n4nd0 shogun-buildbot gsomix --- Log opened Tue Apr 03 00:00:19 2012 -!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun 00:02 -!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] 00:02 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 00:02 -!- tibi_popa [tibi_popa@95.76.45.235] has left #shogun [] 00:03 shogun-buildbot: oh, C# successful, good job! 00:04 What you say! 00:04 shogun-buildbot, test. 00:05 :( 00:05 Ok. 00:05 -!- harshit_ [~harshit@182.68.67.61] has quit [Ping timeout: 265 seconds] 00:06 -!- harshit_ [~harshit@182.68.67.61] has joined #shogun 00:06 hehe 00:07 gsomix: how is it going with your proposal man 00:07 ? 00:07 n4nd0, i'm writing now. 00:08 gsomix: cool! do you find useful the one I sent you? 00:08 yep 00:09 :) 00:09 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has joined #shogun 00:11 Oh. Jimi Hendrix and Duane Allman are ones of a greatest guitarists in history... I can listen to them a very long time. 00:21 But it's time to sleep. 00:22 Goog night. 00:22 -!- karlnapf [~heiko@host86-181-9-49.range86-181.btcentralplus.com] has joined #shogun 00:24 n4nd0: please send me sample proposals also .. 00:29 having sometrouble making mine :( 00:29 harshit_: trouble? why so man? 00:29 For creating the detailed timeline 00:30 what material were you talking about with gsomix ? 00:30 I can send you that part then 00:30 in any case, what projects are you applying for? 00:30 it would be kind of weird to share proposal for the same project :P 00:30 C5.0 + regression solvers of liblinear 00:30 I think yours is SO learning . right ? 00:31 ok, so give me an address and I can send you the timeline for the multiclass project I have done today ok? 00:31 or SO, the one you prefer, they are pretty similar in structure actually 00:31 here : harshit.syal@gmail.com 00:32 but I warn you, I think that mine's not pretty detailed at all 00:32 it's quite basic actually :S 00:32 but still i'll get an idea 00:32 yeah, sure 00:33 please use it just as an idea ... it would be weird if we all send a very similar proposal :P 00:33 yeah sure , i wont copy it :) 00:34 but i thought they ask for weekly timeline 00:34 i mean details of each weak 00:35 week* 00:35 I don't really know 00:35 I suggest you to ask S?ren to get a better idea what do they want 00:35 try to ask when you meet him around here so I can get also an answer for that :) 00:36 sure ;) 00:36 g2g its about 5 here , need to sleep 00:37 yeah sure 00:37 I was going to ask you about that man 00:37 bye 00:37 becuase I see that you are with us in the chat lot of times 00:37 but you have another time ! 00:37 yeah have to manage somehow.otherwise my work wont progress 00:38 hasta luego 00:38 -!- harshit_ [~harshit@182.68.67.61] has quit [Quit: Leaving] 00:38 -!- harshit_ [~harshit@182.68.67.61] has joined #shogun 00:39 forgot to mention, Thanks for sending me your timeline 00:40 -!- harshit_ [~harshit@182.68.67.61] has quit [Client Quit] 00:40 you are welcome, good night 00:41 -!- av3ngr [av3ngr@nat/redhat/x-tfqcnnyvsaidfgdq] has joined #shogun 00:50 -!- karlnapf [~heiko@host86-181-9-49.range86-181.btcentralplus.com] has left #shogun [] 01:03 build #425 of python_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/425 01:19 shogun-buildbot: great!! 01:20 What you say! 01:20 -!- mohataher [52c9fc37@gateway/web/freenode/ip.82.201.252.55] has joined #shogun 02:12 Hi, I'm mohamed, applying for GSoC and I have a question regarding the application form for it. Is there anyone to answer my question now ? :) 02:13 -!- nickon [~noneedtok@dD5774105.access.telenet.be] has quit [Quit: ( www.nnscript.com :: NoNameScript 4.22 :: www.esnation.com )] 02:19 mohataher: hey mohamed! maybe I can help you 02:21 tell me 02:21 Thank you. Regarding sending a pull request, does it have to include any modification to the code/documentation or just a to try it out ? 02:57 it should include a patch 03:00 I think there was a conversation about that in the mailing list 03:01 but definitely, an empty patch is not enough 03:02 -!- LeStealth [~benjamin@CPE001921e5c431-CM00111ae68378.cpe.net.cable.rogers.com] has joined #shogun 03:14 okay, thanks for your answer. I will search for it. 03:15 let me know if you need help to find it 03:16 i found the mailing list and I'm skimming it now. I will let you know if I need help. Thank you. 03:21 I found the answer. Thanks. 03:24 -!- mohataher [52c9fc37@gateway/web/freenode/ip.82.201.252.55] has quit [Quit: Page closed] 03:24 -!- LeStealth [~benjamin@CPE001921e5c431-CM00111ae68378.cpe.net.cable.rogers.com] has left #shogun ["Leaving"] 03:27 -!- hpsoar_ [7c108901@gateway/web/freenode/ip.124.16.137.1] has joined #shogun 03:37 -!- hpsoar_ [7c108901@gateway/web/freenode/ip.124.16.137.1] has quit [Ping timeout: 245 seconds] 03:44 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] 03:45 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has quit [Quit: flxb] 03:58 -!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun 04:32 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 05:07 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 05:26 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 05:52 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] 06:19 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 06:19 -!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Ping timeout: 245 seconds] 06:50 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun 07:55 hi 08:00 hey 08:02 n4nd0, so is our buildbot happy again? 08:07 sonney2k: it looked like that tonight :) 08:08 let's see if it continues well 08:08 sonney2k: yay! everythin all right 08:08 everything* 08:08 shogun: Heiko Strathmann master * rbad4740 / src/shogun/features/Subset.cpp : small cast fix - http://git.io/HefGQQ 08:10 shogun: Soeren Sonnenburg master * r0e4f543 / src/shogun/features/Subset.cpp : 08:10 shogun: Merge pull request #413 from karlnapf/master 08:10 shogun: small SG_ADD type cast fix - http://git.io/_zoRkA 08:10 so shogun-buildbot work! 08:11 n4nd0, heh even warnings down to nothing :) 08:11 sonney2k: cool! 08:11 n4nd0: btw how does the mkl picture look like? 08:44 sonne|work: this is for very few samples 08:48 sonne|work: http://dl.dropbox.com/u/11020840/mkl.png 08:49 n4nd0: could you run this with shading interp ? 08:49 sonne|work: I think it is already set like that 08:49 sonne|work: http://snipt.org/ugjgd6, line 116 08:50 something is wrong then 08:52 contours look pretty ok IMHO 08:52 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 08:53 aham 08:53 I guess the first test to fix this would be to use just 1 kernel 08:53 and if it still looks weird with 1 then figure out what the difference is... 08:54 n4nd0: I just tried - looks still totally broken ... but no time to go deeper for now 08:58 sonne|work: what do you suggest to use a CombinedKernel just appending one more kernel or one PolyKernel alone? 08:58 sonne|work: yeah, I tried changing a couple of things and it broke too :S 08:59 I even used a single GaussianKernel 08:59 still all weird 08:59 aham 08:59 I can try to take it a look and show you the results in some minutes 09:00 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 09:11 -!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has joined #shogun 09:19 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 09:37 -!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun 09:42 uhuh 09:44 blackburn: trying to scare people? :P 09:44 no people scared me 09:45 moaaaar pull requests 09:45 hah 09:45 haha 09:46 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 09:50 -!- Marty28 [~Marty@158.181.76.57] has joined #shogun 09:51 hi 09:51 Marty28: hey! 09:52 -!- harshit_ [~harshit@59.177.193.26] has joined #shogun 09:57 hey harshit_! 10:10 nice work 10:10 btw I didn't understand your benchmark numbers 10:10 how long did it take on covertype? 10:11 for shogun and how long with just matlab? 10:11 -!- harshit_ [~harshit@59.177.193.26] has quit [Ping timeout: 246 seconds] 10:14 -!- harshit_ [~harshit@59.177.193.26] has joined #shogun 10:19 -!- blackburn [~qdrgsm@83.234.54.186] has quit [Ping timeout: 265 seconds] 10:25 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 276 seconds] 10:27 -!- harshit_ [~harshit@59.177.193.26] has quit [Ping timeout: 244 seconds] 10:32 * gsomix went to classes about computer networks. 10:34 -!