Open in new window / Try shogun cloud
--- Log opened Wed Aug 01 00:00:17 2012
blackburnwiking: ping00:17
blackburnn4nd0: I want to make streaming stuff optional00:19
blackburnwhat do you think?00:19
n4nd0sorry I didn't hear the bell00:24
n4nd0but I read your message00:24
n4nd0blackburn: it makes sense to make it optional if it is not normally used00:24
n4nd0but what will we will if it is optional?00:24
n4nd0-> problems with people building shogun and wanting to use it (they won't find it by default)00:25
n4nd0-> compilation time reduced??00:25
blackburnyeah significantly00:25
n4nd0do you think so?00:25
CIA-18shogun: Christian Widmer master * rb1765dd / src/configure : fixed eigen3 detection for the case where includes are located in non-standard directory, e.g. ./configure --includes=/home/user/include/eigen3/ -
CIA-18shogun: Sergey Lisitsyn master * rd866352 / src/configure : Merge pull request #682 from cwidmer/master -
blackburnproblem is that it is not used at all00:25
blackburnI bet just nobody uses it00:25
blackburnand it is not really supported I am afraid00:25
n4nd0didn't pluskid ask about it some time ago during his GSoC?00:26
n4nd0something related to VW00:26
blackburnwell he introduced one algo which is online actually00:26
n4nd0that would go to the optinal part too?00:27
blackburnyeah probably00:28
n4nd0apart from the compilation time, something else?00:28
blackburnnothing :D00:29
n4nd0mmm ok00:30
blackburnprobably wrong idea :D00:31
n4nd0I think it is a good idea00:31
n4nd0but let say it this way00:31
n4nd0why instead of making optional to use them00:31
n4nd0making optional NOT to build shogun with it?00:31
blackburnheh good00:31
n4nd0it would be something like the disable-optimizations we normally use00:32
blackburnwe should find a way to separate interface file00:32
CIA-18shogun: Sergey Lisitsyn master * r6a311bd / (63 files in 9 dirs): Moved streaming classes into a separate folder -
shogun-buildbotbuild #131 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>01:37
-!- n4nd0 [] has quit [Quit: Lost terminal]01:38
-!- n4nd0 [] has joined #shogun01:40
shogun-buildbotbuild #235 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>02:05
shogun-buildbotbuild #195 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>02:08
-!- needsch175 [] has joined #shogun02:12
-!- needsch [] has quit [Ping timeout: 260 seconds]02:16
-!- needsch [] has joined #shogun02:22
-!- needsch175 [] has quit [Ping timeout: 256 seconds]02:22
-!- needsch [] has quit [Ping timeout: 248 seconds]02:28
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]03:04
-!- n4nd0 [] has quit [Quit: leaving]03:28
shogun-buildbotbuild #46 of nightly_default is complete: Failure [failed test]  Build details are at
-!- Cheng [80fa500f@gateway/web/freenode/ip.] has joined #shogun06:36
-!- pluskid [~pluskid@] has joined #shogun08:10
pluskidCheng: hi!08:11
-!- pluskid [~pluskid@] has quit [Ping timeout: 260 seconds]08:29
-!- pluskid [~pluskid@] has joined #shogun08:30
Chengpluskid: Hi, sorry was just chatting with someone in the corridor.08:45
pluskidCheng: then hi again :)08:45
pluskidso shall we talk about our next plan?08:46
Chengpluskid: So, you've finished with RelaxTree?08:46
pluskidI'm currently working on the feedback of my PR to get it merged08:46
-!- uricamic [~uricamic@2001:718:2:1634:fd90:1089:27:b16] has joined #shogun08:47
ChengWhat next? fast label tree, hinge boost, adaptive codes, benchmarks.08:47
pluskidmaybe I will try to port some of my survey to the shogun-tutorial doc08:47
ChengThat's what I have on my list.08:47
ChengIs this still correct?08:47
pluskidthe next might be hingeboost or fast label tree08:47
ChengYes, porting your survey to the tutorial is a great idea.08:48
pluskidconsidering we already have two trees, I might prefer hingeboost, but haven't decided08:48
ChengYes, hinge boost would give a different class of algorithm.08:48
