Open in new window / Try shogun cloud
--- Log opened Sun Mar 20 00:00:38 2016
-!- timsl [54d8e30b@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:26
-!- shogun-notifier- [] has quit [Quit: transmission timeout]01:29
shogun-buildbotbuild #974 of nightly_all is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>, trevor <>, Wu Lin <>03:06
-!- HeikoS [] has joined #shogun03:09
-!- mode/#shogun [+o HeikoS] by ChanServ03:09
-!- HeikoS [] has quit [Client Quit]03:11
shogun-buildbotbuild #1101 of nightly_default is complete: Failure [failed test notebooks]  Build details are at  blamelist: Heiko Strathmann <>, trevor <>, Wu Lin <>03:25
-!- lambday [6a33e9dd@gateway/web/freenode/ip.] has joined #shogun06:10
-!- mode/#shogun [+o lambday] by ChanServ06:10
-!- sonne|osx [] has joined #shogun06:13
-!- lambday [6a33e9dd@gateway/web/freenode/ip.] has quit [Ping timeout: 252 seconds]06:25
-!- sonne|osx [] has quit [Quit: sonne|osx]06:45
-!- lambday [6a33f351@gateway/web/freenode/ip.] has joined #shogun07:50
-!- mode/#shogun [+o lambday] by ChanServ07:50
-!- besser82 [~besser82@fedora/besser82] has joined #shogun09:33
-!- mode/#shogun [+o besser82] by ChanServ09:33
-!- curiousguy13 [~curiousgu@] has joined #shogun10:00
-!- curiousguy13 [~curiousgu@] has quit [Ping timeout: 260 seconds]10:45
-!- sshkhr [7d1409e6@gateway/web/freenode/ip.] has joined #shogun10:51
-!- curiousguy13 [~curiousgu@] has joined #shogun10:58
sshkhrI have a small confusion in the Install wiki instructions. Anyone online to help me out ?11:04
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has joined #shogun11:18
sanujlambday: there?11:21
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 240 seconds]11:32
-!- mizari [] has quit [Read error: Connection reset by peer]11:48
-!- mizari [] has joined #shogun12:07
-!- lambday [6a33f351@gateway/web/freenode/ip.] has quit [Ping timeout: 252 seconds]12:11
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has quit [Ping timeout: 252 seconds]12:17
-!- sonne|osx [] has joined #shogun13:02
-!- sonne|osx [] has quit [Quit: sonne|osx]13:24
-!- Nebula_ [~chatzilla@] has joined #shogun14:03
-!- Nebula_ [~chatzilla@] has left #shogun []14:13
-!- sshkhr [7d1409e6@gateway/web/freenode/ip.] has quit [Quit: Page closed]14:17
-!- lambday [6a33f351@gateway/web/freenode/ip.] has joined #shogun14:35
-!- mode/#shogun [+o lambday] by ChanServ14:35
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has joined #shogun15:05
-!- mizari [] has quit [Ping timeout: 276 seconds]16:04
-!- mizari [] has joined #shogun16:10
-!- livus [~suvil@] has joined #shogun16:12
-!- mizari [] has quit [Ping timeout: 268 seconds]16:28
-!- mizari [] has joined #shogun16:32
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:48
-!- besser82 [~besser82@fedora/besser82] has joined #shogun17:30
-!- mode/#shogun [+o besser82] by ChanServ17:30
-!- HeikoS [] has joined #shogun17:51
-!- mode/#shogun [+o HeikoS] by ChanServ17:51
-!- witness_ [uid10044@gateway/web/] has joined #shogun17:55
-!- livus [~suvil@] has quit [Ping timeout: 240 seconds]17:57
@lambdayHeikoS: there?17:58
@HeikoSlambday: jojo17:58
@lambdayMMD more or less complete17:59
@lambdayHeikoS: will feature branch it from tomorrow17:59
@HeikoScompute_variance computes variance of the estimate?18:00
@HeikoSthat is, variance of the h-terms?18:00
@HeikoSwhere mmd itself is mean of the h-terms?18:01
@lambdayHeikoS: no it is variance estimation under null..18:01
-!- livus [~suvil@] has joined #shogun18:01
@lambdayHeikoS: it uses the variance of h-terms when the estimation method is PERMUTATION.. otherwise, WITHIN_BLOCK_DIRECT is somewhat different18:02
@HeikoSlambday: cool looks like all interfaces seem to wrk?18:02
@HeikoSok I see18:02
@lambdayHeikoS: almost18:02
@HeikoSlambday: is this stuff running on multiple threads already?18:02
@lambdayHeikoS: yep!18:02
@HeikoSlambday: cool stuff :)18:03
@HeikoSlambday: looks very clean to use the interface18:03
@lambdayHeikoS: will add gpu stuff when it's inside shogun18:03
@HeikoSlambday: yeah later is fine for that18:03
@HeikoSlambday: cool so I can start putting examples in soon18:03
@lambdayHeikoS: API-wise it is pretty similar.. but the internals are quite different18:03