- harshit_ [~harshit@59.177.193.26] has joined #shogun 10:49 -!- harshit_ [~harshit@59.177.193.26] has quit [Quit: Leaving] 11:03 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] 11:21 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 11:22 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] 11:22 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 11:23 -!- karlnapf [~heiko@host86-181-9-49.range86-181.btcentralplus.com] has joined #shogun 11:25 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has joined #shogun 11:31 -!- av3ngr [av3ngr@nat/redhat/x-tfqcnnyvsaidfgdq] has quit [Quit: That's all folks!] 11:51 -!- nickon [~noneedtok@dD5774105.access.telenet.be] has joined #shogun 11:56 -!- wiking [~wiking@iridia-pc21.ulb.ac.be] has joined #shogun 12:09 -!- wiking [~wiking@iridia-pc21.ulb.ac.be] has quit [Changing host] 12:09 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 12:09 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has quit [Quit: flxb] 12:26 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun 12:26 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has joined #shogun 12:28 I am taking a look to the seg fault Nicholas has pointed out 12:48 -!- flxb [~cronor@e178173124.adsl.alicedsl.de] has quit [Quit: flxb] 13:03 sonne|work: around? 13:18 yes 13:19 the way to do CV for kernel machines is to use the constructor that takes a machine and no features right? 13:20 the kernel has to be set to that machine 13:20 I think so - karlnapf is the expert for that 13:24 hum? just read my name :) 13:24 well you can specify a kernel before 13:25 depends on what you want to do 13:25 just simple CV for evaluation or parameter search? 13:25 karlnapf: I am trying to fix the example of the guy that sent the mail to the list 13:26 his problem is that when he does CV with LibSVM it seg faults 13:26 ah, ok 13:26 CV means what? 13:26 the kernel has never been set, we should lock that 13:26 cross validation, sorry 13:26 just for simple evaluation or for a parameter search? 13:27 so now I am trying to find out how to do it for LibSVM, I have managed to get something running 13:27 I am afraid I cannot answer that :S 13:27 I can so you the code of the example if so and you can tell me if it is all right, ok? 13:28 mmh, well 13:28 yes sure 13:28 http://snipt.org/ugjie0 13:28 so in the line 45 13:28 ok 13:29 before it was simply LibSVM(), and it turned out to seg fault for this issue that we get to execute a method for the kernel member of LibSVM 13:29 yes you have to set the kernel before training in that case 13:29 and the kernel had never been initialized 13:29 the parameter search does not involve the kernel at all 13:29 so this is cross validation then? 13:29 no its a parameter search 13:29 but this is based on cross-validation 13:30 aham 13:30 since you have to test your current parameters somehow 13:30 and thats done with CV 13:30 ok, I think I get the main idea 13:30 there are two seperate classes for that, you can see them in the example 13:30 my only concern is that now to create the LibSVM 13:30 mmh 13:30 doesnt that example work? 13:30 should 13:30 I have to use a value for C, and the labels 13:31 yes yes, it does 13:31 but idk if it does what it should do ;) 13:31 so the point is 13:31 when the instance of LibSVM is created now 13:31 this values, are used somewhen? 13:31 which values? 13:32 the values for C and the labels 13:32 ah ok 13:32 well the labels are used in training, in the CV, they are needed 13:32 I mean it because C is supposed to be tried out within an interval 13:32 and the C value is just overwritten 13:32 ok 13:32 but the labels are given again in the CV constructor, line 55 13:33 oh, yes, true 13:33 well, then you could create the SVM with an empty constructor 13:33 and set just the kernel by hand instead 13:33 classifier=LibSVM() 13:33 classifier.set_kernel(kernrel) 13:33 I think that is fairly reasonable 13:33 (or so) 13:33 we ineed do sometimes not check for good initialization 13:34 great 13:34 thats currently all a bit complicated i guess :) 13:34 it works as well with the way you've suggested 13:34 nice then 13:35 and I like it more than using the labels in the constructor 13:35 I found that confusing 13:35 yes its nicer that way 13:35 ok 13:35 say n4nd0, are you fernando? 13:36 so I think that I should just add a fix in the method that was seg faulting to fire SG_ERROR in case the kernel is null 13:36 karlnapf: yeah! you are Heiko right? 13:36 I am a bit confused by all these nicknames 13:36 yes :) 13:36 haha 13:36 but yours is even more difficult to guess :P 13:37 yes, an error message if there is no kernelk would be nice 13:37 yes, true ;) 13:37 when blackburn shows up, we can talk about that subset stuff 13:37 karlnapf: sure 13:37 in the meantime I will add a pull request with the fix and answer Nicholas 13:38 nice one 13:38 I will continue on my work now, but will have one eye on the irc, see you! 13:38 thank you, bye! 13:39 sonne|work: jasmine's pull request looks nice! 13:58 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 240 seconds] 14:06 -!- sidsniper [~piggy@14.139.82.6] has joined #shogun 14:27 shogun: iglesias master * rce07314 / (2 files): * fix seg fault when trying to do CV with no kernel initialized, * fix indent - http://git.io/4es7wg 14:30 shogun: Soeren Sonnenburg master * r2b4bdc2 / (2 files): 14:30 shogun: Merge pull request #415 from iglesias/fix-modelsel 14:30 shogun: Small fix to avoid seg fault - http://git.io/vqs8lQ 14:30 -!- sidsnipe [~piggy@14.139.82.6] has joined #shogun 14:32 -!- flxb [~cronor@fb.ml.tu-berlin.de] has joined #shogun 14:32 -!- sidsniper [~piggy@14.139.82.6] has quit [Ping timeout: 260 seconds] 14:33 -!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun 14:33 -!- sidsnipe [~piggy@14.139.82.6] has quit [Client Quit] 14:33 karlnapf: hey there 14:34 hey blackbutrn 14:41 blackburn, hi 14:41 karlnapf: I heard you wanted to kill me because of machine subsetting"? ;) 14:42 krkr ;) 14:42 no, just asking about it 14:42 explain it so me 14:42 ok 14:42 lets take a look 14:42 once you train machines OvO 14:43 you need to subset your trainset 14:43 OvO is one versus one? 14:43 yes 14:43 yes 14:43 https://github.com/shogun-toolbox/shogun/blob/master/src/shogun/machine/MulticlassMachine.cpp 14:44 yes^2 14:44 hah 14:44 2yes 14:44 hehe 14:44 ok, why the need for subsets? 14:44 because it would be painful 14:44 to create new features 14:44 each time you want to construct ovo machine 14:44 no I meant why needs OvO that? 14:44 sorry doorbell 14:44 hmm because all the classes but some (m,n) pair should be ignored 14:45 re 14:46 any better way would be welcome but I can't see any 14:46 blackburn: so this is 'only' to avoid copying of data around right? 14:47 sonne|work: yes probably 14:47 How does LibSVMMutlicalss do this? 14:47 karlnapf: they have some kind of subset index internally 14:47 ok 14:47 and why not subsetting features? 14:48 karlnapf: possible... 14:48 BUT 14:48 what about model selection then? 14:48 blows up 14:48 can we have subsets of subsets 14:48 mmh 14:48 yeah 14:48 Yes, one solution would be to have multiple subsets 14:49 if we had a function to say train(features, subset) it would work 14:49 like we talked about one year ago or so 14:49 yes 14:49 thats possible on the locked data 14:49 or a function that returns a subset given a subset as input 14:49 but thats not good for OvO 14:49 that might be sufficient ... 14:49 I mean 14:49 so a function that computes a new subset based on the current subset 14:49 modelselection on kernel machines without locking is a pain anyway 14:49 only if you can store the kernel in memory... 14:50 (that was never the case in the applications I had) 14:50 yes true 14:50 ok so subsetting the machine 14:51 that seems to be a good solution, however, we got many subset mechanisms then 14:51 and they basically all take subsets of features 14:51 if there was a way to register multiple subsets on features 14:51 we wouldnt need all that stuff 14:51 ok true 14:52 like give each subset a unique identifier 14:52 and pass it before accessing them 14:52 some dynarray of subsets... 14:52 yes 14:52 but actually that would kill performance 14:52 and how to access? 