Open in new window / Try shogun cloud
--- Log opened Wed Mar 28 00:00:03 2012
--- Day changed Wed Mar 28 2012
blackburnI am not shy with my alcohol addiction00:00
n4nd0sonney2k: I realized he gets funnier at nights :)00:00
@sonney2kblackburn, tell n4nd0 about your bear!00:00
blackburnoh its huge00:00
n4nd0what kind of bear are talking about?00:00
blackburnmy bear00:00
blackburnhe drinks a lot of vodka00:00
@sonney2kthe one from siberia00:00
blackburnand eats people00:00
@sonney2kkind of a pet blackburn has at home00:01
n4nd0you're kidding guys00:01
n4nd0I already tried that joke with polar bears with Spanish friends when I came to Sweden00:01
@sonney2kI think he walks out his bear daily - scaring away people00:01
blackburnwanna see my best photo?00:02
n4nd0haha wtf00:03
n4nd0it was just a set-up for the photo?00:03
blackburnunfortunately no00:03
blackburnI should be ashamed of it probably but it is funny as hell00:04
@sonney2kthat's before blackburn is getting up - blood level too high (not enough alcohol-aka-fuel in blood yet)00:04
blackburnactually it was new year celebrations00:05
n4nd0no too bad for new years then00:06
blackburnI've got one more funny photo from school00:06
blackburnn4nd0: this time yes - it is a set-up ;)00:07
n4nd0a Russian friend told me that people are more lucid when they drink ... then I started to believe stories about vodka :)00:07
n4nd0blackburn: good you said it :P00:07
n4nd0blackburn: do you guys have to go with suits to school?00:07
blackburnbut it is not a photoshop or anything00:07
blackburnn4nd0: yes, we used to wear suits in school00:07
-!- Peter_Gr [~Peter_Gr@] has quit [Quit: Leaving]00:08
blackburnhmm I've got uncatchable bug00:15
blackburnn4nd0: sonney2k: good night guys00:19
@sonney2kblackburn, do you know this
@sonney2kseems like someone wrote c45 in c++00:19
@sonney2k...and is applying for gsoc00:19
@sonney2kno idea how good his impl. is though...00:19
@sonney2kbut he could be the guy doing id3 / c45 / c50 / random forests00:20
blackburnor girl according to email00:21
blackburnok see you00:21
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]00:21
n4nd0good night00:21
n4nd0sonney2k: StreamingFeatures cannot be used directly by classifiers right?00:37
n4nd0in order to read from a file, I have to use StreamingFeatures to read the file and fill a e.g. CSimpleFeatures00:37
shogun-buildbotbuild #400 of java_modular is complete: Success [build successful]  Build details are at
-!- jckrz [] has quit [Ping timeout: 246 seconds]02:02
-!- vikram360 [~vikram360@] has joined #shogun02:22
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]02:30
-!- vikram360 [~vikram360@] has joined #shogun02:30
-!- makokal_ [] has joined #shogun02:36
-!- makokal [] has quit [Ping timeout: 245 seconds]02:38
-!- makokal_ is now known as makokal02:38
-!- flxb_ [] has quit [Quit: flxb_]03:06
-!- flxb [] has joined #shogun03:11
-!- flxb [] has quit [Quit: flxb]03:16
-!- zxtx [] has joined #shogun03:43
-!- pluskid [~chatzilla@] has joined #shogun04:00
pluskidFinally figured out why I still cannot submit my application: I mistakenly registered as a mentor...04:01
pluskidWriting to GSoC to fix this04:01
n4nd0pluskid: curious fact :)04:09
pluskidn4nd0: I think maybe I clicked the wrong button when registering :(04:14
pluskidThough I can simply register another account with another email :D04:14
n4nd0fast solucion :)04:15
pluskidn4nd0: you are a GSoC, too, aren't you?04:16
n4nd0well, I try to be one04:16
pluskidCool, what idea are you working on?04:17
n4nd0what about you?04:18
pluskidwhat do you mean by SO?04:18
pluskidI'm trying to work on structured output learning, is it the same topic to you?04:19
n4nd0oh sorry, Structured Output04:19
n4nd0yeah, it seems so :S04:19
pluskidthere are two topics related to SO04:19
pluskidBuilt generic structured output learning framework04:19
pluskidand Bundle method solver for structured output learning04:20
pluskidwhich one are you working on?04:20
n4nd0I am with the first one04:20
pluskidYou have been very active here, I should catch up... >_<04:21
n4nd0well it looks like you have quite a bit of experience in ml (info in your presentation mail) it's me who should catch up actually ;)04:24
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]04:25
zxtxwasn't SO on the list last year?04:26
n4nd0no idea04:27
pluskidno idea, too. Haven't participated last year.04:28
pluskidn4nd0: Yes, both of us have our own strength. But I think I should start to consider applying one more project/organization for the sake of conflict.04:30
