Open in new window / Try shogun cloud
--- Log opened Thu Apr 12 00:00:19 2012
CIA-64shogun: Sergey Lisitsyn master * ra826101 / src/shogun/classifier/svm/LaRank.h : Got rid of larank deprecated warning -
-!- harshit_ [~harshit@] has quit [Quit: Leaving]00:13
-!- PhilTillet [] has joined #shogun00:39
PhilTilletHey :p00:40
-!- PhilTillet [] has quit [Ping timeout: 245 seconds]00:52
-!- PhilTillet [] has joined #shogun00:54
-!- PhilTillet [] has quit [Client Quit]00:55
-!- PhilTillet [] has joined #shogun00:55
-!- PhilTillet [] has quit [Ping timeout: 250 seconds]00:59
-!- av3ngr [av3ngr@nat/redhat/x-bhvghgippgadxuui] has joined #shogun01:05
-!- av3ngr [av3ngr@nat/redhat/x-bhvghgippgadxuui] has quit [Client Quit]01:06
-!- av3ngr [av3ngr@nat/redhat/x-dwqkcbcthexeyton] has joined #shogun01:06
-!- av3ngr [av3ngr@nat/redhat/x-dwqkcbcthexeyton] has quit [Remote host closed the connection]01:10
-!- PhilTillet [~Philippe@] has joined #shogun01:13
-!- PhilTillet [~Philippe@] has quit [Remote host closed the connection]01:35
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]01:51
-!- romi_ [~mizobe@] has joined #shogun02:15
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]02:57
-!- av3ngr [av3ngr@nat/redhat/x-ynyjtavjttwlbxfx] has joined #shogun02:59
-!- vikram360 [~vikram360@] has quit [Ping timeout: 245 seconds]03:05
-!- vikram360 [~vikram360@] has joined #shogun04:12
-!- vikram360 [~vikram360@] has quit [Ping timeout: 265 seconds]04:26
-!- vikram360 [~vikram360@] has joined #shogun05:33
@sonney2k*drum roll*06:08
@sonney2k9 slots06:08
-!- n4nd0 [] has joined #shogun06:23
-!- n4nd0 [] has quit [Remote host closed the connection]06:28
-!- n4nd0 [] has joined #shogun06:42
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]06:44
-!- vikram360 [~vikram360@] has joined #shogun06:47
n4nd0That's awesome!06:48
-!- n4nd0 [] has quit [Quit: AndroIRC - Android IRC Client ( )]07:02
-!- n4nd0 [] has joined #shogun07:02
-!- harshit_ [~harshit@] has joined #shogun07:08
harshit_n4nd0: Hola !07:12
n4nd0harshit_: ey! how is it going?07:16
harshit_hey, my TSVM is almost ready but the problem is it takes too much time to converge07:17
harshit_whereas in svmlin time it doesn't takes that much time07:17
harshit_dont knw whats going on!!07:18
n4nd0harshit_: aham, so there must something different between both implementations right=?07:18
n4nd0mmmm about time of convergence, have you checked if they use a kind of tolerance value07:19
harshit_yeah, but I havnt changed that part much in shogun. that was already in shogun07:19
harshit_tolerance value ??07:19
n4nd0yeah, it was just an idea ...07:19
harshit_how to check that..?07:19
n4nd0looking at their code07:19
n4nd0anyhow, are you confident then that it is not caused by the code you have written but by the other code?07:20
harshit_might be .! dont know exactly at this time .. have to debug a lil more07:21
harshit_n4nd0: Hows your covertree07:21
n4nd0harshit_: I have to discuss with blackburn some details before starting to do the port07:25
n4nd0but it looks promising, it is very fast07:25
harshit_great .. Maybe after completing on KNN we can apply it on K-Means ,07:25
harshit_what sya07:25
-!- blackburn [~qdrgsm@] has joined #shogun07:37
n4nd0blackburn: hey! have you seen the good news?!07:38
n4nd0blackburn: really good :D07:39
blackburnsure better than expectance07:39
harshit_n4nd0: whats the good news ?07:40
n4nd0harshit_: 9 slots!!07:42
harshit_dam awesome !07:43
-!- harshit_ [~harshit@] has quit [Ping timeout: 250 seconds]07:58
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]08:06
gsomixhi all08:16
@sonney2kwe are currently investigating if we can handle all 9 or just 8 (as we requested)08:16
@sonney2kit is a huuuge number08:16
gsomixand good morning08:16
n4nd0gsomix: good morning08:16
n4nd0sonney2k: really, could it be a problem with the number of mentors?08:16
n4nd0gsomix: yeah08:19
