Open in new window / Try shogun cloud
--- Log opened Mon Jan 28 00:00:38 2013
-!- FSCV [~FSCV@] has joined #shogun01:04
shogun-buildbotbuild #264 of nightly_default is complete: Failure [failed test]  Build details are at
-!- FSCV [~FSCV@] has quit [Quit: Leaving]04:00
-!- cavalier- [cavalier-@] has quit [Ping timeout: 256 seconds]05:53
-!- cavalier- [~cavalier-@] has joined #shogun06:12
-!- sonne|work [~sonnenbu@] has joined #shogun09:05
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun09:56
sonne|workblackburn: how do I add a field to the form? I did add an email addr column10:54
sonne|workbut that is not automagically in the form10:54
blackburnsonne|work: noo, use the form in menu10:54
blackburnform (0)10:55
sonne|workand then?10:55
blackburnedit form10:55
sonne|workyes but how do I add it there?10:55
blackburnadd item in the top left10:55
sonne|workI want name and then email address10:55
sonne|workand then what?10:55
blackburnsome field will appear10:55
blackburndrag it to the place you want10:55
blackburnand edit10:55
sonne|workok maybe I did it correctly10:59
sonne|workn4nd0: anything missing?10:59
n4nd0sonne|work: name and email is enough I think11:01
sonne|workn4nd0: well look at what blackburn sent around11:01
n4nd0when? I didn't get anything during the past minutes11:02
sonne|workblackburn: should we ask about vegeterian food?11:02
sonne|workn4nd0: yesterday11:02
n4nd0the google docs, checking atm11:03
blackburnI don't know - is it highly relevant? I get to know only one vegetarian so far :D11:03
sonne|workblackburn: you barbarians!11:04
blackburnsonne|work: how can one do not eat meat :D11:04
n4nd0nothing else I can come up with mmm11:05
sonne|workblackburn: do we need badges?11:08
blackburnsonne|work: would be cool I think11:08
blackburnit could be kind of souvenir :D11:09
n4nd0+1 to the badges idea :)11:18
sonne|workt-shirts too right ;^)11:20
n4nd0sonne|work: yeah! :D11:27
-!- heiko [] has joined #shogun11:28
sonne|workheiko: moin!11:31
sonne|workheiko: please also give feedback about the form!11:31
heikosonne|work, hi!11:31
heikochecking it out ...11:31
heikosonne|work this form is for before the workshop right?11:51
n4nd0heiko: I think is to keep track of the pople who wants to attent to the workshop12:02
sonne|workheiko: to register for the workshop12:02
n4nd0so yeah before the workshop12:02
heikothen I did not understand the question12:02
heikoIf you weren't satisfied with something, could you explain here what it was?12:02
heikoand the rest is in the comments12:02
sonne|workheiko: about shogun?12:02
heikobut this looks good12:02
heikoI see12:02
heikomaybe we should formulate this differently12:03
n4nd0what is this lookalike contest that was mentioned the other day about?12:03
blackburnheiko: you can edit too ;)12:04
heikoblackburn, I see, but I rather comment and let you edit to avoid confusion12:04
sonne|workblackburn: once this is done - will this form send an email to the list or do you have to do that?12:10
blackburnsonne|work: we can set up notifications12:11
blackburnwell one can set up notifications for himself12:12
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Quit: Page closed]13:33
-!- heiko [] has quit [Quit: Leaving.]13:37
-!- FSCV [~FSCV@] has joined #shogun14:07
-!- FSCV_ [~FSCV@] has joined #shogun14:36
sonne|workblackburn: well that would work14:38
sonne|workI could add some rule matching the pattern to forward this to the mailinglist14:38
blackburnsonne|work:  I can receive mails and just answer with CC14:39
sonne|workyeah but that is :/14:39
sonne|workbetter directly post it to the list14:39
-!- FSCV [~FSCV@] has quit [Ping timeout: 276 seconds]14:39
blackburnI have no idea what is the best way14:41
-!- heiko [] has joined #shogun14:48
