
heikoand part of my job will be documentation, so the GP part will be documented soon also00:09
@sonney2klets hope it gets accepted00:09
heikosonney2k, cool! add that00:09
@sonney2kbut heiko, last year only orgs of the size of KDE got in00:09
heikosonney2k, really? ok - but I mean we are growing and have quite a community00:10
heikoalso we need documentation more than anybody else00:10
heikoif a Desktop system is not self-explaining it has failed anyways00:10
wikingheiko: the 5 or 6 of us u mean? ;)00:10
heikowiking, :D00:10
@sonney2kshogun-buildbot, force build nightly_default00:10
shogun-buildbotThe build has been queued, I'll give a shout when it starts00:11
blackburnheiko: growing?00:11
blackburnwho's new? :D00:11
heikowiking, sonney2k, blackburn, actually, I do quite some PR for shogun around here, but people really often complain about documentation and that thats the reason they dont use it00:11
@sonney2kI can believe so00:11
heikoblackburn, the 2012 GSOC student for example00:11
blackburnheiko: wiking?00:12
heikoblackburn, yes for example :)00:12
@sonney2kif n4nd0 wasn't sleeping so much he would be here too I guess :D00:12
@sonney2kwhich reminds me00:12
@sonney2kI wanted to ask him to put his example in applications00:12
wiking?00:12
blackburnwiking: pass ;)00:12
wiking:D00:12
blackburnNOP00:13
wikingnopnop00:13
wiking:>00:13
heikospeaking of sleeping, I will go to bed now, see you all!00:13
wikingnite00:13
-!- heiko [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has left #shogun []00:14
blackburnsonney2k: wiking did I told you I am now mentoring at my current job?00:15
@sonney2kindeed seems like fish got used to the tank - so I rather sleep now too00:15
@sonney2kblackburn, heh yes00:15
blackburnsonney2k: apparently one my student disappeared00:15
blackburn:D00:15
@sonney2kblackburn, maybe you should not mentor shogun then ;-)00:16
wikingblackburn: noup, whom/what? :)00:16
@sonney2kwith these fish it is tough to check if they are all around - they are pretty 2d00:16
@sonney2kso from the wrong angle hard to recognize00:16
blackburnwiking: they gave me 2 students to mentor00:16
blackburnfunny that one (not disappeared one) is of same age as me00:16
blackburnbut knows no java no patterns nothing00:17
wikingblackburn: that happens...00:17
wiking"Little is known about his personal life, beyond a liking for basketball and, according to a leaked US diplomatic cable, Hollywood war movies, possibly acquired during a brief stay in Iowa when he was a young man."00:17
wikinghehehe00:17
wikingi bet this guy has all the chuck norris movies :D00:17
blackburnwiking: who it is about?00:17
wikingblackburn: Xi Jinping00:17
wikingthe distupgrade guy for china00:18
wiking:)00:18
wikingblackburn: don't u read other stuff than slashdot? :D00:18
blackburnahh I am not a big fan of china00:18
blackburnnorth korea is my love00:18
blackburnwiking: I don't00:18
wiking:>00:18
wikingblackburn: pravda? :)00:19
blackburnwiking: da00:19
blackburnwiking: how much time would you able to speak in russian with me? :D00:19
wikingnikoliko? :)00:20
wikingmislim malo bi mi razumeo ako bi hteo00:20
wiking:D00:20
blackburnwiking: no idea :D00:21
wikingali russi nekako ne razumeju sprski tako dobro ka srbi razumeju ruski00:21
wikinge pa tako00:21
wiking;D00:21
blackburnahh00:21
wikingNU PA GADJI!!!00:21
blackburnwiking: I understand kinda 70%00:21
blackburn:D00:21
wikinghehhehe00:21
wikinghttp://www.youtube.com/watch?v=5B9XYZqYuN800:22
wikingsome oldschool stuff :>00:22
blackburnhow russians do not understand srpska at all00:22
wikingu know nu pagadi?00:22
wikingblackburn: yep00:22
blackburnso do serbian00:22
wikingvidis razumes malo ;)00:22
blackburnwiking: come on how can I do not know nu pogodi00:22
wikinghehe00:24
wikingblackburn: well u've born after soviet union00:25
blackburnwiking: before!00:25
blackburn:D00:25
wiking?!00:25
wikingnoway00:25
blackburnmarch 199100:25
wikinggorbachov already fixed thing by then00:25
blackburnyes00:25
wikingso u were born in russia00:25
wikingnot in soviet union00:25
wiking:)00:25
blackburnI was born in soviet union00:26
blackburndo not mix things :D00:26
blackburnsoviet union disappeared in december00:26
blackburnI was born in march00:26