pluskidat the same time, if we could get some help (e.g. reference code) from Gun for adaptive codes, that could also be taken with higher priority08:48
ChengIt is worth emailing him to ask, because Gunnar is the kind of person who keeps everything.08:49
ChengSend him an email and copy me.08:49
pluskidCheng: OK, I'll do that later today08:49
ChengTo benchmark or not to benchmark, that is the question.08:50
pluskidPersonally I support this idea08:50
pluskidbut I wish to postpone it08:50
pluskidbecause I'm not very sure that reasonable results can be obtained within some fixed period of time08:51
ChengPostpone in the sense "not in GSoC"?08:51
pluskidCheng: no, I mean at the tail of GSoC08:51
pluskidbtw, I'm sort of forgetting the end date of GSoC08:51
pluskidI'll check that now08:51
ChengOk, but I was thinking that perhaps you may not have enough time if you don't drop one of the three remaining algorithms.08:52
pluskidCheng: yeah, you are right, I can't implement all of them08:52
pluskidI spent a bit longer than expected to implement Relaxed Tree08:53
pluskidand now I also think shogun-tutorial is important08:53
pluskidAh, end date is Aug. 2008:53
pluskidpencil down08:53
pluskidso I'm afraid there's only time for one more algorithm08:54
ChengI think you already have enough algorithms.08:54
pluskidbesides shogun-tutorial and benchmark08:54
ChengSo if you do not implement any more, it would also be ok for me.08:54
ChengI think it is between Gunnar and hingeboost08:54
pluskidCheng: OK08:55
Chengif you want to implement one more...08:55
pluskidanyway, I'll ask Gunnar for some suggestions first08:55
ChengDo you agree to forget about fast label tree?08:55
pluskidyeah, seems we have to08:56
pluskidand we have enough trees08:56
ChengWhen I suggested the benchmark, I was thinking mostly about its tutorial value.08:56
pluskidCheng: yes, too many algorithms also making users confusing08:56
ChengI don't think it is that important (for you) to do a proper benchmarking.08:56
pluskidCheng: you mean just some guidelines or examples to show to users?08:57
ChengFirst definitely example code.08:57
ChengBut also some sort of way to decide what to use.08:57
pluskidhmm, I agree with you08:57
pluskidactually, I think many of the algorithms are not directly comparable, because they have their different characteristics and application domains08:58
ChengHere's a crazily super dream:
ChengI know we cannot achieve that kind of thing, but this would be something very useful everytime there are multiple algorithms for related problems.08:59
pluskidwow, that site is cool08:59
ChengThat's why perhaps "benchmark" was not the right word.08:59
pluskidOK, I see now08:59
pluskidbtw: I also have an issue with benchmarking, it seems unrealistic to run (and tuning, cross-validation, etc.) MNIST on my laptop09:00
ChengHow about USPS?09:01
ChengI think it is a smaller dataset.09:01
pluskidI have access to a server to my previous school, but my access permission is too low to install softwares (like shogun)09:01
ChengWait... what do you think should go into the benchmark?09:01
pluskidso maybe we might use some smaller dataset09:01
pluskidCheng: I guess at least we should try hard to make the results best?09:02
pluskidlike parameter tuning09:02
ChengUSPS digits is smaller, e.g. available at:
pluskidbut that's also contradicting, because unlike experiments in a paper, where the performance is the most cared thing, I think we should also care about the running time here09:03
ChengI think a proper benchmarking will take way too long.09:03
ChengDefinitely more than 20 days.09:03
pluskidCheng: yeah09:03
pluskidCheng: so do you have any suggestions here?09:04
pluskidCheng: what should we focus on in this "benchmark" and what can be relaxed?09:04
Chengpluskid: My idea is to write some example code for each algorithm.09:04
ChengThe focus is on the tutorial nature.09:04