n4nd0pluskid: probably, I think I will apply for a second project in shogun as well04:30
pluskidme too, because shogun is really an ideal project to work with.04:31
pluskidmaybe we should discuss before we choose our second idea04:31
pluskidto avoid a 2nd conflict :D04:31
n4nd0I already thought of generic multiclass04:32
n4nd0what about you?04:32
pluskidhaven't decided yet04:32
pluskidmaybe I will consider other ideas, since you chose MTL04:33
pluskidI'm still reading the details about other ideas04:33
pluskidI will tell you when I have a decision04:33
n4nd0pluskid: cool, thanks :)04:34
-!- makokal [] has quit [Quit: makokal]04:38
-!- shelhamer [] has joined #shogun04:38
-!- shelhamer [] has quit [Client Quit]04:43
pluskidn4nd0: Maybe I'll choose MTL.... but another MTL: Implement multitask and domain adaptation algorithms04:43
pluskidor Kernel Two-sample/Dependence test04:44
n4nd0I think someone else is with the multitask04:44
n4nd0but it's rather difficult to avoid competence04:45
pluskidyes :(04:45
zxtxdoes anyone ever need multitask learning in practice?04:45
pluskidit seems useful, because many tasks alone have too few information to work out a good model.04:48
pluskidBut I don't know about the real-world application side04:48
zxtxwell it seems you just need a good prior across tasks05:02
zxtxanyway clearly not a project I should be a mentor or student on :P05:03
pluskidzxtx: yes, but I guess in MTL the piror can be learned automatically from the data, instead of made by the expert :p05:06
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]05:19
-!- n4nd0 [] has quit [Ping timeout: 260 seconds]05:29
-!- zxtx [] has left #shogun ["Leaving"]05:37
-!- harshit_ [~harshit@] has joined #shogun06:31
-!- pluskid [~chatzilla@] has joined #shogun07:12
-!- n4nd0 [] has joined #shogun07:28
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]07:33
harshit_hey sonney2k07:38
harshit_hey n4nd007:38
n4nd0hey harshit_ how is it going?07:39
harshit_good . whats up on your side ?07:39
n4nd0just fine :)07:39
harshit_so i read the irclog07:39
harshit_and i think my deep learning thing could not work in this gsoc07:40
n4nd0did you take a look to those references sonney2k talked about?07:40
harshit_but still i can implement it after gsoc ( or if i dont get into gsoc)07:40
n4nd0ok, then it's good07:41
harshit_they are basically big toolboxes07:41
harshit_based on Neural networks07:41
harshit_so extracting deep learning algos from them is a real pain07:41
n4nd0I see07:41
harshit_and you were also talking about implementing decision trees07:42
harshit_C5.0 code seems to be easy, but i wonder why blackburn kept on saying07:42
n4nd0it looks like shogun needs them very bad07:43
harshit_that implementing that is really difficult07:43
harshit_oh i se07:43
harshit_but if C5.0 code is in C then what is so difficult in implementing it in shogun ?07:44
harshit_do you have any idea07:44
n4nd0well probably it is not so difficult as you have said07:45
n4nd0but it is not going to be probably just copy + paste07:45
n4nd0one probably has to adapt it to shogun's structure (e.g. using SGVectors, SGMatrices and the like)07:45
n4nd0I have not really looked into that project though07:46
harshit_i'll have a talk with blackburn though,07:46
harshit_Coz after implementing newtonsvm my project seems to be small07:46
n4nd0I think it's a good idea07:47
harshit_i need something else to add to that project07:47
harshit_do you know of anything that could be added to it !!07:47
n4nd0there must be other tree implementations that could be useful to have07:47
n4nd0didn't they mention something about it? I think I have heard it07:48
harshit_other tree implementations ?? as in07:49
harshit_you talking about c4.5 or other implementation of c5.0 itself07:49
harshit_i think blackburn will give me a closer idea..07:50
harshit_and some trivial questions:07:50
harshit_what are these shogun-builtbot and shogun-toolbox in chat area07:51
harshit_and how is sonney2k always available07:51
n4nd0shogun-buildbot just gives some log information about the state of the build of the project07:52
n4nd0you do a pull request for example, sonney2k merges it07:52
n4nd0and a automatic process of building the project starts07:52
harshit_okay got it07:52
n4nd0this guy shogun-buildbot gives us feedback on IRC about how did that process go07:53
n4nd0checked some of the links it gives if you want07:53
n4nd0I have no idea what shogun-toolbox does07:53
harshit_and second question how is sonney2k always available07:53
harshit_yeah i checked once..07:53