wikinghahaha00:26
wikingcouple of months00:26
wikingdoesn't count :P00:26
blackburncounts!00:26
blackburn:D00:26
wikingso yeah this is like true soviet propaganda cartoon00:27
wikinglove this shit :D00:27
blackburnhey no propaganda at all00:27
blackburnit is very kind00:27
wikingno fucking way: The latest episode was produced in 200600:28
wikingreally?!00:28
wikingi thought that it was only produced like in 1970's00:28
blackburnwiking: well these latest things are not that cool00:30
blackburnwiking: I'd say it is soviet propaganda as much as tom&jerry is american propaganda :)00:30
shogun-buildbotbuild #149 forced00:31
shogun-buildbotI'll give a shout when the build finishes00:31
wikingblackburn: well t&j is like full on american dream propaganda00:31
wiking:D00:31
blackburnwiking: okay then I'd say it contains much less propaganda00:32
blackburn:)00:32
blackburnI can't really see anything really soviet00:32
@sonney2knite00:42
blackburnsonney2k: good night00:43
blackburnwiking: have you ever heard john zorn's music?00:50
shogun-buildbotbuild #149 of nightly_default is complete: Failure [failed tutorial]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/14901:13
-!- blackburn [~blackburn@188.168.3.27] has quit [Quit: Leaving.]01:36
shogun-buildbotbuild #150 of nightly_default is complete: Failure [failed tutorial]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/15004:00
-!- ptizoom [~christian@79-71-87-88.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat]04:32
-!- ptizoom_ [~christian@79-71-87-88.dynamic.dsl.as9105.com] has joined #shogun04:32
-!- ptizoom_ [~christian@79-71-87-88.dynamic.dsl.as9105.com] has quit [Ping timeout: 268 seconds]05:02
-!- ptizoom_ [~christian@85.210.80.20] has joined #shogun05:15
-!- blackburn [~blackburn@83.234.54.215] has joined #shogun06:57
-!- blackburn [~blackburn@83.234.54.215] has quit [Quit: Leaving.]07:27
-!- ptizoom_ [~christian@85.210.80.20] has quit [Read error: Connection reset by peer]08:22
-!- ptizoom__ [~christian@85.210.80.20] has joined #shogun08:22
-!- ptizoom__ is now known as ptizoom08:53
-!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun09:19
-!- heiko [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has joined #shogun09:31
sonne|workheiko: seen that the pdflatex command failed?09:33
sonne|workwhat extra fonts do we need?09:33
heikosonne|work, havent seen, let me check09:34
heikosonne|work another question, do you know when approx. we will get notified whether we are invited to the doc summit?09:34
sonne|workin about a week09:34
heikosonne|work, is texlive-fonts-recommended installed?09:39
sonne|workno idea09:40
sonne|workcannot check from here09:40
heikosonne|work, I think thats it, at least in ubuntu09:41
heikosonne|work, about the doc summit, any guess about our chances? The thing is I would have to fly to Germany and apply for an Express Passport and thats expensive so I would only like to do it if needed. My NIPS travel funding application got neglected so if doc summit doesnt work then I dont need to get the quick passport09:44
sonne|workheiko: well I cannot say more - carol said they want to be quick10:05
sonne|workso hopefully this week10:05
-!- heiko [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has quit [Disconnected by services]10:34
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has joined #shogun10:35
-!- yoo__ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun10:37
-!- yoo__ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Ping timeout: 245 seconds]10:45
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun10:48
yoohi all10:48
sonne|workheiko1: I wonder why you didn't get selected for nips travel funding. having a paper there used to be enough...11:02
sonne|workhi yoo11:03
heiko1sonne|work - no idea :(11:14
sonne|workI remember having received this grant a couple of times11:15
sonne|workbut might be that I had an oral (spotlight!)11:15
sonne|workor some mixtures of papers, demo & workshop talks11:15
heiko1I submitted to a workshop, but these havent been decided yet11:19
heiko1sonne|work, I mean its fine, there will be other times11:19
heiko1also, once I am in a phd programe, it will be much easier11:20