@HeikoSI will do some high level experimental sanity checks once all this is in feature branch18:04
@HeikoSon the base of the existing notebook18:04
@lambdayHeikoS: absolutely..18:05
@lambdayHeikoS: I was thinking of adding all of these as it is inside the feature branch.. so that we can start writing API examples.. then I'll iteratively add proper licensing, documentation and unit-tests18:06
@HeikoSlambday: yes18:06
@HeikoSlambday: then we can parallelise the work a bit18:06
@lambdayyeah that's what I was thinking18:06
@HeikoSlambday: I will first write cookbook examples for all basic cases18:07
@HeikoSlambday: and then do some simple statistical tests to make sure these things behave as expected18:08
@lambdayHeikoS: yeah that sounds like a plan18:08
@lambdayHeikoS: BTW I'll be going home next week on Wednesday.. will come back on Tuesday.. hopefully I can still work while I'm at home..18:09
@lambdaylisitsyn: hola senor18:09
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has joined #shogun18:09
@HeikoSlisitsyn: yoz!18:09
@HeikoSlambday: ok cool18:09
lisitsynHeikoS: git-deploy is on hold as I need some help from wiking18:09
lisitsyn(no rights, as usual)18:10
@HeikoSlambday: good if the feature branch was running before you left ... so I can play18:10
@HeikoSlisitsyn: ah man :)18:10
@lambdayHeikoS: BTW I created an issue for linalg that we discussed.. would start working on that as well18:10
@HeikoSlambday: cool18:10
@HeikoSyeah just checking emails18:10
@lambdayHeikoS: yes I'll do most of it by today/tomorrow18:10
@HeikoSlambday: maybe create an entrance task if there are simple bits18:10
@lambdayHeikoS: first few bytes I'll write.. entrance tasks shouldn't be this much confusing :D I'll write once I refactor some of it, get the skeleton running18:11
@HeikoSlambday: I guess thats a good point18:12
@HeikoSlisitsyn: did you ahve a look on this?18:12
lisitsynHeikoS: this = ?18:12
@HeikoSlisitsyn: the linalg updates lambday thought of18:13
@lambdaylisitsyn: I guess he was talking about
@HeikoSlisitsyn: more clean distinction between library and hardware backend18:13
lisitsynwow that's big18:13
@lambdayread it, it's funny18:14
@HeikoSlambday: haha18:14
@lambdayspecially if you18:14
@lambday're following news18:14
@HeikoSjust reading it18:14
@HeikoSdeep drump ;)18:14
@lambdaytrumped :D18:14
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 268 seconds]18:14
@HeikoShaha man cracking here ;)18:16
@lambdayHeikoS: it was last night's beer talking maybe in the morning :D18:16
lisitsynlambday: ok I like the idea of no switch18:16
@lambdaylisitsyn: I hate that linalg::backend::fuckthisshit stuff18:16
@lambdaylisitsyn: check the API suggestions I proposed later as a comment18:17
lisitsynlambday: I have a suggestion for you how to develop the api18:17
@lambdaylisitsyn: sure man18:17
lisitsynit is - start porting some matrix intensive code18:17
lisitsynyou will get all the troubles18:17
@lambdaylisitsyn: could you elaborate a bit?18:18
lisitsynlambday: just take any svm whatever18:18
lisitsynand try to port it to linalg18:18
lisitsynif it doesn't work - patch linalg :)18:18
@lambdayyeah that sounds like a good plan18:18
@HeikoSjust got a good email for easy installation project18:18
@HeikoSreally want this project to happen ;)18:19
lisitsynlambday: I mean it would be really hard to not miss some detail18:19
@HeikoSlisitsyn: good point actually18:20
lisitsyntry porting that :D18:20
@lambdaylisitsyn: problem with this approach is, we already need to have a good enough stuffs inside linalg to port existing things.. but that should be the goal18:20
@HeikoSlambday: I thought this a while ago too ... rather than making thing broad, porting algorithm by algorithm might be a good idea18:20
lisitsynlambday: ah you can just skip what you've ported so far18:20
@HeikoSlambday: so going deep first might be good18:20
lisitsynthe thing is that mutates the framework18:20
lisitsynso it actually can do this thing18:21
@HeikoSlisitsyn: I guess it needs to be both18:21
lisitsynotherwise we will get all these problems later18:21
@HeikoSsince high level design is also important18:21