n4nd0and well sonney2k may have an IRC client connected all the time07:54
n4nd0but that doesn't mean he is online all the time07:54
n4nd0you should ask him about it if you are curious :)07:54
harshit_i would i want to do it with my account too :)07:55
-!- Marty28 [~chatzilla@] has joined #shogun08:25
-!- vikram360 [~vikram360@] has joined #shogun08:32
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun08:35
-!- hpsoar__ [7c108901@gateway/web/freenode/ip.] has joined #shogun08:37
n4nd0Marty28: hey!08:38
Marty28Could anyone please tell me if the toolbox is ready for Python 3.1?08:38
blackburnMarty28: almost, gsomix works on it actualyl08:38
Marty28Ok, so I will better go on with 2.7 for the next months.08:39
Marty28Is there a random DNA/RNA sequence generator with motifs included in shogun somewhere?08:42
blackburnI haven't seen any :)08:44
Marty28I am currently programming one, but I want it to be realistic.08:44
-!- jckrz [] has joined #shogun08:45
harshit_hey blackburn08:47
blackburnharshit_: hi08:47
harshit_sonney told me  not to use get_computed_dot_feature_matrix() function08:48
harshit_and use add_to_dense() etc08:48
blackburnyes makes more sense08:48
harshit_but the problem is that i want that matrix for cblas_dgemm08:48
harshit_so is there any way out ?08:49
blackburnwhat is the matrix you compute?08:49
harshit_i dont compute any matrix08:49
harshit_its just the feature matrix entered by user08:49
blackburnwhat is the matrix you compute with dgemm?08:49
harshit_actually it has been used at alot of places08:50
harshit_in one place i have to get a subset of feature matrix08:50
blackburngram or covariance?08:50
harshit_with given rows08:50
harshit_gram , covariance ?? i dont get it08:51
blackburne.g. X'X is a gram matrix08:51
blackburnand XX' is a covariance matrix08:51
blackburnwhere X is a feature matrix08:51
blackburnyou may employ linear kernel for that08:52
blackburnor compute it manually08:53
blackburnwhile (i,j) slot contains just a dot product between ith and jth vectors08:53
harshit_okay.. got it08:54
harshit_but that would require complete redesign of whole program08:54
harshit_I dont get that why cant we make use of whole feature matrix08:55
harshit_does that slows down the program ?08:55
blackburnno, but whole feature matrix could be huge08:56
-!- hpsoar__ [7c108901@gateway/web/freenode/ip.] has quit [Quit: Page closed]08:56
blackburnand wouldn't fit memory08:56
harshit_okay.. then i think its worth working on that.08:57
-!- jckrz [] has quit []08:58
blackburnmain point is that you should avoid this access operations and use add and dot ops when possible09:00
-!- hpsoar [7c108901@gateway/web/freenode/ip.] has joined #shogun09:01
harshit_yeah got it , Bcoz thats the reason why COFFIN framework was built .09:03
harshit_hey blackburn i am using CDotFeatures for features but in this class add_to_dense and dense_dot are virtual, so do i need to use some other type of features in my program09:17
blackburnno, it is ok09:17
blackburnit is an abstract class for other features09:17
blackburnso e.g. SimpleFeatures has implementation of these methods09:18
harshit_but then i could not make use of add_to_dense once i type cast to cdotfeatures09:18
harshit_bcoz it wont knw which implementation to use09:19
harshit_i am not sure actually09:19
blackburnyou don't have to know :)09:19
harshit_okay ... i'll try it and get back to you in a moment.09:20
n4nd0blackburn: do you have any example reading from .mat like we do with loadmat in python examples but in C++?09:31
n4nd0blackburn: I am doing one using Streaming objects but I think I am not getting it well09:31
blackburnn4nd0: oh.. no, I never tried to do it with C++09:33
blackburnwhy do you want to work with .mat from C++?09:33
n4nd0I just want to use data files from C++09:34
n4nd0they don't have any specific format, just one line with two floats09:34
n4nd0should I use shogun stuff to read them or standard C++?09:34
n4nd0nice plot09:35
harshit_hey, what is dense vector in shogun ?? i mean why is dense word used in every function in Cdotfeatures09:36
n4nd0harshit_: is just a normal matrix, non sparse09:36
harshit_matrix ??09:37
blackburnvector matrix all the same :)09:37
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]09:51
-!- Marty28 is now known as Marty09:57
-!- Marty is now known as Marty2809:57
-!- flxb [] has joined #shogun10:24
-!- n4nd0 [] has quit [Quit: leaving]10:44
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:46
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:51
-!- gsomix [~gsomix@] has quit [Read error: Operation timed out]10:53