heiko1especially here, people are quite well funded11:20
heiko1yoo, is the example that you used the same I sent in the pastebin?11:23
heiko1or did you change something,11:23
heiko1because it worked here11:24
yoono changes, but you have to change kernels ..11:24
yooCommwordstringker works fine11:24
yoothe prob I reported is with the weightedcommstringker11:24
heiko1ok11:25
heiko1parameters?11:25
yooin my example one of the two kernels was commented11:25
yoosame11:25
yoohttp://pastebin.com/zsKQYPhM11:26
sonne|workheiko1: well you have a nips paper w/o being a phd student. that is already quite something!11:26
heiko1sonne|work, I know - I would have loved to go there, well lets see11:28
heiko1the thing is: the doc summit is in parallel11:28
heiko1so there might be still a chance :) but I am not too optimistic about that11:29
heiko1yoo, confirmed, I get the same error11:29
heiko1yoo, I will have a quick look now11:29
yooone of the features get order = 0 and this cause the error11:30
heiko1yoo, looks like a X-Val based error to mee11:31
heiko1with subsets on the features not updating the order11:32
heiko1sonne|work, the get_order() of string features, what does it do? does it change when a subset is applied?11:32
sonne|workno it should stay fixed11:32
sonne|workheiko1: basically order == n (from n-gram length)11:33
sonne|workso substrings of length n are considered (but this is coded in the features already)11:34
heiko1how can this get zero then?11:34
heiko1in x-val11:34
yooyes it occurs only with xval11:35
yooweightedcommwordstringker manage several order right ?11:36
heiko1yoo, I actually have no idea about this kernel, never looked at it, sonne2y aka sonne|work is the specialist here11:36
heiko1yoo, but I think the bug comes from somewhere else11:36
heiko1let me investigate a bit further11:36
sonne|workyoo: yes spectrum kernel using all orders up to order11:37
heiko1sonne|work, wow didnt know that I thought this was called the blended spectrum kernel, good to know that this is there :)11:39
sonne|workeveryone re-invents the wheel so we have multiple names11:39
sonne|workheiko1: but yes that is the one. caution though it needs special args in the obtain_from_string / translate function (see example)11:40
heiko1sonne|work, I see, this is exactly where erics bug comes from: store_model_features calls copy_subset which copies a certain part of the data and creates a new stringfeatures instance, but does not call this obtain from char11:41
heiko1yoo, so problem solbed, but another one pops up ... :)11:45
sonne|workheiko1: hmmhh I don't know - this translation is only done once11:46
sonne|workso all subsets should be fine11:46
heiko1sonne|work, but this is different11:46
heiko1its store_model_features11:46
heiko1so it creates a copy of the features object where only a certain subset of the data is included11:46
sonne|workheiko1: yes but still11:46
heiko1for example, order is not even set11:47
heiko1in the new instance11:47
sonne|workit should then just copy order and everything and all good11:47
heiko1yes did, but now symbol_mask_table is missing11:47
heiko1so I was wondering whether there was a more systematic way of just copiying till no more assertion errors11:47
heiko1since I might miss something this way11:47
heiko1which chages results or so...11:47
heiko1what about an embed_features call?11:50
heiko1that seems to do it11:50
heiko1sonne|work, hope that is fine?11:50
sonne|workwhat would it do?11:51
heiko1I now even did differently11:52
heiko1result->order=order;11:52
heiko1result->compute_symbol_mask_table(result->alphabet->get_num_symbols());11:52
heiko1then, at least the code runs through11:52
heiko1but I dont know if I changed any semantics this way11:53
heiko1yoo, its fixed, submitting patch soon11:56
heiko1yoo, maybe you can have a look if the results make sense, I am not too sure whether this fix is correct in terms of that data is not changed or something11:57
yooheiko1: thx allright, I will test it. I am trying to figure out what was the pb, could you explain ?11:57
heiko1yoo, the problem was realted to cross.validation11:58
heiko1kernel machines have to store the data for the model somehow11:58