heikoblackburn, I dont get you with check possible answers in form.14:54
blackburnheiko: ehm have you seen the form?14:55
heikoI thought this was the template for it14:55
blackburnI suspect you saw only table :D14:55
heikois it on the website?14:55
blackburnno, use Form (0) - Edit form in menu14:55
heikoI see14:56
heikocan I edit?14:56
heikoah cool14:57
heikoI did not even know that google docs could do this :)14:57
blackburnthat would not be cool to ask users to fill in a table14:57
heikoI thought this was just brainstorming, whatever15:00
heikoI think we should have this "how to set-up shogun" tutorial, but it finds in no category15:00
blackburnheiko: may be in discuss part?15:01
heikoits not really a discussion, is it?15:01
blackburnI don't know15:02
heikoI think this list of ML stuff might be renamed: On which of these topics would you like a tutorial15:02
heikothen we could put that in15:02
blackburnheiko: feel free to edit I have poor grammar :D15:03
heikodone :)15:04
heikook gotta work now, see you later15:05
blackburnsee you15:05
-!- heiko [] has quit [Read error: Operation timed out]15:44
-!- heiko [] has joined #shogun15:45
-!- heiko [] has quit [Quit: Leaving.]15:51
-!- heiko [] has joined #shogun15:51
-!- heiko [] has quit [Quit: Leaving.]16:11
-!- heiko [] has joined #shogun16:12
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun16:15
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:30
-!- heiko [] has quit [Quit: Leaving.]17:49
-!- heiko [] has joined #shogun17:51
-!- SCV__ [~FSCV@] has joined #shogun18:24
-!- FSCV_ [~FSCV@] has quit [Ping timeout: 245 seconds]18:24
blackburnheiko: have you ever played fallout?19:25
heikoblackburn, why you ask?19:25
blackburnheiko: I stumbled upon a photo and created this
blackburnpure russia19:26
heiko :)19:26
heikowell, I have played it :)19:27
blackburnheiko: that one with a cigarette looks like a rad-monster!19:27
blackburnwith camouflage :D19:27
heikonice one :)19:27
blackburnheiko: that's what it was before my GIMP19:27
heikoman thats even worse19:28
blackburnheiko: lol I've accidentally hidden factories19:28
heikoso you made the image less-falloutish19:29
blackburnheiko: I should use that pic to convince sonney2k that things are not the same in different countries19:31
heikowhy doesnt he believe this? :)19:31
blackburnheiko: no idea :) but he even wasn't even surprised when I told him that a man was killed with a champagne bottle in police19:32
blackburndouble even hah19:32
heikoRussian is a wild country I guess19:33
blackburnheiko: sure19:36
blackburnyou will see once you see me19:36
heikohaha :)19:37
heikoso many scars?19:37
blackburnheiko: hmm don't have any :(19:38
heikommh, well then I got more than you :)19:38
blackburnheiko: I was asked about combining linear and gaussian kernels19:38
heikoblackburn, and?19:38
blackburnlinear is like in [-1,1]19:38
blackburnand gaussian is in [0,1]19:38
blackburnanything to do with it?19:39
heikowhat means "in"?19:39
blackburnin range19:39
heikoI dont really get what you mean19:42
heikothe range of the kernel function?19:43
blackburnheiko: they have different value domains yes19:43
heikoand now they have to be combined?19:43
blackburnyeah - is it ok?19:43
heikoas long as the linear kernel matrix is psd, this is fine19:43
heikothe linear one wont be negative when the argument is equal right?19:44
heikosince that would destroy it19:44
blackburnyeah psd is not a problem19:44
blackburnlinear combination is still mercer kernel I know19:44
blackburnbut just these different domains - I never though about that19:45
heikono that is fine19:46
heikokernel always goes to real numbers19:46
heikosince it is an inner product19:46
blackburnheiko: no I am not talking about C :)19:47