14:52 maybe one can add subsets as one wants 14:53 features subsets are currently used implicitly 14:53 and then a 'current' subset is computed based on the subsets in the list 14:53 so accessing features uses only this 'current' subset 14:53 maybe current is the wrong term 'merged subset' is better 14:53 oh yes, they would have a  hierachy 14:54 hierarchy 14:54 still the problem: 14:54 A sets subset A 14:54 B sets another subset on that 14:54 B is not ready yet 14:54 A wants to access 14:54 and because A is just setting the subset once and then accesses implicitly this is a problem 14:55 well I would say that B has to free its subset before passing on to A 14:55 otherwise this gets too difficult 14:55 ok 14:55 I do not really understand why you need any hierarchy or other stuff 14:56 then it would be quite straightforward to implement right? 14:56 what are you trying to solve right now? 14:56 blackburn, the problem is that modelselection subsets features 14:56 and if the multiclass svm also does that, there is a kind of conflict 14:56 a "subset on a subset" 14:56 oh 14:56 would be needed, but currently thats not possible 14:57 yes got it 14:57 and having multiple subset mechanisms (features, multiclass, locking) is a bit ugly 14:57 karlnapf: yes I think this is doable 14:57 multiclass do not do subsetting actually 14:57 it subsets features 14:57 or kernel 14:57 nothing moer 14:57 just some 'add_subset' / 'remove_subset' functions 14:57 and whenever add/remove is called the 'merged' subset is updated (which is what we currently have) 14:58 so no need to change the code we have now - just some additions 14:58 subset union subset? 14:58 actually one needs to change all code that sets a subset 14:58 it needs to use add/remove_subset 14:58 each time you add subset it subsets the subset? 14:58 blackburn: it computes the subset of the subset of the subset - yes 14:59 sonne|work: it may break things then 14:59 -!- harshit_ [~harshit@182.68.249.161] has joined #shogun 14:59 we should set a maximum subset level to avoid abuse of this system ;) 14:59 earlier it was assumed that there is only one subset 14:59 blackburn: what does it break? 15:00 sonne|work: no idea but karlnapf could know probably 15:00 you lost me, what? 15:00 karlnapf: blackburn is asking when setting subsets of subsets would break in the current system 15:00 no I think not 15:01 ok then 15:01 one just HAS to remove subsets after doing work on them 15:01 one issue is that add and remove are not interchangeable or so 15:01 yeah so it should be safe 15:01 I mean add applies it on top of that 15:01 but remove removes all of them 15:01 no 15:02 or I got wrong? 15:02 remove removes 1 15:02 you want to do stack there? 15:02 yes 15:02 yes 15:02 sonne|work : I have updated the pull request, I think now its ready to be merged 15:02 good idea probably 15:02 just an array of subsets 15:02 Perhaps this removing could be possibly  done automatically in some destructor 15:02 I thought you wanted to merge subsets actually 15:02 harshit_: Thanks - I will have a look later 15:02 blackburn: yes for performance 15:02 then we would not have seperate subset systems at three places 15:03 we have one 'merged' subset 15:03 well two subsets is ok probably 15:03 which is computed from the array of subsets 15:03 such that we don't have to go through the whole array of subsets all the time 15:03 karlnapf: which three places? 15:03 yes, all the subset instances are saved but everytime the stack is touched, atemporary "current" one  is created for performance thight? 15:03 modsel, locking, multiclass 15:03 well I see no other way for multiclass ovo 15:04 But when we have this new thing, you can just subset the features before training the class pairs 15:05 and then the machines just train on "all" data (from their perspective) 15:05 fight karlnapf fight 15:05 :) 15:05 lol ;) 15:05 I went kick-boxing yesterday, still have Muskelkater, so, no thanks ;) 15:06 karlnapf: actually I think subset stack is a good idea not only wrt to ovo 15:06 I also like it 15:06 so who wants to implement it? 15:06 i.e. what if you want to do cross-validation on subset of your features? 15:06 thats possible then 15:07 blackburn: features -> you mean subset of dimensions 15:07 or subset of examples? 15:07 oh no sorry 15:07 feature vectors 15:07 that would work then 15:07 yes 15:07 subsets of features would be interesting however it is a different system 15:07 that's fine 15:07 you just would have to check for class indices everytime, not just once in the beginning 15:07 is it required btw? 15:08 yes, subsets of dimensions would be cool for feature selection 15:08 but thats probably another story 15:08 I cou?d implement the stuff 15:08 sonne|work: btw what do you think about sparse linear machine? 15:08 karlnapf: please do if you have time ;) 15:09 I can but you would probably be able to test it better 15:09 Its not ultra urgent, is it? 15:09 blackburn: what for? what does it do? 15:09 karlnapf: no, not urgent at all 15:09 sonne|work: lasso 15:09 blackburn: please elaborate 15:10 because my parents are visiting me tomorrow and I only have a few hours per day, and also want to come up with the hypothesis test interfaces this week 15:10 sonne|work: lasso has only a few non zeroes in [w] 15:10 karlnapf: but it should be rather easy 15:10 maybe n4ndo would want to do it? 15:10 or gsomix? 15:10 I mean it would just mean to add these to add/remove functions 15:10 blackburn: no gsomix is busy with python3 15:11 its ok, Ill do it 15:11 and the merge function 15:11 that's all or? 15:11 there might be trouble with x-val 15:11 sonne|work: do you have any other tasks for gsomix? 15:11 ...can be fixed when it occurs :) 15:11 -!- harshit_ [~harshit@182.68.249.161] has quit [Quit: Leaving] 15:11 blackburn: e.g. octave_modular typemap for sparse features 15:11 sonne|work: because I want somebody to do LARS and he would do that 15:11 I see - then let him do this octave 15:12 blackburn: you still didn't explain what sparselinearmachine is! 15:12 gsomix: are you busy? 15:12 sonne|work: yes,yes - what if w is sparse? 15:12 sonne|work, I read the mail yesterday about the parameter selection for GPs, I think there might be some potential for creating new model selection classes 15:13 blackburn: ahh you mean not to store whole w vector 15:13 rather not to compute outputs with whole one 15:13 it is the case of lasso 15:13 blackburn: or one just adds a sparsify_w function ... 15:13 one could generalize the modsel class to use specific evaluation functions (like model evidence or LOO) and then also specify gradients of them and create new GradientDescentModelSelection class or so 15:14 for training it is tough... 15:14 sonne|work: sparse training you mean? 15:14 yes no need to 15:14 karlnapf: you would probably have to help this student then 15:14 yes, I like that kind of stuff 15:15 blackburn: one way would be to then call compress_w and then apply uses the compressed representation of w 15:15 SNOW 15:15 WTF 15:15 SNOW in april! 15:15 hehe :) 15:15 we just had that on sunday 15:15 it is all dirty here 15:15 I got foggy rainy Londonish sky here :) 15:15 and snow again 15:16 what the fog! 15:16 there was snow in germany? 15:16 sonne|work: ok then lars goes to n4ndo probably 15:17 I hope I do not overload him :D 15:17 blackburn: if you know better what gsomix is interested in - tell me - or gsomix tell me :) 15:18 sonne|work: he is at uni right now probably 15:18 I saw him hour ago 15:19 sonne|work: director stuff btw 15:20 sonne|work: what is # of proposals? 15:20 blackburn: he is more fond of that? 15:20 directors I mean 15:21 sonne|work: I do not know 15:21 I'll ask 15:21 or you 15:21 it is tough though and I guess he would need to write an isolated thing 15:21 blackburn: 21 proposals by now 15:22 3 days left... 15:22 I think new candidates should finish some small successful tasks 15:22 yes - even after deadline but before we decide 15:22 that's why I am afraid of sparse matrix and other stuff 15:23 can you estimate men-day's? :D 15:23 man* 15:24 sparse matrix? 15:24 what? 15:24 sonne|work: software eng managers used to estimate man hours and man days :D 15:25 sonne|work: sparse matrix support for octave 15:25 that is very easy (for me) 15:26 so lets say 2 days max 15:26 ok nice 15:27 sonne|work: will you have time to guide him? 15:27 hey guys, Do you know this possibility of approximating a kernel matric using incomplete cholesky factorisation? 