gsomixtime to vodka08:19
n4nd0c'mon isn't it like 10 around there? :P08:19
@sonney2kn4nd0, yes08:24
n4nd0that's not good08:24
@sonney2knot enough mentors08:24
@sonney2kwe had more ideas but for some ideas we did not get good proposals08:24
@sonney2kso these mentors won't mentor anything08:24
@sonney2kso in fact we are 7-8 mentors08:25
@sonney2khave to leave train08:25
n4nd0tty later08:25
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun08:35
* gsomix going to classes about computer networks.08:46
gsomix*is going08:46
sonne|workn4nd0: which timezone are you in?09:08
sonne|workgsomix: which timezone are you in?09:08
sonne|workblackburn: timezone?09:08
sonne|workI need to collect to know when you all are online...09:08
blackburnsonne|work: both me and gsomix are in +409:09
n4nd0sonne|work: CEST (UTC+2)09:09
n4nd0sonne|work: I think we are in the same09:10
sonne|workn4nd0: yes indeed09:10
sonne|workdoes anyone know which timezone pluskid is in?09:10
sonne|workand heiko?09:10
blackburnI guess +1?09:11
n4nd0btw, if someone had any doubt, apart from being pretty fast, JL's cover tree gives correct results :)09:11
sonne|workn4nd0: so then it is even more clear that we want it :)09:11
n4nd0sonne|work: I'm working on that ;)09:12
sonne|workblackburn: wiking?09:12
sonne|workand jacob?09:13
blackburnsonne|work: belgium, same?09:13
blackburnno idea about jacob09:13
sonne|workharshit? and philippe?09:13
blackburnharshit +5 or +609:13
n4nd0sonne|work: philippe is in France so I guess he has the same09:13
blackburnphilippe france09:13
sonne|worklets better make sure when they pop up09:15
-!- av3ngr [av3ngr@nat/redhat/x-ynyjtavjttwlbxfx] has quit [Ping timeout: 264 seconds]09:40
-!- wiking [] has joined #shogun09:42
-!- wiking [] has quit [Changing host]09:43
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:43
sonne|workwiking: time zone?09:44
sonne|workthat is UTC+2 right?09:44
sonne|work9:45hrs now correct?09:45
blackburnseems you guys are having shortest word competition09:47
sonne|workI've updated the website
sonne|worknot sure if a github banner is a good idea but hey I added it for now10:19
sonne|workalso removed broken links to tuebingen mpg10:19
blackburnyes it is10:19
sonne|workahh but I got the link wrong10:20
blackburnsonne|work: we need new description I think10:20
sonne|workafter gsoc :)10:20
sonne|workand volunteers welcome!10:20
n4nd0what description are you talking about guys?10:21
blackburnn4nd0: shogun is blablabla10:21
n4nd0the paragraph on top of the page?10:21
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:43
-!- pluskid [] has joined #shogun10:48
n4nd0what about int and float in JL's code, should they go to int32_t and float64_t, respectively?10:48
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:49
blackburnsonne|work: it seems jacob is in eastern usa - stands for michigan state univ10:50
sonne|workno overlap in this set I guess then10:51
sonne|worknot good10:51
blackburnsonne|work: may be some saturday evening would work10:51
blackburnhowever it is saturday heh10:52
pluskidcool to hear good news about 9 slots!10:52
sonne|workpluskid: you were UTC+8 right?10:52
pluskidGMT+8, is it the same as UTC+8?10:53
pluskidI'll google it10:53
pluskidsonne|work: yes, UTC+810:53
pluskidthe homepage github link point to , is it intended?10:56
pluskidand the "you" actually has 16 followers10:57
pluskidthough no public repo10:57
pluskidPreviously I thought one mentor can be on multiple project, am I wrong?10:59
blackburnyou mean one person?11:00
blackburnyes that can be11:00
pluskidone person (as a mentor) can be in several projects(ideas), isn't it?11:01
pluskidI'm not sure11:01
sonne|workpluskid: yes11:01
sonne|workbut we try to restrict things11:01
sonne|workto be 1vs111:01
pluskidoh, that might be good for better quality11:02
sonne|workwhich means 7-8 slots is what we will actually use11:03