blackburnheiko: complex numbers19:49
blackburnI mean :)19:49
heikowhy should this negative range be a problem? It is allowed since a kernel is a map X x X => R where X is some set19:49
heikoso what problem should be there?19:49
blackburnhah no problem I just find that interesting19:49
-!- heiko [] has quit [Quit: Leaving.]20:16
-!- tomv_ [] has joined #shogun21:07
tomv_hi. so with blackburn's help I got shogun for python running. but now I am wondering: the FITCInferenceMethod seems to only work when the number of latent features is equal to the number of features.21:08
blackburnhi tomv_21:09
tomv_This is surprising to me, having looked at gpml and also read rasmussen and williams' book...21:09
blackburntomv_: that could be hard as I never managed to get throught rasmussens' book21:10
tomv_well, really the sparse stuff is much better in the overview article IMHO21:10
blackburnRegular and Latent Features do not match in dimensionality21:11
blackburnyou get that, right?21:11
tomv_well, it is unfair to call it a review article. I had been looking at "A Unifying View of Sparse Approximate Gaussian Process"21:12
tomv_or looking at the Large Scale Regression section on
blackburntomv_: do you have some reproducing code?21:13
tomv_gpml does.21:13
blackburnargh heiko has left the room already21:14
tomv_if you look at the picture, the circles are the latent features. unless latent features is something other than inducing inputs.21:14
blackburntomv_: did yuo check that example on fitc we have in examples/undocumented/libshogun?21:15
tomv_actually I used the one in examples/documented/libshogun and tried to port it to python to give a small token of appreciation for your help.21:16
blackburntomv_: so black circles are kind of latent variables?21:17
tomv_but then I stumbled over my idea of how FITC works being literally incompatible to shogun's(?)21:17
tomv_python: /usr/include/eigen3/Eigen/src/Core/DenseCoeffsBase.h:352: Scalar &Eigen::DenseCoeffsBase<Eigen::Map<Eigen::Matrix<double, -1, -1, 0, -1, -1>, 0, Eigen::Stride<0, 0> >, 1>::operator()(Index, Index) [Derived = Eigen::Map<Eigen::Matrix<double, -1, -1, 0, -1, -1>, 0, Eigen::Stride<0, 0> >, Level = 1]: Assertion `row >= 0 && row < rows() && col >= 0 && col < cols()' failed.21:17
tomv_is what I get.21:17
tomv_yes, that is my understanding21:17
blackburnI have no understanding so far as I am stupid code monkey that knows only dimension reduction :D21:18
blackburncould you share that failing code?21:18
blackburntomv_: are you subscribed to the mailing list?21:21
blackburnI am asking because there is a guy who can help you with GPs I think21:23
blackburnheiko strathmann that is - I see you can even talk in person in your common language with more understanding  :)21:24
blackburnman I like your homepage :)21:25
blackburnok let me check example now21:25
blackburntomv_: I get another error message, not eigen3 one21:29
tomv_what do you get?21:29
blackburnsomething about parameters21:29
blackburnI'll try to work it out21:30
tomv_have you compiled without nlopt21:30
blackburnahh right21:30
blackburnokay let me compile with nlopt then21:30
blackburnhmmm wait21:30
blackburnit is with nlopt already21:30
blackburnanyway recompiling21:31
blackburnI get segfault actually21:31
tomv_that should not happen in python, either. though the abort is really bad, too.21:32
tomv_maybe you have optimised the assetions away?21:32
tomv_don't know if that can happen with nlopt21:32
tomv_I am using the stock debian/sid one.21:32
blackburntomv_: segfault happens with python - why not? shogun is C++21:34
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun21:36
n4nd0blackburn: really nice form21:46
blackburnn4nd0: thanks to goolge :)21:47