@HeikoSlisitsyn: ah man that LARS has cholesky updates implemented in it18:21
lisitsynHeikoS: lol18:21
@HeikoSthis is needed in like 10 places in shogun18:22
lisitsynwhy not write good ol' cholesky18:22
@HeikoSwhy is it in the f** lars class ;)18:22
lisitsynpoor french18:22
@lambdayholy shit man this is nightmare18:22
lisitsynok we need property based testing18:22
@lambdayI get scared every time I see cblas18:22
lisitsynotherwise you can't patch any function18:22
@HeikoSlambday: lets do the draft for linalg, once the basic changes are in18:23
@lambdaygood point18:23
@HeikoSwe should try to port whole algorithms18:23
@HeikoSrather than adding single functions18:23
@HeikoSbut rather try to really make use of it18:23
@HeikoSGP stuff is a good one for that18:23
@HeikoSand things like LARS18:23
@lambdayHeikoS: you mean, the entire thing should reside inside linalg?18:24
@HeikoSentire thing?18:24
@lambdaycholesky_delete, cholesky_insert18:24
@lambdayyeah these can be there18:24
@lambdaylisitsyn: dude you found some file to give an example18:25
@HeikoSlambday: oh yes18:25
@HeikoSlambday: cholesky updates are used everywhere18:25
@HeikoSmy python code is full of them18:25
@HeikoSI gotta run now guys, dinner is ready18:25
sanujHeikoS: wait :P18:26
@HeikoSsanuj: jo18:26
@lambdayHeikoS: I haven't yet cooked my dinner :( and it's 2300 in India18:26
@HeikoSmine is warmed up ;)18:26
@HeikoSfrom this morning18:26
@HeikoSreally gotta go18:26
@HeikoSsanuj: gotta talk later18:26
sanujHeikoS: hi, i wanted to ask....shall i make separate pages for svm in cookbook?18:26
@HeikoSIll check more emails tonight18:26
sanujlike linear, kernel etc18:26
@HeikoSsanuj: yes18:26
sanujHeikoS: cool18:27
-!- HeikoS [] has quit [Quit: Leaving.]18:27
lisitsynlambday: have you looked at torch design?18:27
@lambdaylisitsyn: nope.. what's torch?18:27
lisitsynlambday: ah btw don't we need some tensorflow compatibility nowadays?18:27
sanujlambday: dp framework in lua18:27
sanujit's good18:27
sanujdp - deep learning18:28
lisitsynhow's that dp ;)18:28
lisitsynjust kidding18:28
sanujlisitsyn: :P18:28
@lambdayyeah dp is misleading ;)18:28
@lambdayanyway what should I be looking for here?18:28
sanujlisitsyn: even i thought the same18:28
lisitsynlambday: well just general things maybe18:29
sanujlisitsyn: but then i used this for one of my projects18:29
lisitsynI mean I think they have spent some time designing this18:29
lisitsynsanuj: that's funny everybody has a deep learning framework now18:30
lisitsynoh from wikipedia I just got to know yandex uses torch18:30
sanujlisitsyn: yes, deep learning is very popular18:30
@lambdayit's funnier when they make the acronym out of (d)ee(p) as dp18:30
lisitsyngoing to tell all my colleagues we're using torch :P18:31
@lambdaylisitsyn: hahaha18:31
sanujmlpack is also going to add deep learning algorithms18:31
lisitsynI think we should either drop our nns18:31
lisitsynor no idea18:32
lisitsyndevelop it18:32
@lambdaywhat's nns18:32
sanujneural nets18:32
sanuji guess18:32
@lambdayI am really bad at acronyms, aren't I18:32
sanujno :)18:33
sanujthere are many libs for deep learning which are very good and popluar - caffe, torch, tensorflow, theano, keras, CNTK........18:34
sanujlisitsyn: can't we use one of these in shogun?18:34
sanujlike we use opencv18:34
sanujbut i think that's a bad idea18:34
lisitsynsanuj: I think we should have fancy models not available in all these18:35
lisitsynlike bayesian18:35
@lambdaysanuj: I don't think we're gonna use other libs for deep learning - we gotta develop one on our own18:35
lisitsynbut not sure we should even have convolutional nn18:35
lisitsynbecause we will never be as good18:35
sanujbut if we are competitive then people will use it18:36
lisitsynbut for bayesian stuff we should have all the basic stuff18:36
sanuji mean shogun users18:36
@lambdaylisitsyn: why won't we be ever as good?18:36
lisitsynlambday: they already put millions of hours there18:36
lisitsynI think a few people died there :D18:37
sanujyes and some of them support multiple gpu training18:37
@lambdaylisitsyn: okay so the problem is, we don't wanna die18:37
lisitsynpeople trained millions of models with caffe and stuff like that18:37