sonne|work(the downside of things)11:03
wikingblackburn: hey man i'm testing the multi class linear machine... and i'm not soooo satisfied with the speed :D11:07
blackburnwiking: which one?11:08
wikingclassification is super fast :)11:08
blackburnhow much time? d? N?11:08
blackburnif you do modelselection enable save train state option11:09
blackburnto make training a little faster11:09
-!- pluskid [] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]11:10
blackburnwiking: what is epsilon?11:13
blackburnuh that is tough11:13
blackburntry larger no need to 1e-511:13
blackburnyeah 1e-2 or 1e-311:14
wikingit's funny11:16
wikingi have a new kernel11:16
wikingbut it only works with libsvm11:16
wikingother svm die because it cannot find enough support vectors :(11:16
wikingbut with libsvm it works nice11:16
-!- pluskid [~chatzilla@] has joined #shogun11:17
blackburnwiking: is it proper normalized then?11:18
wikingblackburn: you mean the kernel? yes11:19
wikingany suggestion for metal earners?11:21
wikinghahahah metalearners  :)11:21
wikingstupid autocorrection11:21
blackburnmetal earner lol11:21
blackburnwiking: what kind of metalearner?11:22
wikingwell anything that does a consensus in a good way between various ML classifications11:22
wikingbecause i have some good results with different features.... separately11:22
wikingbut when i add them together (even with combined kernel) the result degrades :)11:23
wikingso it's pretty annoying :)11:24
sonne|workwiking: then your normalization sucks11:24
wikingsonne|work: hahahahaha any suggestions ? :)11:25
pluskidblackburn: I find why LinearRidgeRegression is acting weird :)11:25
blackburnpluskid: why?11:25
pluskida small bug11:25
pluskidthe matrices are not initialized11:25
pluskidsimple fix :p11:25
blackburnsonne|work: bad boy11:26
sonne|workwiking: well do you have the data or do I have it ...11:26
* sonne|work smacks sonney2k11:26
* sonne|work luckily I am at work11:26
pluskidbtw I have some suggestions about CMath::twonorm and several related functions11:30
pluskidthe template function is returning the same type as the argument11:30
pluskidfor example, when you calculate twonorm for an int32_t vector11:30
pluskidyou get int32_t result, which would loss a lot of information11:31
wikingpluskid: yeah11:31
wikingah ok11:31
pluskidI suggest we use float64_t or something for all input types11:31
wikingi think in case of qnorm is not like that11:31
pluskidhave to leave for a while11:32
wikingmmm wait i'm mixing now11:32
pluskidcu later11:32
wikingah yeah11:32
wikingqnorm is like that as well11:32
wikingshouldn't be it should be something like pluskid says11:32
wikingthere's anyways an explicit casting there11:34
wikingwhen u return float64_t as class T11:34
sonne|workwiking: in case when we compute a float64_t anyways yes11:35
wikingman it's early here in my head11:35
wikingi meant implicit casting..11:35
wikingbut anyhow yeah it should always return something like float64_t even if the input vector is an int32_t11:35
sonne|workwiking: there is always a trade-off11:41
sonne|workthis can be wanted or not11:41
sonne|workwhen numbers are small the pure int32_t based compuation will use MMX and be very fast11:41
sonne|workand correct11:41
sonne|workif numbers become too big one needs the float64_t based variant11:42
sonne|workcomputation is slow but this is required for large numbers11:42
sonne|workso YMMV11:42
pluskidin our current implementation, float64_t is always used in computaion11:46
pluskidjust when we are returnining, it gets casted to T11:46
pluskidI think the trade-off sonne|work mentioned is true11:46
pluskidbut what we are discussing here is not that issue11:46
pluskidwe are thinking about whether we should return T or float64_t ?11:47
-!- gsomix [~gsomix@] has quit [Ping timeout: 276 seconds]12:07
n4nd0have a GSoC applicant just arrived?12:12
n4nd0a new one I mean12:12