heiko1but since we are working on subsets, these got changed again when testing on different sets in x-val11:58
heiko1thats why they store copies of the data once they computed the alpha terms11:59
heiko1and this method wasnt keeping the order for string features11:59
heiko1Ill send you the patch, its just two lines11:59
yoookay11:59
heiko1yoo, pretty hard to find if you dont know the framework and how cross-validation in shogun works internally11:59
yoogood exercice then ..12:00
-shogungit:#shogun- [shogun] karlnapf opened pull request #809: fixes a bug (master...master) https://github.com/shogun-toolbox/shogun/pull/80912:01
heiko1yoo, https://github.com/shogun-toolbox/shogun/pull/80912:01
-shogungit:#shogun- [shogun] karlnapf pushed 3 new commits to master: https://github.com/shogun-toolbox/shogun/compare/915af5cda5ab...1e93f0dd1c1b12:01
-shogungit:#shogun- shogun/master 30d6b8a Heiko Strathmann: bugfix that allows using string features in cross-validation from store_model_features(). This method did not copy all attributed properly.12:01
-shogungit:#shogun- shogun/master debc3a4 Heiko Strathmann: replaced print by debug message12:01
-shogungit:#shogun- shogun/master 1e93f0d Heiko Strathmann: Merge pull request #809 from karlnapf/master...12:01
heiko1now the code runs with the other kernel, Ill also write to the mailing list in case somebody else had this12:01
heiko1yoo, whats your name to give bug-hunting credits in the Changelog? :)12:02
heiko1I just write Eric (yoo)12:03
-shogungit:#shogun- [shogun] karlnapf opened pull request #810: mention bugfix in NEWS (master...master) https://github.com/shogun-toolbox/shogun/pull/81012:04
-shogungit:#shogun- [shogun] karlnapf pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/1e93f0dd1c1b...91ea878b3d2012:04
-shogungit:#shogun- shogun/master 801071c Heiko Strathmann: mentioned last bugfix12:04
-shogungit:#shogun- shogun/master 91ea878 Heiko Strathmann: Merge pull request #810 from karlnapf/master...12:04
heiko1gotta go now, see you later!12:04
yoo++12:09
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed]12:09
-shogungit:#shogun- [shogun] vigsterkr opened pull request #811: Minor fixes in LatentSVM (master...master) https://github.com/shogun-toolbox/shogun/pull/81112:13
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun12:23
sonne|workwiking: you sure with that patch?13:08
-shogungit:#shogun- [shogun] sonney2k pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/91ea878b3d20...528ccfd8a5cf13:09
-shogungit:#shogun- shogun/master 77332c1 Viktor Gal: Minor fixes in LatentSVM13:09
-shogungit:#shogun- shogun/master 528ccfd Soeren Sonnenburg: Merge pull request #811 from vigsterkr/master...13:09
sonne|workok sure :D13:09
wikingyes13:17
wikingwhy?13:17
sonne|workwiking: never mind13:33
sonne|workX^{-1}^{-1}^{-1} problem13:34
blackburnsonne|work: that must be pretty unstable to compute such matrix :)13:35
sonne|workespecially when neuron spikes are creating the data13:35
blackburnsonne|work: is it possible to hold a code sprint at some conference?13:36
sonne|worknot officially and usually one is *very* busy at a conference13:36
blackburnsonne|work: I believe we need some experts advice or so13:37
sonne|workthen we better get the experts to work with us13:38
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has quit [Quit: Leaving.]13:40
blackburnsonne|work: yes, how?13:43
sonne|workask them :D13:43
blackburnsonne|work: so I write a letter to vapnik okay13:43
blackburn:D13:43
blackburnso he ask his students of students of students to use shogun13:44
sonne|workblackburn: he is retired :D13:44
blackburnsonne|work: no he keeps pushing some his stuff13:44
blackburnwith lectures etc13:44
blackburnsonne|work: just recently he was at MIT13:44
blackburnchiyuan posted a few photos13:45
sonne|workblackburn: where?\13:45
blackburnsonne|work: where what?13:45
sonne|workphotos13:45
blackburnsonne|work: http://instagram.com/p/QIFbhhzHQ1/13:46
blackburnsonne|work: http://instagram.com/p/QDZKOBzHZd/13:46
blackburnvlad is more than full of energy :D13:46
blackburnjust 76 years13:47
blackburnsonne|work: my supervisor was born just nn the same day our little war got over, very active still13:49