ChengHave two or three datasets, such that each algorithm can work on one or two of them.09:05
pluskidCheng: OK, so we polish and tweak the existing libshogun or python_modular example code, add comments and explanations, right?09:05
ChengIt is ok to have a table with empty entries.09:05
pluskidand we add suggestions for algorithm selection09:05
pluskidOK, I think I'm sort of clear now09:06
ChengBut the suggestions for algorithm selection, and also the table comparing runtime and accuracy should be in the tutorial (or somewhere similar).09:06
ChengI'm not 100% sure that these more complex examples should go into python_modular examples directory.09:07
ChengThat's why I gave them a new name "benchmark".09:07
pluskidCheng: OK, I see09:07
ChengThe reason for the python_modular examples is so that people can just hack them to use for their own purposes.09:08
ChengSo, they need to run in a few seconds.09:08
ChengSee my dilemma?09:08
pluskidI think other people might also need example code in their section in shogun-tutorial, so we might end up with a new directory or a new shogun-foobar repo09:08
pluskidCheng: yeah, I see09:08
ChengI think right now, just do everything under the tutorial.09:09
pluskidhmm, OK09:09
ChengSo, have some more complex examples (e.g. with small amount of parameter tuning) in the tutorial.09:09
ChengI'm not also so clear about exactly what kind of tutorial will appear, but it would be nice to have some theory and some code too.09:10
pluskidyeah, and some explaination of how the algorithm works09:10
ChengLike Marsland's book.09:10
pluskidthey already have some content there, but most are still empty09:11
ChengAs part if the multiclass chapter, it would be nice to have the "benchmark"09:11
pluskidthis book looks cool09:12
pluskidI think my friends will like it when I recommend this to them :)09:12
-!- n4nd0 [] has joined #shogun09:12
Chengpluskid: So how about (1) tutorial, (2) one more algorithm, for the next 3 weeks?09:13
pluskiddifficulty with this kind of things is maintaining the code to be compatable with the latest code09:13
pluskidCheng: yeah, I think this is reasonable09:14
pluskidCheng: do you have priority over tutorial by putting (1) and (2)?09:14
ChengNot really. You can do whichever order you like.09:15
pluskidCheng: OK09:15
ChengUm... I should know this, but do you have to write a report or something for GSoC?09:15
pluskidCheng: I'm not sure, I guess no09:15
ChengOne less thing to do. :-)09:16
pluskideven if there's one, that should not be very difficult I think09:16
pluskidsomething like the mid-term survey09:16
ChengSo, you decide between hingeboost and polychotomies?09:16
ChengDoesn't have to be now.09:17
pluskidnot yet, I'll make the decision after feedback from Prof. Gunnar09:17
pluskidanyway, I'll fix the code reviews of my PR first, Soeren gave me a lot of suggestions :D09:18
ChengI feel kind of guilty being a mentor who cannot really program in shogun anymore...09:19
ChengAs long as you have some people to help you have fun, I think it is ok...09:19
pluskidCheng: haha, do mentors at shogun typically programming with the students?09:19
ChengSoeren does.09:20
pluskidCheng: Soeren have to be there, or else all of us will fill shogun with STL stuffs immediately haha :p09:20
ChengBut mostly, the mentors are experts in the method, and not in C__09:20
pluskidCheng: and I think your suggestions are always very helpful for me09:21
ChengWell, if you can figure out how to swig STL to all target languages, then I'm sure it will not be a big problem.09:21
pluskidhaha, I'll not take that risk of nightmares of maintaining so many languages interfaces09:22
pluskidCheng: So I'm clear of all my questions, thank you very much for your time. Do you have any other stuffs that are to be discussed?09:22
Chengpluskid: I think we discussed everything.09:23
Chengpluskid: I'll pull the tutorial to have a look next week.09:24