blackburnn4nd0: what?12:14
n4nd0mailing list12:14
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]12:22
-!- pluskid [~chatzilla@] has quit [Ping timeout: 246 seconds]12:36
-!- pluskid [~chatzilla@] has joined #shogun12:37
-!- n4nd0 [] has quit [Quit: Lost terminal]13:18
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]13:26
-!- vikram360 [~vikram360@] has joined #shogun13:27
-!- n4nd0 [] has joined #shogun13:28
-!- vikram360 [~vikram360@] has quit [Ping timeout: 272 seconds]13:36
-!- vikram360 [~vikram360@] has joined #shogun13:37
-!- gsomix [~gsomix@] has joined #shogun13:42
-!- gsomix [~gsomix@] has quit [Read error: Connection reset by peer]13:43
-!- genix [~gsomix@] has joined #shogun13:43
-!- genix is now known as gsomix13:45
gsomixIt's hot. (~o___o)~13:47
gsomixn4nd0, what about covertree?13:48
-!- PhilTillet [~Philippe@] has joined #shogun13:48
n4nd0gsomix: on progress13:48
gsomixsonne|work, what about slots?13:48
sonne|workgsomix: what about SG_PROGRESS :P13:48
PhilTillethey everybody13:49
sonne|workPhilTillet: what is your timezone?13:50
PhilTilletsonne|work, France, why? :p13:51
sonne|workthat's part of the decision process13:52
PhilTilletoki ^^13:52
PhilTilletHmm, I think it is called CET13:53
PhilTilletor GMT+213:54
sonne|workso it is 13:54 ?13:54
PhilTilletyou're welcome :)13:57
-!- PhilTillet [~Philippe@] has quit [Ping timeout: 246 seconds]14:29
wikingwhat's a good kernel for sparse features apart from linear?14:37
n4nd0well guys I have the first timing results of JL cover tree used from shogun!14:41
n4nd0someone expecting :P?14:42
pluskidshow the data :D14:43
n4nd0there is still some space for improvement14:44
n4nd0so it is promising14:44
pluskidtree construction plus querying14:44
n4nd0yeah, everything there14:44
pluskidsignificantly outperforms qsort14:44
pluskidcorrectness of the results validated?14:45
n4nd0not for this case I have timed14:45
pluskidanyway, it's super cool14:46
n4nd0I have validated them for the non-integrated version, they were the same14:46
n4nd0so I guess that they must remain, it is just the class used for the cover tree points what changes14:46
pluskidI think I should suggest some guys in my lab to try out cover tree in their projects :D14:46
n4nd0I have still some work left to do14:47
n4nd0I wanted to see some results fast14:47
n4nd0pluskid: I wanted to ask you by the way14:47
n4nd0for you accuracy evaluation for KNN14:47
pluskidwhat do you mean?14:48
n4nd0I think I could use it to check that everything is alright using the cover tree, right?14:48
n4nd0let me see, maybe I am not using the right words14:48
pluskidI can't quite catch your meaning14:48
n4nd0I don't really remember the exact words14:48
n4nd0the first patch you did for shogun14:48
pluskidaccuracy evaluation?14:49
n4nd0then it is what I said ;)14:49
n4nd0is that something that can be used to check if the results given by KNN are correct?14:49
pluskidwhy not simply comparing the two results?14:50
pluskidcovertree and qsort14:50
n4nd0yeah sure14:50
n4nd0that's the way I do it14:50
n4nd0I just wondered if your patch was something "more profesional" to do this14:50
pluskidif they are the same, the algorithm should be correct14:50
pluskidI don't think so14:50
pluskidmy patch is for clustering14:50
pluskidhere is classification14:50
n4nd0ok, my bad :S14:50
pluskidI think if two algorithms are the same, they should give the same accuracy14:51
pluskidbut when they give the same accuracy, they are not necessarily the same :p14:51
sonne|workn4nd0: buy you can of course use any of the other evaluation modules that where made for clustering :D14:51
n4nd0sonne|work: what do you think about the result?14:51
sonne|workvery lovely :)14:52
n4nd0sonne|work: but we can make it better14:52
sonne|workthat is factor 4 now right?14:52
n4nd0sonne|work: yeah14:52
sonne|workhmmhh wasn't knn even using multiple threads?14:52