15:27 matrix 15:28 I recently used that and found it quite cool, so perhaps there would be some room for it in shogun 15:28 karlnapf: what is *incomplete* cholesky? 15:28 ah let me google that 15:28 sent you an email 15:29 its an extract from the kernel methods book of john shawe taylor and nello crist.... 15:30 :) 15:30 oh I'll check it later 15:31 karlnapf: did you send the proposal alread 15:31 y? 15:31 yes 15:32 hope you like it :) 15:32 karlnapf: this year we are going to receive 20% of last year ones :D 15:32 what do you mean? 15:32 ah like re-applying? :) 15:33 no 15:34 in total count 15:34 karlnapf: only you and me are re-applying afaik 15:34 last year we had like 70 applications - this year it will be around 30 I guess 15:35 really? how does that come? 15:35 this is because we required patches even before applying 15:35 projects too hard? 15:35 and applications/candidates are much stronger 15:35 mmh, was that a good idea then? the number of slots is determined by the number of applications thight? 15:35 not only I think 15:36 yeah from that perspective not - but who can wade through 70+ applications? 15:36 even reading them is *a lot* of work 15:36 sonne|work: especially if they are done by russians 15:36 indeed 15:36 :D 15:36 hehe :) 15:37 and last year we had like 3 reviewers per student app 15:37 and scores / final ranking 15:37 sonne|work: who did the job? 15:37 we split work somehow 15:37 sonne|work: I mean who have participated in that? 15:38 candidates that we were more sure of we wanted were reviewed by more people then 15:38 everyone 15:38 mentors? 15:38 of course 15:38 mentors 15:38 only 15:38 that is how it works 15:38 I know, I am just curious :) 15:38 I wish blackburn / karlnapf you both would be mentors instead of students 15:38 :D 15:38 Hopefully next year :) 15:39 sonne|work: if it was same-rewarded 15:39 I desperately need money, unfortunately 15:39 so expensive here 15:39 no reward unfortuantely 15:39 yes that's why we want to be students :) 15:40 * sonne|work wants to be student to 15:40 too 15:40 -!- pythonroar [d2198538@gateway/web/freenode/ip.210.25.133.56] has quit [Ping timeout: 245 seconds] 15:40 How is work at tomtom btw? 15:40 they have great data... 15:41 and lots of 15:41 and even more data analysis problems 15:41 what are you actually doing there? Like what kind of problems? 15:41 so quite interesting 15:41 sonne|work: I bet your year salary is larger than gsoc reward ;) 15:41 I am working in the HD traffic team - that is the one predicting jams / any kind of live traffic stuff 15:42 sonne|work: btw can I intern there? ;) 15:43 blackburn: you should go to university and become professor 15:43 would be funny 15:44 sonne|work: why should I? 15:44 otherwise we will never have students to work for you on shogun 15:44 lol :) 15:44 hah good reason to devote whole life for useless stuff there 15:44 karlnapf: nothing to laugh here ;) you have the same destiny 15:46 sonne|work: what about you then? 15:46 I left academia... 15:46 we are all converging on growing shogun students ;) 15:47 yes I heard that before but no way back? 15:47 actually we've got one kind of brilliant  haha 15:47 sonne|work: what I think is there is no need to worry we are not the mentors 15:49 karlnapf: I am attempting to grow+educate 2 kids to become shogun developers :D 15:49 first shogun girl? 15:49 we would have to wait at least 15 years more though 15:50 blackburn: I do - every help is welcome (as always) 15:50 sonne|work: you do worry you mean? 15:50 nowadays programmers start 6 15:50 yes 15:50 hehe, your kids 15:50 I got a BASIC interpreter for my AMIGA for my 10th birthday from my father ;) 15:51 sonne|work: I think being mentors would mean we have no time for develop.. 15:51 I got my first computer when I was 7 15:51 And my parents always wanted me to do computer science, but I refused and went studying jazz guitar first ;)   so dont put too much pressure up, sonne|work ;) 15:53 karlnapf: my son is 11 months and want to play games on my android phone :) 15:55 but I don't understand how I got interested in this 15:55 sonne|work: he is also used to multiclass stuff already 15:55 maybe because my parents always wanted to limit computer-time 15:56 hehe, dont let him become a super-nerd ;) 15:56 yes same here 15:56 so I guess that is the key 15:56 tough limits (he wants to break) 15:56 sonne|work: you should deny him to code perfectly later 15:57 * sonne|work thinks that karlnapf would have finished this sub-subsetting business by now if we wouldn't fool around... 15:57 hehe 15:57 I am studying ;) 15:58 and yes deny him to construct theory better than vapnik did 15:58 but youre right, I should stop chatting 15:58 take care guys :) (wave) 15:58 my son, never develop general pattern recogntion theory better than VC and PAC 15:58 sonne|work: that's the thing you should say everyday ;) 15:59 karlnapf: hope to see you in chat more often (idling) 15:59 cu 15:59 yeah 16:00 yes, will leave it on more, uni just stopped so I can be online all day : bye 16:00 snow stopped 16:03 yay! 16:04 karlnapf: bye 16:04 sonne|work: have you used clearcase ever? 16:05 no 16:05 if someone asks you to use it 16:05 better run 16:05 or hide somewhere 16:05 I heard it is legal to kill person that suggests to use clearcase 16:06 -!- pluskid [chatzilla@111.120.76.142] has joined #shogun 16:08 -!- flxb_ [~cronor@141.23.80.206] has joined #shogun 16:09 -!- flxb [~cronor@fb.ml.tu-berlin.de] has quit [Ping timeout: 264 seconds] 16:11 -!- flxb_ is now known as flxb 16:11 sonne|work: that paper I showed to you is to appear in shity 'computer optics' 36(2) :D 16:15 computer optics??? 16:16 yes that is the name of journal.. 16:17 sonne|work: do you think I am crazy? I am not welcome to Russia :D 16:17 am not,* 16:18 the world is crazy... 16:18 sonne|work: it is a kind of local journal with rather easy submission 16:20 sonne|work: and I needed it to 'power up' my bachelor's work 16:20 the most funny thing there is abstracts 16:21 'In this paper is considered a method of forming a sequence of reliefs based on the specified reference terrain images by using a criterion based on computation a conjugation indexes between a current reliefs and a reference reliefs. An example of the implementation is given.' 16:21 some lower machine learning conference would also have worked 16:21 anyways got work to do cu 16:21 sonne|work: can you believe it is *real* abstract? 16:21 oh sure 16:21 see you 16:21 conference would work but no time for that 16:21 blackburn i've decoded the email!!! :DDD 16:23 wiking: hahah 16:23 took a while right? 16:24 well yeah 16:24 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun 16:24 n4nd0: can you remind me what is your 'contribution roadmap'? ;) 16:25 wiking: so what does he want? 16:25 a little bit of change 16:25 not so much 16:25 basically we would remove the whole CLatentFeatures 16:25 blackburn: what do you mean, contributions in shogun? 16:25 n4nd0: yes 16:25 and just have CFeatures 16:25 but they would contain all the features: x,y,h 16:26 wiking: and include things to latentlabels? 16:26 ah 16:26 how to separate it when? 16:26 well you don't need to 16:26 then* 16:26 blackburn: from the very beginning, CPLEX compilation, Mahalanobis distance, QDA, OvO, SPE 16:26 optimizer doesn't need to know it 16:26 blackburn: not that much :S 16:26 what part of the feature is what 16:26 the only thing that after the results 16:26 the user of the tool will have to know what is what 16:27 n4nd0: sorry I meant what is next? 16:27 blackburn: aham! 16:27 wiking: ah I see 16:28 blackburn: so first we should manage to get merged SPE, did you see I updated it? 16:28 and then still one _could_ define an own function to speed up the optimization algo 16:28 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] 16:28 wiking: that is hardly implied from the msg :D 16:28 blackburn: later, what about the OvO issue? I see you have been discussing long about it 16:28 but if it's not provided it should just do a brute force search 16:28 i'm just wondering now 16:28 n4nd0: we shall wait for new subset stack concept ready 16:29 whether the whole class should be inherited.... i.e. have a protected virtual function for finding the argmax of the latent variable 16:29 blackburn: ok 16:29 n4nd0: I've got a new task for you thought, LARS 16:29 interested?;) 16:29 wiking: whole class of? 16:29 -!