Chengpluskid: if that is ok with you.09:24
Chengpluskid: or whenever you are ready, and tell me to.09:24
pluskidCheng: not sure there will be enough of multiclass stuff there by then09:24
pluskidCheng: OK, I'll write you an email when I have something there :)09:25
Chengpluskid: then some later time then...09:25
n4nd0pluskid: hi pluskid, do you how can I push a .mat to the repo?09:31
n4nd0pluskid: I believe I have to do it pushing to the shogun-data repo and doing a PR09:32
n4nd0I am not sure though09:32
pluskidn4nd0: I think you have to fork shogun-data in github to make a PR09:32
pluskidbut I haven't tried that09:32
n4nd0I think I have actually done it once09:33
n4nd0haha I don't remember09:33
pluskidI put some small toy dataset, but blackburn helped me to do that, so I didn't make a PR09:33
pluskidif your .mat is not big, maybe blackburn can help you to put there directly09:33
pluskidcloning a whole shogun-data is really slow for my internet connection09:34
n4nd0I already have it here to tell the truth09:34
pluskidhmm, then that might be easier :)09:34
n4nd0yeah, I will just add my file there09:35
n4nd0thank you!09:35
pluskidyou are welcome :)09:36
n4nd0pluskid: when are you going to USA btw?09:36
pluskidn4nd0: fly is on Aug. 2309:36
n4nd0pluskid: not that much left!09:36
pluskidright after GSoC pencil down time09:37
n4nd0it must be really exciting for you, right?09:37
pluskidyeah, to tell the truth, I can't wait any more  :p09:37
pluskidfirst time to be abroad09:37
n4nd0and to start your phd!09:38
pluskidn4nd0: btw, did you watch the movie "the PhD movie"?09:38
pluskidmovie for the "Piled Higher and Deeper" comics09:38
n4nd0n4nd0: no I haven't09:40
n4nd0is it good?09:40
pluskidit's interesting09:40
pluskidreflect many of the "real life" of graduate students I think09:40
pluskidthough a bit exaggerating :D09:41
n4nd0haha ok09:41
pluskidthough I like their comics better than their movie :D09:41
n4nd0pluskid: where can I find the comics?09:45
n4nd0I never heard about them09:45
n4nd0thank you09:47
pluskidur wel~09:48
-!- Cheng [80fa500f@gateway/web/freenode/ip.] has quit [Quit: Page closed]10:21
-!- pluskid [~pluskid@] has quit [Ping timeout: 252 seconds]10:43
-!- pluskid [~pluskid@] has joined #shogun11:03
-!- n4nd0 [] has quit [Quit: leaving]11:23
-!- koen-shogun [~ksande@] has left #shogun []11:35
-!- koen-shogun [~ksande@] has joined #shogun12:06
@sonney2kpluskid, around?12:24
@sonney2kpluskid, if you have time I would like to talk a bit... let me know when you are around - friday before the meeting maybe?12:36
-!- needsch [] has joined #shogun13:23
pluskidsonney2k: hi13:51
-!- pluskid [~pluskid@] has quit [Quit: Leaving]14:16
-!- pluskid [~pluskid@] has joined #shogun14:30
-!- uricamic [~uricamic@2001:718:2:1634:fd90:1089:27:b16] has quit [Quit: Leaving.]15:16
-!- uricamic [~uricamic@2001:718:2:1634:fd90:1089:27:b16] has joined #shogun15:16
-!- n4nd0 [] has joined #shogun15:29
-!- pluskid [~pluskid@] has quit [Ping timeout: 260 seconds]15:31
-!- pluskid [~pluskid@] has joined #shogun15:31
-!- n4nd0 [] has quit [Quit: leaving]17:31
-!- yoo [575b08cb@gateway/web/freenode/ip.] has joined #shogun17:35
-!- yoo [575b08cb@gateway/web/freenode/ip.] has quit [Client Quit]17:35
-!- heiko [] has joined #shogun17:48
-!- uricamic [~uricamic@2001:718:2:1634:fd90:1089:27:b16] has quit [Quit: Leaving.]17:48
-!- pluskid [~pluskid@] has quit [Quit: Leaving]18:05
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun18:19
CIA-18shogun: Heiko Strathmann master * r1434a79 / (9 files in 3 dirs): Merge pull request #686 from karlnapf/master (+7 more commits...) -
CIA-18shogun: Heiko Strathmann master * r95dd125 / examples/undocumented/libshogun/statistics_linear_time_mmd.cpp : removed an assertion that (still) causes trouble -
CIA-18shogun: Heiko Strathmann master * r36f5a27 / examples/undocumented/libshogun/statistics_linear_time_mmd.cpp : Merge pull request #687 from karlnapf/master -