-!- flxb_ [] has joined #shogun10:57
-!- flxb [] has quit [Ping timeout: 260 seconds]10:59
-!- flxb_ is now known as flxb10:59
Marty28ah, easysvm has a generator for DNA with motifs11:05
-!- harshit_ [~harshit@] has quit [Ping timeout: 252 seconds]11:18
-!- hpsoar [7c108901@gateway/web/freenode/ip.] has quit [Quit: Page closed]11:59
-!- flxb [] has quit [Quit: flxb]12:10
-!- flxb [] has joined #shogun12:11
-!- flxb [] has quit [Client Quit]12:11
-!- OgreMagi [~vinaygarg@] has joined #shogun12:13
-!- OgreMagi [~vinaygarg@] has left #shogun []12:13
-!- gsomix [~gsomix@] has joined #shogun12:23
-!- harshit_ [~harshit@] has joined #shogun12:26
-!- pluskid [~chatzilla@] has joined #shogun12:27
-!- Peter_Gr [~Peter_Gr@] has joined #shogun12:30
harshit_hey, does anyone here know the working of add_to_dense_vec() ??12:41
harshit_problem is that it gives me the results, considering features matrix as row major but actually feature matrix is in column major12:42
-!- in3xes [~in3xes@] has joined #shogun12:46
-!- gsomix [~gsomix@] has quit [Ping timeout: 244 seconds]13:13
-!- PSmitAalto [82e9b263@gateway/web/freenode/ip.] has joined #shogun13:25
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun13:29
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]13:29
-!- gsomix [~gsomix@] has joined #shogun13:29
-!- in3xes [~in3xes@] has quit [Ping timeout: 246 seconds]13:32
-!- in3xes [~in3xes@] has joined #shogun13:45
PSmitAaltoHey guys14:07
PSmitAaltoA question about using MKLMultiClass, will this use CPLEX if I have compiled with CPLEX support?14:07
-!- in3xes_ [~in3xes@] has joined #shogun14:14
-!- in3xes [~in3xes@] has quit [Ping timeout: 260 seconds]14:18
-!- Peter_Gr [~Peter_Gr@] has quit [Ping timeout: 265 seconds]14:21
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 246 seconds]14:33
-!- in3xes_ [~in3xes@] has joined #shogun14:46
-!- in3xes__ [~in3xes@] has joined #shogun14:51
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 244 seconds]14:54
-!- flxb [] has joined #shogun14:55
-!- hpsoar_ [7c108901@gateway/web/freenode/ip.] has joined #shogun15:04
-!- vikram360 [~vikram360@] has quit [Ping timeout: 264 seconds]15:07
hpsoar_I'm wondering why there's no comments or suggestions to my proposal for gsoc15:07
hpsoar_I suppose that someone may review it and give some judgement and suggestions15:08
-!- vikram360 [~vikram360@] has joined #shogun15:19
-!- in3xes__ is now known as in3xes15:29
-!- in3xes [~in3xes@] has quit [Quit: Leaving]15:29
-!- hpsoar_ [7c108901@gateway/web/freenode/ip.] has quit [Quit: Page closed]15:34
-!- Peter_Gr [~Peter_Gr@] has joined #shogun15:45
-!- flxb [] has quit [Read error: Connection reset by peer]15:55
-!- n4nd0 [] has joined #shogun15:59
-!- vikram360 [~vikram360@] has quit [Ping timeout: 244 seconds]16:06
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun16:24
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]16:24
-!- flxb [] has joined #shogun16:24
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun16:40
harshit_hey n4nd017:10
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]17:10
harshit_i hav an issue with add_to_dense_vec() function17:10
harshit_hav you used it anywhere ?17:10
-!- vikram360 [~vikram360@] has joined #shogun17:12
n4nd0harshit_: I think I have not used it, but tell me anyway17:14
harshit_firstly i need  to confirm that whether matrices are stored in shogun in column major order or not17:17
harshit_i think they are stored in column major .. yes17:17
n4nd0yes they are17:18
harshit_problem is that add_to_dense_vec gives me the results, considering features matrix is stored as row major17:18
n4nd0so you are using CDotFeatures?17:19
harshit_do you knw any place where i can upload small scripts ?17:19
harshit_yes sort of coz simplefeatures is derived from CDotFeatures only17:19
n4nd0if you mean code I use this
harshit_nice site, have a look at a small portion of my code :
harshit_here when i use CMath::display_matrix(sgv.vector,x_d,1); it shows me 1 2 317:23
harshit_rather than 1 4 717:23
harshit_same is the case with add_to_dense_vec17:23
n4nd0let me think of it a moment17:24
n4nd0so sgv is the first vector of features17:26
n4nd0i.e. the first column of matrix17:26
n4nd0so then I think it's normal that it's 1 2 317:26
harshit_it should be first row of matrix !17:27
n4nd0why the first row?17:28
n4nd0as we said they are stored column wise in shogun17:28
harshit_but still if then if it first vector (or first coloumn ) then it should contain x_n elements17:29
n4nd0and here x_n is?17:29