n4nd0I have not configured anything for threads14:53
n4nd0so I think it is just with one14:53
n4nd0the cover tree admits an heuristic to compute the distance14:54
n4nd0there is an upper bound14:54
n4nd0while computing the distance, if the intermediate result is already higher than this upper bound, one can cut the comptuations14:54
n4nd0computations, sorry14:54
n4nd0right now I am not doing that, I am just using the method distance in CDistance14:55
n4nd0sonne|work: what about including in CDistance this distance computation with the upper bound?14:55
sonne|workn4nd0: feel free...14:55
n4nd0sonne|work: cool14:55
n4nd0I have still some time to have fun with this :D14:56
n4nd0I will keep you update14:56
sonne|workit is actually not so easy14:56
n4nd0sonne|work: why?14:56
sonne|workyou would need to modify *every* distance function14:56
n4nd0my idea was14:56
n4nd0just to define it in Euclidian14:56
n4nd0do people use KNN with other than that?14:56
sonne|workto stop doing computations once a certain distance is reached14:57
sonne|workbut agreed euclidian is most important14:57
n4nd0I can include the heuristic in Euclidian for the moment14:58
n4nd0for the others it would work just fine14:58
-!- PhilTillet [] has joined #shogun15:01
-!- vikram360 [~vikram360@] has quit [Ping timeout: 245 seconds]15:14
-!- harshit_ [~harshit@] has joined #shogun15:19
n4nd0see you later people15:29
-!- n4nd0 [] has quit [Quit: leaving]15:31
gsomixsonne|work, i have just returned from uni.15:38
gsomixbut... done15:38
gsomixsonne|work, so what about slots?15:38
pluskidbrowse irc log ;)15:39
sonne|workgsomix: ahh 9 but we will return 115:41
sonne|workso 815:41
sonne|workwhich is still *a lot*15:41
gsomixsonne|work, is it the final decision? or should google make sure that shogun could have 8 students?15:45
sonne|workI don't understand15:45
harshit_sonne|work: Do you remember the wrapper code for SVMlin.cpp.15:46
sonne|workharshit_: hardly15:46
sonne|workbtw which timezone are you in?15:46
harshit_Actually I am making its TSVM available in shogun,I have dont most of the part , but optimization is really slow15:46
gsomixsonne|work, awww, I'm stupid. Never mind, I already figured out.15:48
sonne|workharshit_: timezone?15:49
-!- blackburn [~qdrgsm@] has joined #shogun15:51
CIA-64shogun: Jacob Walker master * rb98500e / src/shogun/regression/GaussianProcessRegression.cpp : Hopefully fixed strange memory bug in CGaussianProcessRegression::GetCovarianceMatrix -
CIA-64shogun: Sergey Lisitsyn master * rb95b089 / src/shogun/regression/GaussianProcessRegression.cpp : Merge branch 'master' of git:// -
harshit_about +5:3015:55
harshit_btw congrats on getting your requested no of slots, perhaps even more :)15:55
sonne|workharshit_: thx15:56
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120314124128]]15:56
sonne|workdoes any one of you blog?15:58
-!- pluskid [~pluskid@] has joined #shogun15:59
-!- pluskid [~pluskid@] has quit [Client Quit]16:00
gsomixcu, guys16:01
blackburnI do not, only twitter16:01
-!- pluskid [~pluskid@] has joined #shogun16:01
gsomixsonne|work, too only twitter16:02
harshit_sonne|work: not yet.16:03
wikingblackburn: what's your twitter acc16:04
blackburnwiking: @qdrgsm16:05
wikingsonne|work: i do but it's really unrelated to what i do :P16:05
shogun-buildbotbuild #644 of r_static is complete: Success [build successful]  Build details are at
wikingblackburn: will be funny to read your twits :)16:05
blackburnoh I twit some shit :D16:06
wikingyeah i can see that16:06
wikingbuild success? :))16:07
-!- romi_ [~mizobe@] has quit [Remote host closed the connection]16:07
wikingwho uses shogun from r? :)16:09
wikingi mean i dislike the whole R as it is so ... ;)16:09
shogun-buildbotbuild #666 of cmdline_static is complete: Success [build successful]  Build details are at
wikingany ideas who could i debug16:10