lisitsynwe'd have to go this way again18:38
sanujand we have limited resources18:38
lisitsyndoesn't sound possible for me18:38
sanujcompared to google and facebook :P18:38
lisitsynwe should take some short route18:38
@lambdaysanuj: you're forgetting baidu18:38
sanujlambday: oh yeah18:38
@lambdaylisitsyn: we don't wanna be excellent at a single thing, do we...18:39
lisitsynyeah but even to be competitive at neural networks we would need like 5 people working on that18:39
@lambdayamen to that18:40
@lambdaymaybe next year18:40
lisitsynbut I hope neural networks die next year18:40
@lambdaynah they are here to stay18:40
@lambdayI feel illiterate because I never worked on any18:40
lisitsynunfortunately hehe18:40
sanujeverything is getting deep in ml :P18:40
@lambdayall this SVM and feature selection and other convex/submodular optimizatons - that's all I've worked on :(18:41
lisitsynI know all the stuff but I have no time18:41
sanujlambday: there are very cool deep learning courses online ;)18:41
sanujif you want to learn18:42
@lambdaygotta get into an ML company.. this oracle is sucking my blood out for nothing18:42
@lambdaysanuj: I hardly get any time bro :(18:42
lisitsynlambday: well I don't get gps as well as I do neural networks18:42
sanujlambday: even i'm graduating in placed at samsung :(18:42
@lambdayafter months I've been active doing some good work :'(18:42
lisitsynbecause you have to think a bit when it comes to gps18:43
lisitsynneural networks are dumb as hell18:43
@lambdaysanuj: congratulations18:43
@lambdaylisitsyn: really18:43
@lambdaythat's another thing I don't know18:43
sanujlambday: not sure what to do :P18:43
sanujfeel like doing research18:43
@lambdaygot the book, got the lecture serieses, but hardly get time18:44
@lambdaysanuj: good idea.. I'm leaving my job18:44
sanujlambday: haha18:45
@lambdaysanuj: gotta get another one for now cause I'm getting married.. after that, I'm out!18:45
sanujlambday: oh good luck ;)18:45
sanujand congratulations :)18:45
lisitsynlambday: are you going to have that fancy thing with one thousand people?18:45
@lambdaysanuj: thanks :D18:45
sanujlisitsyn: haha18:46
@lambdaylisitsyn: pretty close - I am expecting 600+ people at lesat - like one of those bollywood movies18:46
@lambdaywithing singing and dancing and colors18:46
@lambdayhaha yeah that's how we roll ;)18:46
@lambdaylisitsyn: I told HeikoS once about this 600+ people thing - he was shocked :D18:47
lisitsynhahah that's too funny18:47
lisitsynyou don't even know them, do you?18:47
@lambdaylisitsyn: never talked to >= 85% of them in my life18:47
@lambdayand in case I did, I don't remember18:48
sanujlisitsyn: it's complicated18:48
sanuji'm also from india :)18:48
lisitsynsanuj: yeah I know :)18:48
lisitsynyou're guys having fun there18:48
@lambdaylisitsyn: actually, over last few months, I started to kinda root for this large wedding thing - I guess that's the India in me talking :D18:49
lisitsynlambday: when is it happening?18:49
@lambdaylisitsyn: 22nd Jan 2017, if things go smooth18:50
lisitsynholy sheatz and you're already preparing?18:50
sanujlisitsyn: haha18:50
@lambdaylisitsyn: ...and we're late at that, as it turns out18:50
lisitsynwell I guess that caused by that 600+ count18:50
@lambdaythings get booked even before 1 year18:50
@lambdayand there are a lot of things to think of man, you won't understand18:51
@lambdaythings **have** to be the way the bride wants it18:51
sanujlisitsyn: there will be many more events before lambday gets married18:51
@lambdaysanuj: donald trump as US president18:51
lisitsynmake shogun work again pleaze18:52
sanujlambday: are u planning to move to USA?18:52
sanujlisitsyn: +118:52
@lambdayfor example18:52
@lambdaysanuj: it's tough, maybe I'll apply for PhD at some US schools but cannot settle there18:52
sanujlisitsyn: how difficult is that MCMC stan thing? assuming i don't know much about MCMC18:52
lisitsynsanuj: what thing exactly?18:53
sanujlambday: PhD is cool (y)18:53
lisitsynsanuj: the library itself?18:53
@lambdaysanuj: and tough! ;)18:53
sanujthis :)18:53
sanuji won't write a proposal for this18:54
lisitsynah ok18:54
@lambdaysanuj: you're from which college?18:54