- flxb [~cronor@141.23.80.206] has quit [Remote host closed the connection] 16:29 or have a class that only has a function like this.. and that class should be inherited and redefine that function for finding the argmax 16:29 blackburn: LatentLinearMachine 16:29 aha 16:30 -!- flxb [~cronor@fb.ml.tu-berlin.de] has joined #shogun 16:30 blackburn: tell me about it 16:30 any input on it? 16:30 i'd rather go with the latter option 16:30 blackburn: http://www.machinelearning.org/proceedings/icml2005/papers/053_GeneralizedLARS_Keerthi.pdf?? 16:30 it feels better for me 16:30 n4nd0: generalized? simple would work as well :) 16:30 wiking: then go with latter ;) 16:31 blackburn: give me a pointer if so 16:31 n4nd0: scikits learn has one 16:31 n4nd0: I'll take a look on spe later 16:31 blackburn: cool 16:31 n4nd0: maybe this one? http://www.stanford.edu/~hastie/Papers/LARS/LeastAngle_2002.pdf 16:32 n4nd0: https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/linear_model/least_angle.py 16:32 yes this one 16:32 you wish to include LARS in shogun for, solving LASSO ? 16:32 yes 16:32 regression, cool 16:32 LARS is cool, but I heard that there are many more efficient algorithm for L1 regularized problems? 16:33 aham, e.g.? 16:33 pluskid: I have not much expertise with linear regression, what would you suggest? 16:33 I do not know the literature (not familiar with optimization), actually, I always use LARS in my own work :p 16:34 blackburn: I heard some discussion about something like ??? gradient descent algorithms, but I do not understand the details 16:35 I know gradient descent 16:35 But I do think LARS has an advantage that it gives you the whole path 16:35 I think many other algorithms only gives you the solution for a particular \lambda (regularization coefficient) 16:35 Maybe any of you could check out this package: http://www.public.asu.edu/~jye02/Software/SLEP/ , if you are good at optimization 16:36 wow 16:37 nice reference 16:37 hmm that's something I should take care of 16:38 group based lasso is of interest for my project 16:38 hmm, I think those ideas interesting, too. But I never used them in practice yet, except simply L1 16:39 simple 16:39 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 16:40 Trying to find the discussion thread about LASSO optimization I have read before 16:41 Got it! 16:43 I list some references here, in case any one would be interested in. 16:44 They said that this algorithm is very fast: http://www-stat.stanford.edu/~jhf/ftp/glmnet.pdf 16:44 And for really huge data, online learning algorithms also exist: 16:44 http://jmlr.csail.mit.edu/papers/v10/langford09a.html 16:44 http://jmlr.csail.mit.edu/papers/v10/duchi09a.html 16:44 That's it. :) 16:46 pluskid: thanks! nice references 16:46 Time for sleep here. 16:46 good night/afternoon/morning guys! 16:46 -!- pluskid [chatzilla@111.120.76.142] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 11.0/20120312181643]] 16:47 oh 16:47 good night 16:47 good sth ;) 16:49 so after the ideas that you suggested yesterday and the new one today, LARS ... 16:51 let's go for LARS because it is the newest :) 16:51 n4nd0: what have I suggested yesterday? 16:52 the tree 16:53 the cover tree for knn too 16:53 aha 16:53 parzen windows 16:54 ... 16:54 :P 16:54 n4nd0: what about parzen window one with covertree internally? later we would merge good design from it to knn 16:55 blackburn: idk if you're serious now 16:55 yes I am 16:56 so parzen windows can be implemented with cover trees?? 16:56 good question 16:56 probably not 16:58 :D 16:58 n4nd0: actually it could if there was some threshold based covertree 17:00 but it would require patching covertree that is present in shogun 17:00 blackburn: let's do that new threshold based covertree and write a paper :P 17:02 I am afraid it is way too simple idea 17:02 original covertree supports it IIRC 17:02 blackburn, is covertree only for euclidian distances? 17:05 or can it work with any distance? 17:05 sonney2k: any 17:05 it requires nothing but pairwise distances 17:05 ok then n4nd0 this definitely is my favourite then covertree knn :) 17:06 i'll focus on that then sonney2k 17:08 shogun: Harshit Syal master * r85ffdcd / src/shogun/classifier/svm/NewtonSVM.cpp : removed unwanted SG_SPRINT - http://git.io/_YLBPA 17:09 shogun: Soeren Sonnenburg master * ra934dcb / (7 files in 5 dirs): 17:10 shogun: Merge pull request #409 from harshitsyal/master 17:10 shogun: Newton SVM - http://git.io/EAZk4Q 17:10 I'll be back later, see you guys 17:15 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] 17:15 shogun-buildbot, ? 17:15 -!- sidsniper [~piggy@14.139.82.6] has joined #shogun 17:17 hi blackburn 17:19 hi 17:19 how can i test dim red algos? is there any data? 17:19 sidsniper: what kind of testing you want to do? 17:19 reducing dimension of feature vector 17:20 how did you test the correctness of algoriths implemented? 17:20 manually ;) and compared to other implementations 17:21 so there must be some dataset which u had tested them upon? 17:21 yes, I tested it on swissroll, ORL and MIT-CBCL 17:22 swissroll is the simplest test I'd say 17:22 okay 17:23 so these are availaible online right? 17:23 sidsniper: yes 17:23 ok thanks :) 17:23 sidsniper: have you doubts about its correctness? 17:23 i have yet to test them..i m sure they will be correct :) 17:24 I'm going to implement unit-tests as well.. 17:24 if you can give me links from where you have implemented them it would be great 17:24 ORL and CBCL are face database right? 17:26 sidsniper: implemented what? 17:26 sidsniper: yes 17:26 implemented the algorithms 17:26 dim red 17:27 hmm I mainly refer to dr toolbox 17:27 http://homepage.tudelft.nl/19j49/Matlab_Toolbox_for_Dimensionality_Reduction.html 17:27 nice :) 17:28 i have to implement some of them for my project 17:28 can u suggest how to proceed? 17:28 i mean which are the easier ones 17:28 sidsniper: LLE is pretty simple 17:30 diffusion maps are as well 17:30 they are all actually pretty simple 17:30 but pretty slow in case of straightforward implementation 17:31 so implementation part is hard? 17:31 depends on speed you need 17:31 I had to implement superlu and arpack wrappers 17:31 and to parallelize it 17:31 but in matlab it could be 20-liner 17:32 okay 17:32 as for correctness 17:33 but we have to do it in R language 17:33 they are all ill-posed 17:33 will that be tough? 17:33 no idea 17:33 i saw that shogub also suppports R 17:33 shogun* 17:33 how much data you have? 17:33 -!- harshit_ [~harshit@182.68.249.161] has joined #shogun 17:33 cureently i dont have 17:33 harshit_: congrats your code is merged ;) 17:34 i will first try on swissroll as suggested by u 17:34 sidsniper: yeah try ;) 17:34 hey blackburn: is there any way to get precision , recall, accuracy etc on tested dataset in shogun itself ? 17:34 and thanks i didnt check it , 17:35 harshit_: yes, ContingencyTableEvaluation 17:35 ok heading home finally 17:35 see you 17:36 -!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] 17:36 github is strange doesn't notify you for getting your code merged ! 17:36 harshit_: which project are u applying for? 17:37 sidsniper :Its a collection of things actually 17:39 sonney2k, sonne|work slap. 17:39 2 projects for which i am applying are: integrate regression in liblinear + implement some cdotfeatures 17:40 and one is trees + liblinear 17:40 sidsniper : what about you ? 17:40 okay 17:40 dman 17:47 sonney2k, sonne|work I want to finish my proposal today. That's all I want. :) 17:50 however I can do smth else. 17:50 build #455 of octave_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/455  blamelist: harshit.syal@nsitonline.in 18:00 build #442 of csharp_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/442  blamelist: harshit.syal@nsitonline.in 18:04 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun 18:08 build #431 of lua_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/431  blamelist: harshit.syal@nsitonline.in 18:09 -!- khayyam [~chatzilla@189.166.217.6] has joined #shogun 18:11 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 18:13 build #434 of java_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/java_modular/builds/434  blamelist: harshit.syal@nsitonline.in 18:13 n4nd0: hey , do you know why all these build are are failing to compile ? 