wikingwhy i don't get support vectors for certain svm-s?16:10
wikingwith my kernel?16:10
wikinggnmpsvm for that matter...16:11
gsomix>> build #666 >:]16:11
-!- pluskid [~pluskid@] has quit [Quit: leaving]16:12
-!- pluskid [~pluskid@] has joined #shogun16:12
sonne|workwiking: likely trivial solution is better...16:14
shogun-buildbotbuild #645 of octave_static is complete: Success [build successful]  Build details are at
harshit_Good news , TSVM is working,just need some modifications and examples and then people can use unlabeled data as well ,!!!16:15
wikingsonne|work: ?16:15
sonne|workcan happen for noise data that just outputting a constant is better16:15
wikingbut libsvm works fine that's the strange thing16:16
wikingwith same features and same kernel16:16
wikingeverything is the same just the optimizer is different16:17
shogun-buildbotbuild #632 of python_static is complete: Success [build successful]  Build details are at
sonne|workwait one is a true multiclass svm the other not16:20
sonne|workso it is a big difference16:20
harshit_blackburn: How can we print SG_SDEBUG() and SG_SINFO() messages in octave_modular16:20
harshit_thanks :)16:22
-!- harshit_ [~harshit@] has quit [Quit: Leaving]16:36
CIA-64shogun: Soeren Sonnenburg master * r2d7f35d / src/configure :16:38
CIA-64shogun: use ${SWIG} instead of explicit swig in configure16:38
CIA-64shogun: thanks Paul B?nau for the report -
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]16:51
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun17:04
@sonney2kwiking, you still havent' documented things
wikinghehehe i have a paper deadline tonight at 2am :)))17:08
@sonney2kbut time to hang around here I see. must be important ;-)17:08
-!- nickon [] has joined #shogun17:09
@sonney2kgsomix, what's up with the SG_PROGRESS patch?17:10
wikingwell i cannot make faster the cpus17:10
wikingso in the meanwhile it's calculating :)17:10
-!- vikram360 [~vikram360@] has joined #shogun17:10
wikingheheh yeah actually it's a cvpr paper should be important :>>.17:10
@sonney2kahh and PhilTillet, you did the nearest centroid right - so there are also some warnings in there
@sonney2kfor you17:11
@sonney2kwiking, I see - shouldn't then you *write* the paper?17:11
wikingheheheh it's done17:11
wikingneed to fill it out with stats17:11
wikingpull request sent17:14
PhilTilletsonney2k, true, i'll fix that asap17:15
blackburnsonney2k is in 'work, slaves!' mood today17:17
CIA-64shogun: Viktor Gal master * r9af8f41 / src/shogun/preprocessor/PNorm.h : Fix doxygen comment for getter/setter functions -
CIA-64shogun: Soeren Sonnenburg master * r87b031d / src/shogun/preprocessor/PNorm.h :17:19
CIA-64shogun: Merge pull request #443 from vigsterkr/master17:19
CIA-64shogun: Fix doxygen comment for PNorm -
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]17:47
-!- pluskid [~pluskid@] has quit [Read error: Connection reset by peer]18:02
gsomixsonney2k, I'm corrected. before I asked about the slots. :318:42
-!- blackburn1 [~qdrgsm@] has joined #shogun18:59
wikingthanks god for deadline extensions ! \o/19:01
@sonney2kmore time to waste on experiments19:02
@sonney2kwhy do they do this?19:02
@sonney2kgsomix, ?19:02
@sonney2kblackburn1, what does gsomix mean to say?19:02
wikingsonney2k: ;D19:02
blackburn1let me check logs19:03
@sonney2kblackburn1, regarding slavery - long abandoned :-D but all these things I keep repeating to do take like 5 minutes of work and I mentioned them a couple of times already - so I want them to be done.19:03
blackburn1sonney2k: oh you take these words seriously ;)19:04
-!- blackburn1 is now known as lisitsyn19:04
lisitsynok let it be so19:04
lisitsynwho stole my nickname19:05
lisitsynah long stayed web connection19:06
lisitsynsonney2k: ok he said he is corrected however no idea what does it mean :D19:07
lisitsynnevermind he just surprised with 9 slots19:07
@sonney2karent we all19:08