sanujbut if i get selected for some other project, then i could learn MCMC also18:54
sanujlambday: IIT guwahati final year18:54
@lambdaysanuj: ah I've been there :D18:55
@lambdaynice place18:55
sanujlambday: i know you did your masters from IIT-B18:55
lisitsynsanuj: stan has a good codebase18:55
lisitsynI don't think its really tough18:55
@lambdaysanuj: great girls! mostly bong professors18:55
sanujlisitsyn: okay18:55
sanujlambday: haha18:55
sanujlambday: i'm scared about graduating...dunno what i'll do18:56
lisitsynsanuj: all you need to know about MCMC is that it is what you use when you're lazy or can't compute some integral ;)18:56
-!- HeikoS [] has joined #shogun18:56
-!- mode/#shogun [+o HeikoS] by ChanServ18:56
sanujlisitsyn: okay18:56
lisitsynsanuj: in this project the plan is to employ its solvers and autograd18:57
@lambdaysanuj: on don't worry you'll be fine.. figuring things out on the way - that's the key :)18:57
sanujlisitsyn: i was planning to submit my proposals for these 2 projects18:57
lisitsynHeikoS: you're back?18:57
@HeikoSlisitsyn: not really ;) but listening on one ear18:57
lisitsynsanuj: good!18:58
sanujlambday: thanks18:58
lisitsynsanuj: plugin thing would give one all the fame18:58
sanujlisitsyn: yeah, this is my first priority...guess this is important for shogun18:59
sanujlisitsyn: and i would get to learn about ml in general18:59
@lambdaygo plugin19:00
lisitsynHeikoS: are our gp classification sampling based?19:01
-!- Nebula_ [~chatzilla@] has joined #shogun19:01
@HeikoSlisitsyn: no19:01
lisitsynHeikoS: what kind of approximation?19:01
@HeikoSep and mostly variational19:01
lisitsynI see19:01
lisitsynHeikoS: btw does sampling works at all?19:02
lisitsynHeikoS: takes unreasonable time?19:02
lisitsynor? %)19:02
@HeikoSlisitsyn: it works very nicely19:03
@HeikoSmost general method19:03
@HeikoSbust obviosly slower19:03
lisitsynHeikoS: I see19:03
@HeikoSvariational methods are quick but not accurate19:03
@HeikoSif you need accuracy of uncertainty, mcmc is king19:03
@HeikoSlisitsyn: but if you only care about predictive RMSE, then EP or variational is totally fine19:04
@HeikoSif model is complex, variational methods fail19:04
@HeikoSbut for GPS ,which are simple, all good19:04
@HeikoSreally depends19:04
sanujHeikoS: i got this education from a shogun talk on youtube by emtiyaz :)19:04
@HeikoSthese methods are not better than one another19:04
@HeikoSsanuj: haha :)19:04
sanuji watched it today19:04
@HeikoSemt was at UCL and gave a talk on Friday19:04
lisitsynHeikoS: nice to know, thanks!19:05
lisitsynHeikoS: I think we can do some kind of experiment19:05
@HeikoSlisitsyn:  which is?19:05
lisitsynI can implement a few methods to use our class_list.cpp19:05
lisitsynand then disable everything in swig, but these methods19:05
lisitsynthis is not plugins but we get this narrow interface we expect from plugins19:06
lisitsynwe're already collecting classes by their names19:06
@HeikoSwhats the goal?19:06
@HeikoSah I see19:06
@HeikoSto design the interface?19:06
-!- Nebula_ [~chatzilla@] has left #shogun []19:06
@HeikoSgood entrance task!19:06
@HeikoSlisitsyn: yeah19:07
sanujnice talk guys, i gotta leave.19:07
lisitsynsanuj: see ya19:07
@HeikoSsanuj: take care!19:07
sanujbye :)19:07
@HeikoSlisitsyn: I also goota go soon19:07
lisitsynHeikoS: I think I have to try it first because we can't merge it yet19:07
sanujsome PRs coming tomorrow19:07
@HeikoSlisitsyn: yeah definitely19:07
lisitsynso not really a good task for a student19:07
@HeikoSit is a good start for a project19:07
-!- mizari [] has quit [Read error: Connection reset by peer]19:07
@HeikoSlisitsyn: well19:08
-!- sanuj [0e8bc402@gateway/web/freenode/ip.] has quit [Quit: Page closed]19:08
lisitsynthey need merged code ;)19:08
@HeikoSafter your draft is done19:08
lisitsynok lets see19:08
@HeikoSlisitsyn: feature branch drafts are also fine I guess?19:08
lisitsynthat's what I am going to try19:08
@lambdayvery useful19:08
@HeikoSyeah thats the way to go I guess19:09
@HeikoSworked well for the cookbook as well19:09
@HeikoSthe build was only broken a bit19:09
@HeikoSlisitsyn: speaking of, can you fix ctags on bsd?19:09
@HeikoSok gotta really go now .... see you later19:09