18:15 As in does shogun-buildbot compiles every commit that i made earlier 18:15 harshit_: so shogun-buildbot compiles commit by commit 18:15 harshit_: exactly, can it  be that some of those commits where not complete? 18:16 I think there were some commits that I made earlier,in which there were some conflicts 18:17 I can see this error ^ 18:17 make[1]: *** No rule to make target ../../shogun/classifier/svm/NewtonSVM.h', needed by modshogun_wrap.cxx'.  Stop. 18:17 So what do I need to do now 18:17 for me it looks like NewtonSVM.h was missing at that moment from the interfaces 18:17 but I am not sure 18:17 harshit_: don't worry about it 18:17 harshit_: it should compile the following commits soon and recover the state 18:18 ohkay got it 18:18 build #430 of python_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/430  blamelist: harshit.syal@nsitonline.in 18:18 harshit_: you can monitor it a bit if you are interested 18:18 http://shogun-toolbox.org/buildbot/waterfall 18:18 just if you are curious to know how it works 18:19 I find it quite cool 18:19 (I think I like everything that is colorful :P) 18:19 ^ dont blame me, :( 18:20 So that was the reason why blackburn said to make big commits rather than small ones 18:20 looks great 18:21 exactly 18:21 -!- naywhaya1e is now known as naywhayare 18:21 there's another way it seems 18:21 that sonney2k proposes using git commit -amend 18:22 I don't know how to use it yet though 18:22 what does -amend do ? 18:22 build #430 of r_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_modular/builds/430  blamelist: harshit.syal@nsitonline.in 18:22 I have not really gone into it yet, but I think it is used to replace already published-commits 18:24 take a look to man git commit 18:24 oh great, I think we need to use it more often then 18:24 and search for --amend 18:24 yeah doing the same 18:25 build #426 of ruby_modular is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/ruby_modular/builds/426  blamelist: harshit.syal@nsitonline.in 18:27 -!- harshit_ [~harshit@182.68.249.161] has quit [Quit: Leaving] 18:30 -!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun 18:34 build #456 of octave_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/456 18:35 build #443 of csharp_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/443 18:42 great, our friend shogun-buildbot is healthy again 18:42 build #432 of lua_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/432 18:48 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 18:48 build #435 of java_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/java_modular/builds/435 18:58 -!- flxb [~cronor@fb.ml.tu-berlin.de] has quit [Quit: flxb] 18:59 -!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 276 seconds] 19:02 -!- PhilTillet [~Philippe@tillet-p42154.maisel.int-evry.fr] has joined #shogun 19:03 Hey!! 19:05 :) 19:05 hi! 19:05 build #431 of python_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/431 19:06 sup? 19:06 bah not much 19:06 what about you? 19:06 our friendly shogun-buildbot is almost recovered again :) 19:07 lol :D 19:07 well not much 19:07 i'm tired 19:08 and hungry 19:08 dinner time? 19:09 well 19:09 sort of 19:09 but can't have dinner 19:09 the canteen is not opened yet 19:10 aham 19:11 build #431 of r_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_modular/builds/431 19:12 -!- harshit_ [~harshit@182.68.249.161] has joined #shogun 19:15 build #427 of ruby_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/ruby_modular/builds/427 19:19 -!- Juggy [~Razor@14.139.82.6] has joined #shogun 19:20 Hi blackburn 19:21 Can you give me the link for tutorial on Dimensionality Reduction in shogun? 19:21 @blackburn: are you there ? 19:33 -!- harshit_ [~harshit@182.68.249.161] has quit [Ping timeout: 252 seconds] 19:40 -!- harshit_ [~harshit@182.68.249.161] has joined #shogun 19:44 -!- PhilTillet [~Philippe@tillet-p42154.maisel.int-evry.fr] has quit [Ping timeout: 264 seconds] 19:46 harshit_, I am doing some whitespace cleanups and will commit soon - please have a brief look then. I hope this helps to make clear what I had in mind coding style wise 19:46 sonney2k: I am sorry to insist but, did you have time to take a look at multiclass proposal? 19:47 sonney2k, moin 19:50 n4nd0, sorry not yet 19:50 gsomix, so how is it going? 19:50 sonney2k: all right, I am sorry that I rushed 19:51 shogun: Soeren Sonnenburg master * rcfab94a / (2 files): whitespace and coding style fixed in newtonsvm - http://git.io/wnQ-rw 19:52 sonney2k, i'm working on my proposal and python3 interface now. 19:54 sonney2k, are there other tasks for me? 19:55 harshit_, please have a look at http://git.io/wnQ-rw 19:56 -!- puneetgoyal [~puneet@115.242.16.126] has joined #shogun 19:56 gsomix, most important now is to write your proposal 19:57 gsomix, apart from that  it depends a bit what you are most interested in doing 20:00 I guess python3 stuff is close to final, so other simple task is to write swig typemap for octave modular 20:00 -!- PhilTillet [~Philippe@npasserelle10.minet.net] has joined #shogun 20:00 sonney2k : sorry having issues with Internet 20:01 yeah I've seen... 20:01 harshit_, btw can you write some more description about newton svm in the class description and mention the paper there? 20:02 harshit_, ahh and before I forget - you didn't mention how shogun's newtonsvm performs on covertype data set vs. matlab version 20:03 would be interesting to know 20:03 sure, i'll do it today 20:03 sonney2k, ok. i will consider this issue later. 20:04 sonney2k, what are the "most used" CKernel children? DotKernel I suppose, but any other? :p 20:04 and for comparison there was a problem running the dataset on matlab version, so i compared it with liblinear results on that dataset 20:04 liblinear took about 3-4 sec to train vs my algo which took about 6-8 sec 20:05 and the final weight matrix was almost similar 20:05 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 20:05 gsomix, please ask questions / update us - I am not really sure if the tasks you have are too boring for you. 20:06 PhilTillet, the most used kernel is the gaussian  kernel 20:06 sonney2k : so after inclusion of documentation in NewtonSVM.cpp, do i need to issue a new pull request ? 20:06 -!- PhilTillet_ [~Philippe@tillet-p42154.maisel.int-evry.fr] has joined #shogun 20:06 harshit_, sure 20:06 PhilTillet_, the most used kernel is the gaussian  kernel 20:07 harshit_, what was the problem on matlab? 20:07 yes, Actually every DotKernel could be more or less implemented the same way on OpenCL 20:07 but might be different for some other kernel classes 20:07 :p 20:07 brb eat :D 20:07 sonney2k: matlab said there is some problem with minres() function when i used that dataset 20:08 but otherwise with other dataset it ran perfectly 20:08 -!- PhilTillet [~Philippe@npasserelle10.minet.net] has quit [Ping timeout: 250 seconds] 20:09 harshit_, and if you use a subset of this data set (say first 10000 examples only?) 20:09 I think that dataset is pretty big 20:09 to handle 20:09 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] 20:09 oops forgot to do that ! 20:09 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 20:09 PhilTillet_, dotkernel is most important... / maybe a function to compute whole kernel/distance matrix based on opencl would be nice too 20:10 I'll do it today and report you, for now preparing my gsoc proposal ( i think i am running late on that) 20:10 harshit_, yes that is most important true 20:10 sonney2k: after submitting my proposal i want to start working on lbp features implementation . 20:11 do you think opencv's implementation is portable ? 20:12 harshit_, we have one in shogun that needs fixing - matlab / c code is avilable 20:13 sonney2k: LBP features is already in shogun ! I thought that needs to be implemented from starting 20:15 harshit_, it is in unfinished state 20:16 anyways that is great I can start working on it ~ 20:16 please just give me any link to the matlab/c code from which it was ported 20:17 harshit_, when it is in you can do some nice demo using opencv and these LBP features - to detect gender of a person based on an image of the face 20:17 sonney2k : actually I have done something like that using deep learning "sparse autoencoders" 20:18 I think it would be easy for me 20:18 great thought 20:18 karlnapf, around? 