sonne|workwell yeah I've seen his talk a few times - he really is into his stuff still13:49
blackburn9 may 1945 :)13:49
blackburnsonne|work: you have seen that for real?13:49
blackburnI think I'd really like to listen to him13:50
wikingblackburn: http://videa.hu/videok/jarmuvek/szetrobbant-a-lada-megsem-halt-qu5Z2ey02o7lIBa213:56
-!- gsomix [~gsomix@188.168.13.104] has joined #shogun13:57
gsomixhi13:57
blackburnwiking: 2105 FTW14:01
wiking:)14:02
sonne|workblackburn: yes I have seen him talking about god einstein and himself :D14:16
sonne|workblackburn: and he was very upset when oliver said sth. like you don't need kernels :)14:17
blackburnsonne|work: don't need kernels for what?14:17
sonne|workSVMs14:17
sonne|workand no need to do anything in the dual14:18
sonne|workhaha14:18
blackburnhe meant linear primal svm?14:18
sonne|workbut olivier will mentor next year so we won ;)14:18
blackburnsonne|work: what he wants to mentor?14:19
sonne|workyeah. this guy http://www.kyb.mpg.de/fileadmin/user_upload/files/publications/attachments/primal_%5b0%5d.pdf14:19
sonne|workblackburn: well I talked to him - I told you that helps at times ;)14:19
blackburnah so chapelle14:20
blackburnhe is an author of some SVM IIRC?14:20
blackburnsonne|work: sometimes I feel SVM is stupid14:21
blackburn:D14:21
sonne|workreplace SVM with $METHOD14:21
blackburnsonne|work: we are all too tied to SVM14:21
blackburnsonne|work: it seems to be so bad we have no methods but SVM14:22
wikingyeah so heiko will make us GP classifiers :)14:22
blackburnwiking: they are not better I heard14:23
sonne|workblackburn: what is ``better'' then?14:23
wikingblackburn: complexity is high but it's autotuning14:23
blackburnsonne|work: no idea - accuracy-wise may be14:23
wikingso you dont really need to tune your parameters14:23
wikingbut fuckthisshit14:24
blackburnwiking: autotuning is interesting but performance?14:24
wikingwhy my code is not working argh14:24
blackburncan one do large scale?14:24
wikingas said earlier14:24
wikinghigh complexity14:24
blackburncomplexity? algorithmic or like in VC?14:24
wikingo(n^3)14:24
wikingif i remember right14:24
blackburnwiking: fuckthishit14:24
wikinghehehe14:25
wikingwell getsome/loose some14:25
wikingas usual14:25
blackburnand we came that everything is useless again14:25
blackburn:D14:25
wikingwell if u give a neural net enough data and time, it'll learn god of course :)14:25
sonne|workwiking, blackburn - matrix inversion!14:26
wikingwhere? :D14:26
sonne|work*EXPENSIVE*14:26
wikingis it coming :)14:26
sonne|workGPs14:26
blackburnsonne|work: n^3/3 at least yeah14:26
* wiking hides from matrix inversions :D14:26
sonne|workand as blackburn said - rock stable stuff14:26
sonne|workeven more stable than QP optimizers14:27
sonne|workhaha14:27
blackburnsonne|work: I am solving problems with det=1e-9 here come on14:27
blackburn:D14:27
wiking:>>>14:27
blackburnif we are talking about illposed problems dimension reduction that is14:27
sonne|workblackburn: at least positive :)14:27
blackburnsonne|work: not always, IEEE754 FTW14:28
blackburnheh14:28
blackburnI add 1e-9 to diag14:28
blackburnto avoid negative14:28
sonne|workeven the gaussian kernel can be negative w/ floating point numbers14:28
wikingi've heard that in new york they are planning to build a sandy-bridge14:41
wikinghehehehe14:41
wiking:)14:41
wikingi know it was a bad one... :P14:42
blackburnheh14:45
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has joined #shogun14:45
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun14:57
-!- gsomix [~gsomix@188.168.13.104] has quit [Ping timeout: 255 seconds]15:23
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has quit [Quit: Leaving.]15:28
-!- gsomix [~gsomix@188.168.13.104] has joined #shogun15:42
-!- gsomix [~gsomix@188.168.13.104] has quit [Ping timeout: 260 seconds]15:49
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has joined #shogun15:59
-!- gsomix [~gsomix@188.168.13.104] has joined #shogun16:21
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed]17:03
-!- gsomix [~gsomix@188.168.13.104] has quit [Quit: Leaving]17:41
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 268 seconds]18:18
wikingsonne|work, sonney2k around?18:20
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun18:47