n4nd0what are the values of x_n and x_d here?17:29
harshit_x_n: no of features .. ie no rows17:30
harshit_and x_d is no of dimensions17:30
n4nd0yeah ... but I mean the particular values you are using17:30
harshit_see matrix is [1 4 7;2 5 8 ;3 6 9]17:31
harshit_ohh its 5 for x_n and 3 for x_d17:31
n4nd0in any case I think you got it the other way round17:31
n4nd0look at the constructor in CSimpleFeatures
n4nd0the one in line 3217:32
n4nd0num_feat (your x_d here) is for the rows17:32
n4nd0and num_vec (your x_n) for the number of cols17:32
n4nd0got it?17:34
harshit_but i have used set_feature_matrix17:34
harshit_and in set_feature_matrix() first argument is num_feat and second is num_vec17:34
harshit_and i have used x_d for first17:34
harshit_and x_n for second17:35
n4nd0yeah but why did you make that?17:36
harshit_just for understanding this function17:36
n4nd0I mean17:36
-!- jckrz [] has joined #shogun17:36
n4nd0SGMatrix<float64_t> matrix(x_n,x_d);17:36
n4nd0first x_n and later x_d17:37
n4nd0the other way round17:37
harshit_coz SGMatrix constructor says to have first arg as no of rows and second as no of columns17:37
harshit_if i do other way around i get an assert error17:38
harshit_hey u figured out something17:38
n4nd0I don't know if you see what happens17:38
n4nd0it is like if you have this matrix17:38
harshit_may be i should not use SGMatrix17:39
n4nd0[1 3 5; 2 4 6]17:39
n4nd0and you build it doing SGMatrix< float64_t >(2, 3)17:39
harshit_wait i'll give it a try17:39
n4nd0and later you do features(matrix, 3, 2)17:39
n4nd0you will tell to features that the matrix is17:40
harshit_in first case ie with the matrix you wrote17:41
harshit_sgv vector displays 1 2 317:41
harshit_whereas it should display 1 3 517:42
n4nd0each feature vector is a column17:43
harshit_okay agreed17:43
n4nd0I know that the notation might be confusing17:43
harshit_but then column should have 2 members17:44
harshit_take a practical example17:44
n4nd0I don't know what you mean with but then column should have 2 members17:44
n4nd0ok I think I got it17:45
n4nd0yeah it should be like that17:45
n4nd0but the problem is when you do features->set_feature_matrix(matrix.matrix,x_d,x_n);17:45
n4nd0changing the order17:45
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]17:45
n4nd0you create it with x_n, x_d17:45
n4nd0and set it with x_d, x_n17:45
n4nd0when your matrix really means17:46
n4nd0[1 3 5; 2 4 6]17:46
n4nd0you make believe to features that it is17:46
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun17:46
harshit_but doing that results in an error17:46
n4nd0I guess the error must come from another thing that is wrong17:47
harshit_i think best way to debug this is to drop SGMatrix for now and feed simplefeatures normal matrix17:47
n4nd0mmm I don't think so17:47
n4nd0when you change to x_d, x_n17:48
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]17:48
n4nd0everything must work well until SGMatrix<float64_t> sgmatrix=features->get_computed_dot_feature_matrix();17:48
harshit_wait let me use gdb17:49
harshit_it is good till add_to_dense_vec () function17:51
harshit_and sgv displays : 1 217:51
harshit_and matrix is= [1 4; 2 5 ; 3 6]17:52
-!- PSmitAalto [82e9b263@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]17:53
harshit_dam this is really confusing17:54
harshit_i think i need to have a closer look at how other implementations make use of add_to_dense_vec() function17:56
n4nd0I think you've to think how it is stored in memory18:00
n4nd0is just an array18:00
n4nd0a vector is an array a matrix is also an array18:00
harshit_yeah right18:00
n4nd0for matrices the dimensions are the ones that define them, and are really important18:00
harshit_but see here dimensions are alright i mean whe i use get_dim_feature_space it returns 318:02
harshit_exactly the value i used while defining matrix18:02
harshit_and when i use add_to_dense_vec it also returns 3 values18:02
harshit_but it returns those 3 values considering the matrix as row major18:03
harshit_but actually it is stored in column major in features18:03
harshit_i think we need blackburn around here for this ..18:04
harshit_or perhaps sonney2k18:05
harshit_i think i am messing up with x_n and x_d ..18:05
harshit_function could not be wrong ..18:06
-!- Miggy [~piggy@] has joined #shogun18:07
harshit_Hey dont waste your time on this (if you are ), i'll ask blackburn or sonney2k for this18:09
n4nd0there is no problem18:09
n4nd0but I don't really understand what is it that you don't understand now :)18:10
n4nd0harshit_: what is it that doesn't make sense for you?18:11