lisitsynHeikoS: ctags - ok, will check19:10
lisitsynsee you19:10
-!- HeikoS [] has quit [Quit: Leaving.]19:11
-!- HeikoS [] has joined #shogun19:13
-!- mode/#shogun [+o HeikoS] by ChanServ19:13
-!- lambday [6a33f351@gateway/web/freenode/ip.] has quit [Ping timeout: 252 seconds]19:17
-!- lambday [6a33f351@gateway/web/freenode/ip.] has joined #shogun19:34
-!- mode/#shogun [+o lambday] by ChanServ19:34
-!- curiousguy13 [~curiousgu@] has quit [Remote host closed the connection]19:34
@lambdaylisitsyn: shit shogun doesn't build on my laptop!19:35
-!- Saurabh7 [~Saurabh7@] has joined #shogun19:58
-!- witness_ [uid10044@gateway/web/] has quit [Quit: Connection closed for inactivity]20:08
-!- Saurabh7 [~Saurabh7@] has quit [Ping timeout: 248 seconds]20:31
-!- livus [~suvil@] has quit [Ping timeout: 260 seconds]20:32
lisitsynlambday: why?20:46
@lambdaylisitsyn: some issue with gpumatrix.. GPUMatrix.cpp:40: error: redeclared with 4 template parameters20:49
@lambdayI have outdated viennacl20:49
-!- GandalfTheWizard [~ian@] has quit [Ping timeout: 244 seconds]21:00
-!- GandalfTheWizard [~ian@] has joined #shogun21:02
-!- GandalfTheWizard [~ian@] has quit [Ping timeout: 248 seconds]21:09
-!- sonne|osx [] has joined #shogun21:11
@lambdaylisitsyn: can't we have sub namespaces under shogun? seems like class-list.cpp only works with shogun namespace21:21
lisitsynlambday: well we'd have to patch it then I guess21:22
@lambdaylisitsyn: seems so.. I just noticed this21:23
@lambdaylisitsyn: how bad it would be if I add some c++14 checks in cmake and enable -std=c++14 if a compiler supports this? not saying that we'd make it hardcode requirement21:25
lisitsynlambday: no idea21:26
lisitsynwe are not C++11 still..21:26
@lambdaylisitsyn: would it hurt? it's not like build will break if it's not there21:27
lisitsynlambday: I don't think it would hurt21:27
@lambdaylisitsyn: actually the auto param type in lambdas is quite useful..21:28
@lambdayprobably that's the only c++14 stuff I used in my code21:28
@lambdaynah it's better if I change those21:31
-!- Nebula_ [~chatzilla@] has joined #shogun21:32
-!- Nebula_ [~chatzilla@] has quit [Client Quit]21:34
-!- sonne|osx [] has quit [Quit: sonne|osx]21:41
lisitsynlambday: still here?21:54
@lambdaylisitsyn: yessir21:54
lisitsynlambday: going to show you something21:54
@lambdaylisitsyn: sure what is it21:54
-!- GandalfTheWizard [~ian@] has joined #shogun21:55
-!- shogun-notifier- [] has joined #shogun21:56
shogun-notifier-shogun: Sergey Lisitsyn :narrow_interface * a9d5428 / / (3 files):
shogun-notifier-shogun: Introduce a function to create SGOBject by name21:56
lisitsynah shogun-notifier- made this for me21:56
lisitsynlambday: check the unit-test21:56
@lambdaylisitsyn: hehe classy.. this is useful!21:57
shogun-buildbotbuild #3668 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>21:57
@lambdaylisitsyn: maybe useful for class_list21:57
lisitsynlambday: well it uses the class_list21:58
lisitsynlambday: the thing is21:58
lisitsynI want to try to add some more methods and then disable *everything* in swig :)21:58
lisitsynand see how fast it is21:58
@lambdaylisitsyn: how do you plan to use this one for that?21:59
lisitsynlambday: we could change how objects are created under the hood later22:00
lisitsynso it uses plugins22:00
lisitsynHeikoS: oh u here?22:01
lisitsynlambday: there is a thing with new_sgserializable22:01
lisitsynthere is a hack for templated classes22:02
@lambdaylisitsyn: that's what I was thinking - how to instantiate template classes22:02
@lambdaywhat's the hack22:02
lisitsynlambday: some enum22:02
lisitsynlambday: check the definition at shogun/base/class_list.h22:03
@lambdaylisitsyn: only for primitive typees22:03
lisitsynlambday: yes22:03
lisitsynI don't think we need templates for non-primitives22:03
@lambdaylisitsyn: in the present stuff I am working on, I used a few :|22:05
lisitsynlambday: should they be interfaced?22:05
@lambdaylisitsyn: they are base classes, so, yes22:06
@lambdaylisitsyn: I didn't put all these stuffs under shogun yet... so I missed to consider a few things22:07
@lambdaynow having the pain22:07
lisitsynlambday: I see22:07
@lambdaylisitsyn: e.g.