tomv_blackburn: well, segfault in python means the lib has a bug.21:49
blackburntomv_: exactly21:50
n4nd0tomv_, blackburn where do we have a segfault?22:01
blackburnn4nd0: in parameters stuff22:01
n4nd0mmm I see22:01
tomv_actually, I don't have one, I get assertion errors (which, I would imagine, would lead to a segfault if the assertion is skipped due to optimization options in nlopt)22:02
n4nd0tomv_: mmm if you get an assertion error then execution *must* stop22:15
n4nd0ahhh ok I see what you mean22:16
tomv_hm, why?22:16
n4nd0well because the assertion works like that22:16
n4nd0assert(boolean expression)22:16
n4nd0if the boolean expression is false, then stop22:16
n4nd0tomv_: agree?22:16
tomv_well, to be honest, I might think that exiting the python interpreter because a function call ran into an assertion error is excessive.22:17
tomv_it's assert, not falloverifitsnot.22:17
n4nd0tomv_: mmm22:17
n4nd0but an assert is used to stop if the expression given is false right?22:18
n4nd0I mean say I want to compute the distances between two sets of features22:18
n4nd0if one of them is empty22:18
n4nd0then it makes sense to stop rightaway22:18
n4nd0because it doesn't make sense to compute the distance between some features and nothing22:19
tomv_well, actually, the c++ spec says that the program should abort.22:20
tomv_so yes.22:20
tomv_not "spec" , reference22:20
n4nd0tomv_: ok anyway, all this was almost philosophic :)22:20
n4nd0what do you have in your code a failed assertion or a seg fault?22:21
tomv_and it says that compiling with NDEBUG the assertions are stricken.22:21
tomv_I have a failed assertion, blackburn has the segfault.22:21
tomv_running the above.22:21
n4nd0I cannot test it right now22:24
n4nd0but if you have a failed assertion, I'd suggest to check the inputs you are giving there22:24
tomv_I am afraid sleeping might be the first order of business for me. good night and thanks for your interest.22:26
-!- tomv_ [] has quit [Quit: Ciao]22:26
n4nd0night :)22:27
-!- heiko1 [] has joined #shogun22:35
blackburnheiko1: !22:35
blackburnwe missed you22:35
heiko1I hope so :)22:35
blackburnheiko1: there was a guy who complained about GPs22:35
blackburnheiko1: yes it doesn't work22:36
heiko1blackburn, I know22:36
blackburnheiko1: do you know about FITC?22:36
heiko1Jacob implemented that riught?22:37
heiko1I couldnt even make regression work when I tried it :)22:37
blackburnbut you said you are going to be a GP master22:37
heiko1Yeah, I want to take this over22:37
blackburnI have a feeling you didn't so far22:37
heiko1worked with GP over the last months22:37
heiko1well its a lot of work22:38
heiko1quite a big framwork22:38
heiko1and not so much time here :)22:38
heiko1I even reserached new methods that I want to implement :)22:38
blackburnI am confused with a lot of things in shogun22:38
heiko1but gotta do something else now22:38
heiko1so see you in a bit22:38
blackburnheiko1: okay22:38
n4nd0blackburn: so how many days do you think we will be shoguning in Berlin?23:01
blackburnn4nd0: I expect only tow23:02
n4nd0it sounds fine23:02
n4nd0the plan is to see if we can be at TU the second day right?23:02
n4nd0and what is this lookalike contest?! hehe23:03
blackburnjust joke (*I hope*)23:03
n4nd0are we going to look for people that look like you? hehe23:03
n4nd0I think the competition thing for the audience is a very nice idea23:04
n4nd0hehe yeah was going to ask sth but I answered myself instead :D23:19
@sonney2kblackburn, the issue tomv_ is having is the one I am debugging...23:58
@sonney2kno success yet though23:58
* sonney2k is happy to be online again23:58
@sonney2kbricked the router and hat to short circuit some pins on the flash chip23:59
--- Log closed Tue Jan 29 00:00:38 2013