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 252 seconds]18:58
-!- heiko1 [~heiko@host86-177-179-234.range86-177.btcentralplus.com] has left #shogun []19:28
@sonney2kwiking, ?20:07
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun20:07
-!- blackburn [~blackburn@188.168.13.104] has joined #shogun20:08
wikingsonney2k: i saw that yesterday you were talking about narray with heiko20:25
wikingsonney2k: i'm missing a resizeable matrix in shogun currently20:26
wikingso i was wondering if u guys came to a conclusion...20:26
@sonney2kshogun-buildbot, force build nightly_default20:31
shogun-buildbotbuild #151 forced20:31
shogun-buildbotI'll give a shout when the build finishes20:31
@sonney2kwiking, SGNDArray is what we talked about20:32
wikingsonney2k: and do i recall it right that u were working with dna sequences? :)20:32
@sonney2kwiking, an N-dimensional array20:32
@sonney2kwiking, what is a resizable matrix for you?20:32
@sonney2kwiking, yes but?20:33
wikingsonney2k: well i currently have this in the new sosvm20:33
wikingstd::vector<SGSparseVector<float64_t> >20:33
wikingi can resize both the vector and the inner sparsevector20:33
@sonney2kwiking, why not SGSparseMatrix?20:33
wikingwithout loosing the content of it20:33
@sonney2kwiking, well you could use SGSparseMatrix.20:34
wikingsonney2k: sgsparsematrix has no realloc option20:34
wikingor i didn't see and i was blind20:34
wikingno it doesn't have a line where it uses sg_realloc20:35
wikingnor a function call for resizing20:35
@sonney2kwiking, std::vector is no different20:35
wikingsonney2k: i can resize the vector w/o problems20:35
@sonney2kit is slow slow slow when resizing20:37
@sonney2kyou need to copy everything20:37
@sonney2kwhy not allocate an empty SGSparseMatrix with whatever max number of vectors you need?20:37
@sonney2kwiking, yeah but it same you could do with SGSparseMatrix20:37
wikinghttp://www.cplusplus.com/reference/stl/vector/resize/20:37
wikingsonney2k: you mean to add a new function to it? :)20:37
wikingand about the dna thing i was just looking for data set and thought you might have something or know somebody who would have something20:38
@sonney2kwiking, yes sure - it is not exactly magic to do so. but I would rather resize in big chunks - but no idea if the resizing is time critical in your case20:38
@sonney2kwiking, what exactly are you looking for?20:39
wikingno it's it's not time critical that much... i mean there are other parts of that sosvm that are way slower so this wouldn't really slow down the thing20:39
wikingsonney2k: motif finding20:39
@sonney2kwiking, ok - and you really need sparse not SGStringList right?20:43
wikingindeed20:43
wikingthe two things are not really related20:43
@sonney2kwiking, no I mean instead of your sparse vector / matrix list20:44
wikingwell i only need to be20:47
wikingto resize both horizontally and vertically the matrix20:48
wikingwithout all the time maintaining20:49
wikingthe values20:49
@sonney2kwiking, well SGStringList is the dense variant of SGSparseMatrix20:50
@sonney2kwiking, actually a list of vectors20:51
wikingyeah but it's the same deal20:52
wikingi cannot resize this dynamically20:53
@sonney2kwiking, well the same way20:53
@sonney2kanyways you could do a couple of things but they all take quite some time20:54
shogun-buildbotbuild #151 of nightly_default is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/15121:06
wikinghehe http://imgur.com/t5Qvz21:10
wiking:>21:10
wikingblackburn: ^21:11
blackburnwiking: :)21:13
@sonney2khurray21:21
@sonney2khttp://shogun-toolbox.org/doc/en/latest/shogun_tutorial.pdf21:21
wikinghahaha21:31
wikingwtf is this turbo boost ?21:31
wikingare we back to the 90s?21:31
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]21:36
blackburnwiking: turbo boost?21:48
wikingblackburn: intel's new shit :)21:48
wikingin the sandy/ivy-bridge arch21:48
-!- blackburn [~blackburn@188.168.13.104] has quit [Quit: Leaving.]22:27
-!- blackburn [~blackburn@188.168.13.104] has joined #shogun22:31
-!- ptizoom [~christian@85.210.80.20] has quit [Quit: Ex-Chat]23:26
-!- ptizoom_ [~christian@85.210.80.20] has joined #shogun23:26
--- Log closed Tue Oct 30 00:00:17 2012