20:20 sonney2k, boring problem - this is also the problem. :) 20:20 sonney2k, yes hi 20:20 whats up? 20:20 karlnapf, would you mind if I give you github commit rights - I mean you are a core developer without this right for some time now... 20:21 you can still do the same workflow 20:21 why should I mind this? :) 20:21 Should I have a look at people's requests then? 20:21 karlnapf, I knew that you like to break things 20:22 karlnapf, if you have time - that's welcome 20:22 sure, of course then 20:22 I will still ask you guys to take a look on stuff which involves larger changes 20:22 sonney2k, I am just wondering what is our favorite way of doing intending in python, spaces or tabs? 20:24 Juggy: still there? 20:24 sonney2k: finally! 20:25 Juggy: http://shogun-toolbox.org/edrt/tutorial.pdf probably that one, it is unaccurate though 20:25 in* 20:26 PhilTillet_: btw I have something really big to propose 20:28 I have some collection of dimensionality reduction algorithms there.. ;) 20:29 karlnapf, https://github.com/shogun-toolbox - welcome to the team but now officially :D 20:29 nice one :D 20:29 thanks 20:29 sonney2k: what a long way home 20:29 -!- gsomix [~gsomix@188.168.5.227] has quit [Read error: Operation timed out] 20:29 karlnapf, well *I* do tabs as usual 20:29 I prefer tabs as well btw 20:30 blackburn, sing some russian love song for heiko please - to celebrate this event! 20:30 yeaaah :) 20:30 sonney2k: love song? like 'it is ok to be gay'? 20:30 * sonney2k turns the volume to the max! 20:30 blackburn, gogo 20:30 blackburn, no sth. like ~~~oh~~~heiko~~~in~~~your~~~code~~~ohh~~~we~~~trust~~~ 20:31 ??????????? ???? ??????? ???? ?????? 20:31 karlnapf, regarding spaces - it might be that python3 wants spaces ... not sure 20:31 ? ?????????? ? ????????? ???? ????? 20:31 sing louder blackburn, cant hear you 20:31 ???? ??????? ?? ???? ???????? 20:31 ?? ??????????? ????????????????????????! 20:31 sonney2k, ok, but we can convert, just asking 20:32 kind of old love song hahaha 20:32 just like sonney2k asked 20:32 I love these letters 20:32 want to learn Russian at some point 20:32 funny it is a war song probably hahah 20:32 WW2 yes 20:32 Ill double check with google translate 20:32 karlnapf: would be pretty senseless 20:32 Gr?ne Ahorn Riegelahorn, aber ;) 20:33 *rotflbtc* 20:33 but I can heart it, the friendly google woman sings it 20:33 http://www.youtube.com/watch?feature=player_detailpage&v=CXMEto5NXb8 20:34 karlnapf: this one is better hah 20:34 -!- sidsniper [~piggy@14.139.82.6] has quit [Ping timeout: 250 seconds] 20:34 nice 20:34 karlnapf: why do you want to learn russian? 20:35 Want to learn another language which is not romanian 20:35 and I want to travel to the place 20:35 karlnapf: you are very welcome btw ;) 20:36 -!- Juggy [~Razor@14.139.82.6] has quit [Ping timeout: 260 seconds] 20:36 thanks blackburn 20:36 its very nice to be in your team guys 20:36 karlnapf: seriously I would like someone of you to visit me here 20:36 I was also serious, really wanna go there, so perhaps next year or so, well meet 20:37 sonney2k: let you organize shogun code sprint at berlin? ;) 20:37 also a nice idea 20:38 -!- gsomix [~gsomix@188.168.2.47] has joined #shogun 20:38 at my flat with my kids jumping around - why not 20:40 how big is your flat? ;) 20:40 no but seriously we could do that from gsoc money 20:40 hehe, well I know many people to stay at 20:40 yes would be kind of cool to meet 20:41 blackburn, how is your german? :) 20:41 karlnapf: ~ to your russian ;) 20:41 ich bin mude 20:42 ich bin immer muede 20:42 karlnapf: wo bist du? 20:42 ich bin in england 20:42 sonney2k: ??? ????? 20:42 ich bin sehr m?de 20:42 ???? 20:43 yes I know umlaut is there 20:43 too lazy to type it 20:43 ? 20:43 how to type it on russian keyboard? 20:44 karlnapf: do I remember that correctly - u mit dem umlaut = ? 20:44 ? 20:44 using compose key 20:44 u+right alt+" 20:44 yes 20:44 karlnapf: I had been studying german up to 7th grade at school 20:45 another very strong proposal got submitted... many are from phd students this time 20:45 for two years probably 20:45 sonney2k: w/o contributions? 20:45 ah really, thats more than my russian 20:45 which project? 20:45 libqp 20:45 >> phd students *__* 20:45 yes, phd all over the place, gsoc is very academic 20:46 my belief is proposal w/o contributions would not work 20:46 karlnapf, last year that was not the case 20:47 sonney2k, probably means that stuff is more advanced, or they checked last years projects or so 20:47 I mean if you dont do machine learning the stuff is pretty tough 20:48 even if you do its not trivial 20:48 many things this year are really tough 20:48 for example for SO we didn't have any proposal last time 20:48 this year quite a bit 20:48 this year we have mighty n4nd0 20:49 SO? 20:49 structured ouput 20:49 tput 20:49 ah ok 20:49 yes, cool thing is that 20:49 n4nd0: btw checking your spe right now 20:50 anyway we have 9 mentors so max slots is 9 20:52 lets hope to get all that ! 20:52 uh 20:52 oh 20:52 pretty impossible probably 20:52 karlnapf, actually in retrospect I am happy that we got just 5 slots last year 20:53 it really was a lot of work 20:53 sonney2k, ok I believe that 20:53 this year though we have n4nd0, blackburn, karlnapf and others helping others 20:53 it is not soo much all on my shoulders any more 20:53 yes, you did almost everythign last year 20:53 so maybe this process can scale 20:53 and actually students this year all submitted github pull requests before 20:54 with pretty nice work that goes deep into shogun 20:54 last year we were switching from svn -> git in the process 20:55 this year we got the buildbots 20:55 we were doing conversions like double* -> sgvector 20:55 we created buildbots 20:55 etc 20:55 -!- flxb [~cronor@g231218093.adsl.alicedsl.de] has joined #shogun 20:55 lots of infrastructure that is now there and helps to keep the project in a sane state 20:55 to conclude - we should be realistically able to handle 9 students 20:56 sonney2k: what was there before gsoc 2011? 20:56 sonney2k, wow that would be amazing. 20:56 do you know how google allocates slots? 20:56 only in terms of application # 20:56 or also other stuff? 20:56 karlnapf: no, there are some criteria 20:58 I'll refer in a min 20:58 karlnapf: http://code.google.com/p/google-summer-of-code/wiki/NotesonStudentAllocations 20:59 -!- Juggy [~Razor@14.139.82.6] has joined #shogun 20:59 -!- sidsniper [~piggy@14.139.82.6] has joined #shogun 20:59 blackburn: cool man, I was having dinner, tell me if I should do / say sth 21:01 karlnapf: to be fair we fit 1), 2), 4) 21:01 sonney2k, just trying out a request 21:01 does the workflow change? 21:01 do I still make pull requests and then merge them myself? 21:01 n4nd0: I like it pretty much 21:01 karlnapf: I think this type of commits would be better to be direct 21:02 ok so only pull request for larger stuff 21:02 and direct means I clone the actual shogun repo? 21:03 karlnapf: yes 21:03 karlnapf: ah yes 21:03 you need to push to 21:03 ok then 21:03 google, shogun and github 21:04 karlnapf, btw you can from now on just do (C) Heiko S... no need to mention TU or so any longer 21:04 sonney2k: btw 21:04 I had idea before 21:04 Written and Copyright now are pretty same 21:04 blackburn, karlnapf I disagree 21:04 sonney2k: about? 21:04 I am very much favoring pull requests 21:05 and then have someone else looking at it 21:05 sonney2k: for small fixes? 21:05 but yeah I know blackburn you disagree 21:05 blackburn, for small fixes pull request + self merge 21:05 this takes more time.. 21:05 ok like examples and stuff, but its always good to get stuff reviewed when its more complicated 21:06 and avoids quite a bit of stupid errors 21:06 sonney2k: these stupid errors emerge pretty rarely.. 21:06 blackburn, I have a talent of finding stuff like this ;) 21:07 I still don't like this kind of self-merge 21:07 shogun: Heiko Strathmann master * rf5c30fd / examples/undocumented/python_modular/modelselection_grid_search_kernel.py : added example for modelselection of different kernels - http://git.io/y1hxQA 21:07 shogun: Heiko Strathmann master * r67c3782 / examples/undocumented/python_modular/modelselection_grid_search_kernel.py : fixed copyright and excluded TUB - http://git.io/2pNQHw 21:07 shogun: Heiko Strathmann master * r2801f8d / examples/undocumented/python_modular/modelselection_grid_search_kernel.py : 21:07 shogun: Merge pull request #417 from karlnapf/master 21:07 shogun: new example - http://git.io/ehsJzw 21:07 karlnapf: sonney2k: one more thing I would like to suggest 21:08 is to use #XXX stuff 21:08 for example if you are pretty ready to merge PR no need to wait 21:08 you may commit with 'blabla Closes #XXX' 21:09 and actually we can use not only PRs but issues as well 21:09 we can commit with #XXX to incorporate commit to this issue 21:10 -!