@sonney2kit is like x-mas, easter, birthday all on one day19:09
lisitsynshogun is a serious thing for you ;)19:09
@sonney2know I only hope 3 students stick with us after this gsoc19:10
gsomixsonney2k, awww. my English is very bad. I mean, that I corrected SG_PROGRESS patch.19:10
@sonney2kgsomix, ahhh!19:10
lisitsyneven I didn't get it19:10
gsomixI use too many passive voice. hehe :(19:11
@sonney2kso that we finally have some kind of developer community19:11
lisitsynsonney2k: have you seen page generator at github?19:11
lisitsynI can't use it while I am only pushermaan19:11
gsomixlisitsyn, where is blackburn?19:12
lisitsynnickname is in use by some hanged webchat connection19:12
-!- lisitsyn is now known as shogun-blackburn19:12
shogun-blackburndo you like it more?19:12
wikingsonney2k: the fork me label on the page was cool!19:12
-!- shogun-blackburn is now known as blackburn_19:13
gsomixshogun-blackburn, it sounds like bot.19:13
CIA-64shogun: Soeren Sonnenburg master * r19f81a5 / src/shogun/classifier/GaussianNaiveBayes.cpp :19:13
CIA-64shogun: Merge pull request #439 from gsomix/sgprogress19:13
CIA-64shogun: add SG_PROGRESS messages to Gaussian Naive Bayes -
blackburn_gsomix: LIKE A BOSS19:14
gsomixblackburn_, huh.19:14
@sonney2kblackburn_, you remain vodking19:14
blackburn_sonney2k: why should I ever stop19:14
blackburn_commit changes! like a boss! promote shogun! like a boss! github workflow! like a boss!19:15
blackburn_my own classes! like a boss!19:15
gsomixI heard word `vodka'.19:16
blackburn_gsomix: I want you to write an essay 'how my life changed when I start shoguning and vodking'19:18
gsomixblackburn_, okay, master. =___=19:18
@sonney2kblackburn_, hilarious19:18
PhilTilletlol the chan is going crazy :p19:25
@sonney2kI think the only sane participant is shogun-buildbot19:26
PhilTilletmaybe one day the buildbots will have enough artificial intelligence to get crazy aswell :D19:27
@sonney2klets hope they don't listen19:29
@sonney2kblackburn_, can we merge pluskid's LARS etc patch?19:30
blackburn_sonney2k: I didn't take a closer look yet but as it is already tested - probably we can19:33
blackburn_sonney2k: how would you like to solve Subset problem?19:35
blackburn_no typemap for index_t19:35
blackburn_so should be a typemap or it should be int32_t as it is but without index_t19:35
blackburn_ah you commented it already19:36
CIA-64shogun: Sergey Lisitsyn master * rba71ded / (2 files): Exposed Subset in modular #445 -
-!- harshit_ [~harshit@] has joined #shogun19:42
-!- blackburn_ [~qdrgsm@] has quit [Quit: Leaving.]19:48
-!- blackburn1 [~qdrgsm@] has joined #shogun19:52
@sonney2kindex_t is int32_t so it should just work20:07
blackburn1sonney2k: yes but [1,2,3] is probably not int32_t20:14
@sonney2kthat is why I cast things20:15
blackburn1sonney2k: have you seen 'refactoring' PR?20:16
@sonney2knot sure what to do with it20:17
blackburn1sonney2k: I have no words to comment it :D20:17
blackburn1sonney2k: there are some PRs sometimes you can stare and stare and stare and nothing to say20:18
@sonney2kparts are ok parts are not  - and it is soo huge that I don't know20:18
@sonney2kfor example I like the const declarations instead of #defines20:18
@sonney2kbut then it is too big to be merged at once20:19
blackburn1sonney2k: are you ok some code uses stl vector in LARS?20:28
blackburn1I am not against it while it can be easily refactored next day20:28
@sonney2kblackburn1, yes - just tell him20:34
-!- mohataher [c4cd9077@gateway/web/freenode/ip.] has joined #shogun20:34
mohataherhi, I applied for a GSoC, anywyas, I was working on implementing Convex Multi-class Feature Learning algorithm and I didn't really find any alternative functions in C++ equivalent to Diag(k) in matlab20:36
mohataherany suggestions ?20:36
mohataherif not, I can manage to implement and test it otherwise20:36
-!- n4nd0 [] has joined #shogun20:40