harshit_If add_to_dense_vec is returning 3 elements i.e  no. of dimensions i.e one of feature vector18:11
harshit_then it should be 1 4 718:12
harshit_first datapoint18:12
n4nd0your matrix was [1 4 7;2 5 8 ;3 6 9]18:12
n4nd0your ; is a row delimiter right?18:13
n4nd0I mean18:13
n4nd01 4 718:13
n4nd02 5 818:13
n4nd0the first point is 1 2 318:13
n4nd0so then?18:14
harshit_oh sorry18:14
harshit_got it18:14
n4nd0the key is to remember that they are just arrays18:15
n4nd0just a T* (T is float64_t, int32_t or whatever)18:15
harshit_i was confusing coz i thought 1 4 7 as my first datapoint always18:16
n4nd0you're welcome18:19
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]18:31
harshit_bye n4nd0 you really provide great help18:31
-!- vikram360 [~vikram360@] has joined #shogun18:32
n4nd0harshit_: :)18:33
-!- n4nd0 [] has quit [Quit: leaving]18:33
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds]18:35
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun18:48
-!- flxb [] has quit [Read error: Connection reset by peer]18:57
-!- harshit_ [~harshit@] has quit [Quit: Leaving]19:11
-!- flxb [] has joined #shogun19:15
CIA-64shogun: Evgeniy Andreev master * rda550bd / (3 files in 3 dirs): fixed python typemaps -
CIA-64shogun: Soeren Sonnenburg master * r355cb51 / (3 files in 3 dirs):20:06
CIA-64shogun: Merge pull request #399 from gsomix/python3_interface20:06
CIA-64shogun: updates on python3 interface -
CIA-64shogun: Soeren Sonnenburg master * rc69fb70 / (3 files):20:12
CIA-64shogun: Merge pull request #398 from karlnapf/master20:12
CIA-64shogun: First bunch of fixes for migration -
gsomixsonney2k, moin.20:20
@sonney2kgsomix, hey20:23
@sonney2kdobrui wetscher :)20:24
gsomixsonney2k, dobrui. :)20:24
-!- blackburn [~qdrgsm@] has joined #shogun20:25
@sonney2kgsomix, I wanted to comment on your patch:20:25
gsomixI'm listening.20:26
@sonney2kpython3's integers are long?20:26
@sonney2kand python3's strings are always unicode?20:27
@sonney2kand python3 supports some kind of bytestrings that are mutable?20:27
@sonney2kif so we should for strings use these mutable byte based strings20:27
@sonney2k(at least in the future support it)20:28
@sonney2kblackburn, hi hi20:29
blackburnsonney2k: heyaa20:29
gsomixsonney2k, integers are long. And strings are unicode, but there is old-style strings - PyBytes.20:29
@sonney2kgsomix, are these PyBytes deprecated or the new mutable string type?20:30
blackburnoh that is so funny they call everything with py- suffix20:30
@sonney2kblackburn, 8 proposals by now...20:31
@sonney2kblackburn, I am happy they do - C has no namespaces...20:31
@sonney2kand actually namespaces are crank ...20:31
@sonney2kwhen a project doesn't use them one is screwed20:31
blackburnsonney2k: yeah that's ok but every library has this py-*20:32
blackburnI said suffix? :D20:32
blackburnprefix probably lol20:32
gsomixsonney2k, new. PyBytes has just replaced PyString.20:33
blackburnsonney2k: I heard you like proposals so n4ndo put a proposal into your melange20:33
@sonney2kgsomix, so is PyBytes immutable or mutable?20:33
@sonney2kblackburn, I know20:34
blackburnI am in mood to joke - again damn20:34
blackburnsonney2k: so what about C4.5 girl?20:34
gsomixsonney2k, immutable.20:35
@sonney2kI don't know - the google code project is very new (didn't exist a few days ago)20:35
@sonney2kand we have to see how it goes to integrate that stuff in shogun / compare with j45 from weka20:36
blackburnsonney2k: actually we really need to know # of slots..20:36
gsomixsonney2k, hmm. wait.20:38
gsomixsonney2k, yes, bytes is immutable. I'm confused by old documents.20:41
-!- lamer13329602986 [] has joined #shogun20:45
* gsomix had returned from dance floor. Now he is bit stupid.20:46
-!- jckrz [] has quit [Ping timeout: 265 seconds]20:48
* blackburn is getting angry with girl-like laugh of his neighbor :D20:50
@sonney2kgsomix, so b'string' is a byte string20:50
gsomixsonney2k, yep20:50
@sonney2kand 'string' is a normal string right?20:50
blackburnno more u'stuff'?20:51
blackburnugliest part of python for sure20:51
@sonney2kgsomix, b'string' works in python 2.x too right?20:51
@sonney2kblackburn, yeah luckily no longer needed in python320:51
gsomixsonney2k, it works. In my python2.7. :)20:52
gsomixBut type(b'string') returns 'str' in python2.7.20:53
@sonney2kgsomix, ok I guess we have to support both the b'string' variant and the unicode variant (checking that only 0...127 is used) in shogun20:53
@sonney2knot now but we should keep that in mind20:54
gsomixsonney2k, ok.20:55
gsomixI remember it.20:55