lisitsynlambday: what's test type?22:10
@lambdaylisitsyn: custom struct22:10
lisitsynlambday: can you make it dynamic? ;)22:10
@lambdaylisitsyn: doable - painful :D22:11
lisitsynlambday: I mean you don't really get much speedup by that22:12
lisitsynbut interfacing is painful22:12
@lambdaylisitsyn: yeah that's what I am noticing22:12
@lambdaylisitsyn: BTW when the template param is another SGObject, we're good I guess22:12
@lambdayPT_SGOBJECT is there22:12
lisitsynlambday: yeah somewhat ok22:13
@lambdaynot saying that I'll make these test types SGObjects.. but it's about another case where I used a static polymorphism22:13
lisitsynlambday: I'd keep static polymorphism for something internal ;)22:14
lisitsynlambday: oh I just realised 'object' is not a good name hah22:16
lisitsynin the python context22:16
@lambdaylisitsyn: how about instance?22:20
lisitsynlambday: yeah good22:21
@lambdayit works like a factory - instance sounds reasonable22:21
lisitsynlambday: anyway, this function is not expected to be used directly22:21
lisitsynI will add some classifier which is just instance + cast22:21
lisitsynand stuff like that22:21
-!- besser82 [~besser82@fedora/besser82] has joined #shogun22:39
-!- mode/#shogun [+o besser82] by ChanServ22:39
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 268 seconds]23:03
@HeikoSlisitsyn: im here now23:11
@HeikoSlambday: hi23:11
@HeikoShad a discussion?23:11
lisitsynHeikoS: I am currently experimenting with disabling everything but this very method23:11
lisitsynobject(const char*)23:11
@lambdayHeikoS: hey.. yeah23:11
@HeikoSscary stuff ;)23:11
lisitsynHeikoS: without actual classes swig compiles instantly :)23:13
lisitsynthat's predictable tho23:13
@HeikoSlisitsyn: thats also desirable ;)23:17
@HeikoSlisitsyn: that will be a good draft for gsoc project23:17
@HeikoSlisitsyn: is it that tricky?23:18
lisitsynHeikoS: it seems not really23:20
lisitsynHeikoS: I am now having fun with swig segfault23:21
lisitsynusual mac problem now23:21
lisitsynI don't even feel angry anymore23:22
@lambdayHeikoS: merging this thing to shogun is gonna give me some pain23:24
@HeikoSlambday: the flash?23:24
@lambdayHeikoS: yep23:24
@HeikoSlambday: whats the painkiller? :D23:24
@lambdayHeikoS: apparently, patience :|23:24
@lambdaysent a patch though - with the internals..23:25
@lambdaywill send SGObjects patches one by one23:25
* lisitsyn compiles python because it is easier than to use system one23:26
@HeikoSlambday: haha23:28
@HeikoSlambday: in a feature branch?23:29
@lambdayHeikoS: yes23:29
@HeikoSlambday: cool, travis passes?23:29
@lambdayit was doing some previous build, last time I checked23:29
@HeikoSlambday: hey btw just push into this, no nee to do PRs in this feature branch23:29
@lambdayI am adding documentation and making the code look Shogun style23:29
@lambdayHeikoS: oh23:30
@HeikoSmerge it, then I can compile here :)23:30
@lambdayHeikoS: cool23:30
shogun-notifier-shogun: lambday :feature/bigtest * ba67219 / src/shogun/statistics/experimental/internals/ (34 files):
shogun-notifier-shogun: added first version of bigtesting framework23:30
shogun-notifier-shogun: Soumyajit De :feature/bigtest * fbe78df / src/shogun/statistics/experimental/internals/ (34 files):
shogun-notifier-shogun: Merge pull request #3103 from lambday/feature/bigtest23:31
shogun-notifier-shogun: added first version of bigtesting framework23:31
@lambdayHeikoS: didn't add any of the SGObjects yet.. so not yet usable :(23:31
@HeikoSlambday: ok23:31
@HeikoSill compile anyways23:31
@lambdaygot gazillions of errors23:31
shogun-buildbotbuild #3669 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>, Soumyajit De <>23:32
@HeikoScompiling ...23:34
lisitsynHeikoS: we need features 2.023:38
lisitsynseparating features layout and structure23:38
lisitsynto mix up categorical and numeric things23:38
@HeikoSlisitsyn: yes please23:40
@HeikoSlisitsyn: maybe can be optional part of plugin framework23:40
@HeikoSbut not sure23:40
@HeikoSmaybe something for later23:40
@HeikoSlambday: compiles23:40
@HeikoSbut I cant run I guess?23:40
@lambdayHeikoS: nah the wrappers are not there yet23:41
@lambdayHeikoS: BTW can you please try compiling flash once? it's the same API..23:42