n4nd0hi all20:40
blackburn1mohataher: we have one in CMath afaik however it is not any complex20:41
blackburn1n4nd0: hey20:41
mohataherblackburn1 : I have just looked up the class, seems like I didn't find any functions equivalent to Diag(k), do you know the name of the function in Math.h ?20:51
blackburn1mohataher: create_diagonal_matrix20:52
mohataherIs it added recently ? my version doesn't have this function, I will update my version and get back to you if I needed. Thanks20:57
blackburn1mohataher: depends how old is your version ;)21:00
-!- mohataher [c4cd9077@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:10
gsomixbtw, congratulations on the day of space exploration.21:25
shogun-buildbotbuild #207 of nightly_none is complete: Success [build successful]  Build details are at
@sonney2kblackburn1, have you heard anything from jacob?21:58
gsomixshogun-buildbot, good work!21:58
shogun-buildbotWhat you say!21:58
blackburn1sonney2k: anything like?21:59
@sonney2kyou merged his patch but was he here in the channel?21:59
* sonney2k is trying to figure out the timezone he is in21:59
blackburn1sonney2k: yes21:59
blackburn1sonney2k: michigan21:59
@sonney2kblackburn1, what is his nick?22:00
blackburn1sonney2k: just puffin44422:02
blackburn1sonney2k: check 18-4822:02
@sonney2kso yesterday22:07
@sonney2kI was hoping he is here today22:07
@sonney2kblackburn1, btw I tried to return a slot but got an internal server error :D22:07
blackburn1sonney2k: but he fixed the error22:07
blackburn1however I hope it is not his last commit heh22:08
@sonney2koliver said that what he did was 10-20% of the gsoc work22:09
@sonney2kbut yes agreed22:09
-!- harshit_ [~harshit@] has quit [Ping timeout: 240 seconds]22:09
blackburn1sonney2k: 10-20%? seems to be small project22:12
@sonney2klike all of our projects ;-)22:12
n4nd0sonney2k: good night22:13
blackburn1sonney2k: no I am planning more that I can :(22:14
-!- nickon [] has quit [Quit: ( :: NoNameScript 4.22 :: )]22:15
-!- harshit_ [~harshit@] has joined #shogun22:22
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]22:25
-!- wiking [] has joined #shogun22:37
-!- wiking [] has quit [Changing host]22:37
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:37
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]22:37
-!- wiking [] has joined #shogun22:37
-!- wiking [] has quit [Changing host]22:38
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:38
wikingblackburn1: here?22:44
wikingso whoever made gmm is around?22:45
wikingAlesis Novik22:45
blackburn1no, he is doing phd in england and seems to be pretty busy22:45
wikingi was wondering if anybody would mind if it would be possible that u could preset the kmeans clusters22:45
wikingwhen you want to do gmm22:45
blackburn1kind of warm start?22:46
wikinge.g. u already have a preknowledge about the clusters.22:46
blackburn1I don't mind if you do that ;)22:46
wikingsince now i'm testing gmm22:47
wikingon a very small dataset22:47
wikingand it's horribly slow22:47
wikingand as a matter of fact i do have the cluster centers from kmeans...22:48
wikingso i thought maybe it'd be good to save some time..22:48
wikingbut first i'll add some debugging code to gmm22:48
wikingand dtrace it or something22:49
wikingto see where goes the most calc22:49
wikingmost of the cpu time is used by dlasr22:52
wikingwhich is basically max_likelihood calc22:56
wikingit's from lapack22:58
blackburn1yes I know that function23:00
blackburn1I mean I am surprised ;)23:00
-!- harshit_ [~harshit@] has quit [Quit: Leaving]23:18
gsomixgood night, guys23:21
n4nd0good night gsomix23:23
n4nd0nice video the one of the space by the way ;)23:23
-!- blackburn1 [~qdrgsm@] has quit [Ping timeout: 244 seconds]23:33
-!- blackburn [~qdrgsm@] has joined #shogun23:35
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 252 seconds]23:54
--- Log closed Fri Apr 13 00:00:19 2012