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] 21:10 I believe it makes more sense to group commits with issues 21:10 ...if we used issues 21:10 let me show how it works 21:10 -!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun 21:11 https://github.com/shogun-toolbox/shogun/issues/418 21:11 build #198 of nightly_none is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/198 21:11 here we would have docu improvements 21:11 karlnapf: about the answer to the issue, can you elaborate a bit more on that? what is the idea then? not to serialize SGNArray stuff? 21:13 karlnapf: I think that might be a problem because for example QDA has some members of that type 21:13 n4nd0, no that was just a general thought 21:13 this is possible (and very nice I think) 21:13 but its more complicated than on the first impression 21:13 and once I add new commit 21:14 karlnapf: I see, I have no idea how serialization stuff works in shogun 21:14 if you dive into the load_serializable method of CSGObject you will that there are many places where these types have to be handled 21:14 n4nd0, is it important that you need that? 21:15 shogun: Sergey Lisitsyn master * rb1b17cb / src/shogun/converter/DiffusionMaps.h : Added note about lapack performance with diffusion maps #418 - http://git.io/4zEjlQ 21:15 it appears in #418 21:15 https://github.com/shogun-toolbox/shogun/issues/418 21:15 karlnapf: you mean if I need to use that type in QDA? 21:15 sonney2k: karlnapf: what do you think? 21:15 n4nd0, yes 21:16 karlnapf: at the beginning I was simply using SGMatrix * 21:16 karlnapf: the idea is that I need a list of matrices 21:16 blackburn, I think thats nice, but is it easy to handle? 21:16 karlnapf: but that SGMatrix* turned out to be a bit problematic also with this issue, the serialization 21:17 karlnapf: yes all you need is to recall # of issue 21:17 n4nd0, i see 21:17 karlnapf: so sonney2k proposed to use SGNDArray instead 21:17 blackburn, yeah I like that - we had this same feature with trac before but also there rarely used it - just because we never really managed to use issues all the time 21:17 sonney2k: that would be faster for doc fixes and other small fixes 21:17 too much pull requests would slow down things I think 21:18 n4nd0, ok, probably good idea then 21:18 blackburn, I am saying lets use it but I don't know how who will file issues all the time 21:18 not sure if we are enough developers yet for that stuff 21:18 n4nd0, I just hate to touch the migration stuff ;), but should be doable, if you want, take a look 21:18 karlnapf: so the thing is that right now serialization in QDA wouldn't work right? 21:18 n4nd0, yes 21:19 sonney2k: hmm just write somewhere (doc fixes goes to #418) 21:19 not too bad now, but should be done on the long run 21:19 karlnapf: I will write that in the doc. at least 21:19 would be nice to make some other keyword referencing 21:19 karlnapf: yeah, I agree, sooner or later it must be done 21:19 blackburn, can I close an issue from via a commit message? 21:19 n4nd0, yes, perhaps during the summer :) 21:20 sonney2k: closes #418 21:20 k 21:20 sonney2k: just like pull request 21:20 but no need to switch between fork and origin 21:20 sonney2k: fixes, fixed, fix, closes, close, closed 21:21 and of these words 21:21 any* 21:21 ok have to clean up my damned traffic sign recognition paper be back in a hour or so 21:24 sonney2k, http://pastebin.com/DuSe6Qs1 at now. my last fixes and 2to3. 21:24 blackburn: see you later then, maybe we can finish with SPE then? 21:24 n4nd0: hmm one more thing 21:24 please add reference 21:25 and I'll check it once again this night and merge 21:25 blackburn: for the doc? 21:25 yes 21:25 blackburn: ok 21:25 -!- khayyam [~chatzilla@189.166.217.6] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 11.0/20120312181643]] 21:25 blackburn: I am a bad guy ... I am always like saying TODO doc :P 21:25 heh better have todo than forgot 21:25 gsomix, you need to install liblzma-dev 21:26 -!- blackburn [~qdrgsm@83.234.54.186] has quit [Quit: Leaving.] 21:26 gsomix, and libjson0-dev 21:26 sonney2k, done 21:31 blackburn can you give me the link to the tutorial on dim red? 21:36 sidsniper: I think blackburn is out now working on his stuff 21:37 sidsniper: what tutorial do you want? 21:37 the one that is in shogun 21:38 tutorial showing how to use those algos 21:38 ok, have you taken a look to the examples? 21:38 maybe that is enough for your needs? 21:38 sonney2k, but doesn't work. hmm 21:40 on python3. 21:40 and on python2. 21:41 reboot 21:41 -!- gsomix [~gsomix@188.168.2.47] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] 21:41 -!- Afan [~Afan@c-76-29-50-209.hsd1.il.comcast.net] has joined #shogun 21:48 Hello again 21:48 :) 21:48 -!- PhilTillet_ is now known as PhilTillet 21:51 -!- gsomix [~gsomix@83.234.169.122] has joined #shogun 21:52 sonney2k, works. 21:53 -!- uricamic [5546c513@gateway/web/freenode/ip.85.70.197.19] has joined #shogun 21:53 is there who knows about the DR tutorial in Shogun? 22:01 I am not able to fine :( 22:01 *find 22:01 is there a list of implemented algorithms ????? 22:02 you can check them under the directory converter 22:03 http://www.shogun-toolbox.org/doc/en/current/classshogun_1_1CConverter.html 22:04 there you can a find a list in doxygen doc 22:04 but doc might not be up to date 22:05 n4nd0, strong MC proposal... 22:07 yours I mean 22:08 sonney2k: oh thank you :) 22:08 -!- PhilTillet_ [~Philippe@tillet-p42154.maisel.int-evry.fr] has joined #shogun 22:08 -!- PhilTillet [~Philippe@tillet-p42154.maisel.int-evry.fr] has quit [Remote host closed the connection] 22:08 -!- gsomix [~gsomix@83.234.169.122] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] 22:21 sonney2k, do you have any idea of what "big idea" blackburn was talking about? 22:38 big idea? 22:47 I guess I missed sth 22:47 except he is dreaming of world domination again ;-) 22:47 PhilTillet_: btw I have something really big to propose 22:49 I have some collection of dimensionality reduction algorithms there.. ;) 22:49 XD 22:50 many dim reduction methods are expensive so it sounds like he wants some opencl magic :) 22:51 haha 22:51 PhilTillet_, btw how are you progressing with your patch? 22:51 They are all about eigenvalues right? 22:51 Well, pretty hard 22:51 I cleaned my source 22:52 so that there is a clear OpenCL structure 22:52 to compile programs etc 22:52 and now I copy with direct memcopy, without using ublas 22:52 I mean, for now my patch is mostly the opencl code 22:54 and basically a shogun::ocl::svm::init() 22:54 which compiles all the opencl code 22:54 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 23:03 memory management is one of the first thing in my GSoC Timeline, but won't be able to do it for the patch 23:04 :D 23:04 -!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun 23:18 -!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] 23:18 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 23:18 -!- Marty28 [~Marty@158.181.76.57] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 11.0/20120310010446]] 23:18 -!- karlnapf [~heiko@host86-181-9-49.range86-181.btcentralplus.com] has quit [Quit: Leaving.] 23:22 sonney2k, hope I'll be able to submit an opencl patch (not perfect from a design point of view XD) soon 23:28 shogun: Soeren Sonnenburg master * r21ae0ec / src/shogun/classifier/svm/NewtonSVM.h : add ifdef HAVE_LAPACK for NewtonSVM - http://git.io/iX7SlQ 23:31 -!- puneetgoyal [~puneet@115.242.16.126] has quit [Ping timeout: 246 seconds] 23:32 sonney2k, I'll have some design question though :p 23:36 -!- uricamic [5546c513@gateway/web/freenode/ip.85.70.197.19] has quit [Quit: Page closed] 23:51 -!- PhilTillet_ [~Philippe@tillet-p42154.maisel.int-evry.fr] has quit [Quit: Leaving] 23:57 -!- harshit_ [~harshit@182.68.249.161] has quit [Ping timeout: 244 seconds] 23:57 --- Log closed Wed Apr 04 00:00:19 2012