-!- heiko [] has left #shogun []19:15
shogun-buildbotbuild #132 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>19:17
shogun-buildbotbuild #133 of bsd1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>19:20
shogun-buildbotbuild #236 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Heiko Strathmann <>19:45
shogun-buildbotbuild #196 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Heiko Strathmann <>19:49
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:51
-!- blackburn [~blackburn@] has joined #shogun20:05
blackburnhey there20:13
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun20:16
shogun-buildbotbuild #237 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Heiko Strathmann <>20:22
shogun-buildbotbuild #134 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>20:26
-!- ckwidmer_ [8ca3fe9d@gateway/web/freenode/ip.] has joined #shogun20:41
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:41
shogun-buildbotbuild #238 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Heiko Strathmann <>20:46
shogun-buildbotbuild #197 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Heiko Strathmann <>20:48
CIA-18shogun: Sergey Lisitsyn master * r735abe5 / (3 files): Fixed libshogun examples -
shogun-buildbotbuild #135 of bsd1 - libshogun is complete: Success [build successful]  Build details are at
@sonney2kblackburn, lots of breakage today?22:36
blackburnsonney2k: no not really22:36
shogun-buildbotbuild #239 of deb3 - modular_interfaces is complete: Failure [failed test ruby_modular]  Build details are at  blamelist: Sergey Lisitsyn <>22:46
shogun-buildbotbuild #198 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at
-!- audy [] has quit [Changing host]23:10
-!- audy [~audy@unaffiliated/audy] has joined #shogun23:10
blackburnsonney2k: around?23:15
blackburnwiking: may be you? :)23:19
CIA-18shogun: Sergey Lisitsyn master * r235b136 / (4 files in 3 dirs): Fixes for multitask logistic regression -
-!- n4nd0 [] has joined #shogun23:42
n4nd0hey blackburn23:51
blackburnn4nd0: hey23:51
n4nd0I've just seen your comment23:51
n4nd0but I must be already half-sleep :P23:51
n4nd0what do you mean with it?23:51
blackburnn4nd0: I just posted it23:52
blackburnwhich one?23:52
n4nd0will it come back?23:52
blackburnyou commented out ifdef23:52
n4nd0oh fuck23:52
blackburnand in mosek.cpp too23:52
n4nd0it is because sometimes, I don't know why even if I put manually the USE_MOSEK in .config, shogun was not built for those parts23:53
blackburnn4nd0: btw I bet it is a good experience for ya to play with mosek23:53
n4nd0blackburn: yeah, it has been useful I think23:54
blackburnso mosek detection is bad?23:54
n4nd0mmm not really23:54
n4nd0it was something with the .config and the ifdefs23:54
n4nd0that even if I had in my .config (manually introduced) -DUSE_MOSEK23:55
n4nd0those parts were not compiled23:55
n4nd0I ignore why23:55
blackburnohh your PR is terribly big23:55
n4nd0I am sorry :(23:56
n4nd0about the USE_MOSEK thing23:56
n4nd0I think I should modify and amend to a commit23:56
n4nd0otherwise, I am afraid that the buildbot will fail23:56
n4nd0because it doesn't have mosek installed23:56
blackburnfeel free :)23:56
n4nd0I will have to amend previous commits23:57
blackburnehmm do we have mosek detection at all?23:57
n4nd0I did it some days ago23:57
n4nd0Soeren told me not worry about it23:57
blackburnwell I'll do that then23:57
n4nd0that we didn't plan to include it23:57
n4nd0I don't know23:58
blackburnso you wanted *not to have mosek in configure*?23:58
n4nd0I had mosek in the configure23:59
n4nd0but it is not autodetected23:59
n4nd0for example23:59
n4nd0I run ./configure --disable.... blah blah blah23:59
n4nd0and later I modify the file .config created23:59
n4nd0adding in INCLUDES -DUSE_MOSEK23:59
n4nd0the corresponding compilation flags23:59
--- Log closed Thu Aug 02 00:00:01 2012