@lambdayjust wanna make sure that the sanity tests work23:42
@lambdayHeikoS: just run the sanity script23:42
@HeikoSlambday: ok cloning23:42
shogun-notifier-shogun: youssef :develop * 18686e5 / src/shogun/multiclass/QDA.cpp:
shogun-notifier-shogun: remove rotations transpose23:45
shogun-notifier-shogun: Heiko Strathmann :develop * a77c561 / src/shogun/multiclass/QDA.cpp:
shogun-notifier-shogun: Merge pull request #3102 from youssef-emad/develop23:45
shogun-notifier-shogun: Remove transpose of rotations in QDA23:45
lisitsynSHEATZ swig swig swig23:45
lisitsynY U NO WORK23:45
@HeikoSlambday: I get lots of no such file or dir23:46
@lambdaylisitsyn: changing that base template to non-template.. making the enums as static int.. essentially serves the same purpose23:46
@lambdayHeikoS: err.. this is an old school makefile.. maybe I missed something23:46
shogun-buildbotbuild #3670 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>, youssef <>23:47
@lambdayHeikoS: which dir is missing23:47
@HeikoSah sorry23:48
@lambdayHeikoS: oh crap it won't compile without c++14.. your compiler doesn't have that! :|23:48
@lambdayHeikoS: I'll finish the merging with Shogun tomorrow..23:48
@HeikoSi can install one23:48
@lambdaythen we can check23:49
@HeikoSgot c++ 4.8.4-223:49
lisitsynI want to see viktor's face when you merge C++14 code23:49
lisitsynHeikoS: your grandma C++23:49
@lambdaylisitsyn: I swear I don't23:49
@lambdayI am using 5.323:49
@HeikoSlisitsyn: hey btw we should push the c++11 thing soon23:49
@HeikoSsonney2k: is also for it23:49
@HeikoSI think wait until next year and then do it23:49
@HeikoSlets face it, nobody uses shogun anyways23:50
@HeikoSso no need to block ourselves23:50
@HeikoSI mean I am all in for multiplatform and compatible etc etc23:50
@HeikoSbut with this, I think we can compromise23:50
lisitsynyes if you're grep23:50
lisitsynor curl23:50
@HeikoSgot more important issues23:50
@HeikoSyeah exactly23:50
@HeikoSbut we can do it after gsoc23:51
@lambdaywould be great to get rid of those macros23:51
@HeikoSone thing for sure: I dont want these double implementations23:51
lisitsynI give up this shit doesn't work again23:51
lisitsynswig + mac = segfault23:51
@HeikoSlambday: just reading your thing in detail and laughing here23:53
@lambdayHeikoS: which one?23:53
@lambdayah linalg :D23:53
@HeikoSlambday: yeah23:54
@HeikoSthats a good method name23:54
@HeikoSlisitsyn would like it23:54
@lambdaygot the inspiration from eigen guys23:54
-!- travis-ci [] has joined #shogun23:55
travis-ciit's Wu Lin's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []23:55
@HeikoSlambday: hey I agree to change this soon, before we get too many patches and students get used to linalg23:55
@HeikoSalready happening, we have like 3-5 patches on this23:55
@lambdayHeikoS: yeah.. this is the first thing I am gonna do after I finish merging flash23:56
@HeikoSlambday: ah man I love these second details you write there23:56
@HeikoSlinalg::elementwise(m).custom([](auto& v) { v = something_fun(v); });23:56
@lambdayHeikoS: BTW how did you like the proposed API for elementwise23:56
@HeikoSvery good23:56
@lambdayHeikoS: yeah these would be neat23:56
@HeikoShaha thats what I just meant23:56
@lambdaywill create issues for students to achieve these effects23:57
@HeikoSlambday: yeah thats a good idea23:57
@HeikoSlambday: maybe we can do this in another feature branch and ppl can send their patches there?23:57
@HeikoSlambday: not sure how to best organise this23:57
@HeikoSlambday: should happen before gsoc start definitely23:58
@lambdayHeikoS: absolutely.. I won't take more than a week to do the rafactoring.. I'll be merging in some feature branch.. but then I think it's better to put it back to dev23:58
@lambday(similar to your argument for shogun) no one uses linalg anyways..23:59
@HeikoSlambday: yeah haha23:59
@lambdayso we won't break anything hopefully23:59
@HeikoSlambday: should be ok23:59
@HeikoSwe have the tests etc23:59
@HeikoSlinalg has like 100% coverage23:59
@lambdayHeikoS: getting rid of the NATIVEBLABLA thing would be the biggest gain API-wise IMO23:59
--- Log closed Mon Mar 21 00:00:05 2016