gsomixsonney2k, can i send changes in examples?20:56
gsomixprint ' ' -> print(''), and others.20:57
@sonney2kthese are all compatible so it is a good idea20:58
-!- flxb_ [] has joined #shogun20:59
gsomixsonney2k, and what about examples, which are not working at now?21:00
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun21:00
* gsomix is enslaved by google translate. The thing was invented by the devil. =___=21:01
-!- flxb [] has quit [Ping timeout: 246 seconds]21:01
-!- flxb_ is now known as flxb21:01
@sonney2khi wiking - are you still with us? torturing  alexander with questions?21:01
wikingsonney2k: yeye21:01
wikingi'm here21:02
blackburnsonney2k: hmm actually I forgot to CC you probably21:02
wikingjust got back21:02
wikingblackburn: yeah actually we could do an abstract class21:02
wikingand that would be the 'minimizer'21:02
wikingso that one has to implement (inherit) a class from that one21:02
wikingdid you mean that in your email?21:02
blackburnbetter this way21:03
wikingok that's good as well21:03
wikingmore c++ like21:03
wikingi can change now the api in 10 mins21:03
@sonney2kgsomix, difficult to say21:03
wikingany other comments?21:03
wikingblackburn: the LatentLabel would actually contain two parts21:04
blackburnwiking: well only ones I have listed21:04
wikingy (the usual labeling) and h and that can be 'anyting'21:04
wikingthat is the latent variable21:04
wikingand how you interpret that21:04
wikingit's really up to your application21:04
@sonney2kgsomix, quite a couple are not working but it needs probably quite some work to figure out why ...21:04
wikingsonney2k: yeah i did a first api attempt for the latent svm.... do you want to be in the flow of emails?21:05
gsomixsonney2k, ok, i will think about it.21:05
@sonney2kwiking, I would even prefer if you do this discussion in public on the mailinglist21:05
wikingsonney2k: actually why haven't we started there i have no idea :))))21:06
wikingi second this idea as well :P21:06
gsomixsonney2k, while I will solve the problems in the covertree.21:06
blackburnsonney2k: wiking: hmm could be considered unfair21:06
wikingthe only thing is that i have a fierce competition ... :)21:06
wikingsomebody else wants as well to do the latent svm21:06
wikingi've just seen a week ago on the mailing list21:07
blackburnit would look like we already chose wiking21:07
wikingbut then again i do all my implementation public21:07
@sonney2kblackburn, no21:07
@sonney2klast year a couple of people started doing some project21:07
wikingso that could be taken an improved by anybody...21:07
@sonney2ksome we took some we didn't take21:08
@sonney2kand we even merged most patches21:08
blackburnwiking: sonney2k: ok then it should be in the way 'how would I do it if I was accepted' :D21:08
@sonney2kif someone wants to do a project - why should he waste time submitting patches he is not really interested in21:09
@sonney2kso he could start directly with the project and send patches for that21:09
@sonney2ke.g. harshit is doing newtonsvm for some time now21:09
wikingsonney2k: i would send patches... just simply referring to my fork's branch....21:09
@sonney2kpull requests whatever21:10
wikinghihi yeah sure... anyhow then i'll send the next email about it on the mailing list21:10
blackburnwiking: yeah just make work in progress request21:11
blackburnall the discussion could be there21:11
@sonney2klast year we had people working on shogun i/o stuff, sgd-qn, java interface already before gsoc started21:14
blackburnI forgot21:14
@sonney2kblackburn, I have no idea what you did?21:15
blackburna lot of vodking21:15
@sonney2kblackburn, lets call you vodking from now on :D21:15
blackburnsonney2k: evaluation stuff21:16
blackburnI reworked it a little (completely)21:16
CIA-64shogun: Soeren Sonnenburg master * r6fea6c9 / src/interfaces/python_modular/swig_typemaps.i : remove debug code / string typemap -
@sonney2kvodking - I know and GNB, conjugateindex...21:18
blackburnsonney2k: do not remind meconjugateindex crap21:18
blackburnit was developed in december actually21:18
blackburnbut GNB during pre-gsoc yeah21:18
@sonney2kand actually you were mentoring students21:18
@sonney2kbtw, is GNB aready converted to logspace?21:19
@sonney2kwould be some simple task for a patch21:19
blackburnbut covertree for KNN is still vacant21:19
blackburnsonney2k: you said you were planning to try GNB on some task, did you?21:20
@sonney2kreally many examples work now :)21:20
@sonney2kerr tests21:20
blackburnsonney2k: which fails?21:24
@sonney2kquite a couple of errors we had before21:25
@sonney2ksome nan's in substracts due to preprocessors21:25
-!- Peter_Gr [~Peter_Gr@] has quit [Ping timeout: 248 seconds]21:26
-!- Peter_Gr [~Peter_Gr@] has joined #shogun21:28
CIA-64shogun: Soeren Sonnenburg master * rcfb5860 / testsuite/python_modular/ : add -f option to tester to only show failed tests -
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]21:33
@sonney2kgsomix, btw could you please use assert_version to check for the python version21:43
@sonney2kgsomix, in configure - an example is assert_version doxygen $_doxygen $_doxygen_version 1.6.021:44
-!- vikram360 [~vikram360@] has joined #shogun21:44
@sonney2k_doxygen contains yes/no/auto (depending on value it will fail)21:44
@sonney2k_doxygen_version is the version string and 1.6.0 is what is required21:44
@sonney2kblackburn, what's up with converters - will they ever get a test?21:53
blackburnsonney2k: yes21:53
blackburnbut tests are rather complex for that shit21:54
shogun-buildbotbuild #594 of cmdline_static is complete: Failure [failed test_1]  Build details are at  blamelist: heiko.strathmann@gmail.com21:54
blackburnsonney2k: arpack+superlu / arpack / lapack21:55
blackburnthree eigensolvers21:55
blackburnwhile weight matrices are computed with pretty high precision and could be checked21:56
blackburneigenvectors are pretty random21:56
@sonney2ksome kind of sanity check - like objective value?21:56
blackburngreat idea btw21:56
blackburnbut it could vary as well21:57
blackburnI'll check later21:57
@sonney2k(I wouldn't accept any mloss paper that doesn't have tests btw)21:57
-!- vikram360 [~vikram360@] has quit [Ping timeout: 264 seconds]21:57
@sonney2kbut I am no longer AE so you might be lucky21:57
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]21:58
@sonney2ksome of the tests in shogun are maybe overly strict21:58
shogun-buildbotbuild #572 of r_static is complete: Failure [failed test_1]  Build details are at  blamelist: heiko.strathmann@gmail.com21:58
blackburnsonney2k: well do you think it is ok to resubmit in say may?21:58
@sonney2kcurrently I expect the objects to be exactly the same21:58
blackburnis there any time limit for that?21:58
@sonney2kI am not aware of a limit21:59
blackburnI can't promise I implement all the tests even in april21:59
@sonney2kthat makes sense if you test on the same machine with no difference in anything (numerical libraries, compiler, ...)21:59
@sonney2kbut for e.g. svm just using atlas will make the tests fail21:59
@sonney2kand the solution will only be within 1e-3 or whatever svm eps is22:00
blackburnsonney2k: may be test should contain required precision22:00
@sonney2kproblem is I often introduced problems that were like 1e-10 off somewhere but results should have been *exactly* the same22:01
@sonney2kblackburn, yeah but this can hide problems too22:01
@sonney2kmaybe we should generate these test data on the buildbot22:01
blackburnsonney2k: one reason why I didn't implement such test is it is really easy to check graphically22:01
@sonney2kand then be more tolerant on other machines22:02
shogun-buildbotbuild #573 of octave_static is complete: Failure [failed test_1]  Build details are at  blamelist: heiko.strathmann@gmail.com22:02
@sonney2kblackburn, I don't agree. a PCA solution might *look* ok but variance of that other dimension might be 1e-6 larger...22:02
@sonney2kor swiss-unroll might look almost perfect but still isn ot22:03
@sonney2kis not22:03
blackburnsonney2k: I meant LLE and such stuff22:03
blackburnactually there is no degree of freedom in lle22:03
blackburnvariance one*22:03
@sonney2kblackburn, what do you think wrt my test idea above?22:04
blackburnsonney2k: no idea if it would work well22:04
@sonney2kblackburn, it will work well until we change build bot / os / gcc / libs22:05
@sonney2kand on other archs we can at least test...22:05
blackburnyou are right probably22:05
@sonney2kso we need some precision for returned outputs in these functions22:05
shogun-buildbotbuild #561 of python_static is complete: Failure [failed test_1]  Build details are at  blamelist: heiko.strathmann@gmail.com22:06
@sonney2keither pairs ((output1, prec2), (output2, prec2))22:06
@sonney2kor (out1,out2), (prec1,prec2)22:06
@sonney2kor some global precisions variable22:06
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:07
blackburnsonney2k: actually I can employ some student for writing tests for edrt :D22:07
blackburnobjective function would be provided so it is pretty easy to do22:07
blackburnin python22:07
@sonney2ksounds like you are a lot faster doing it yourself22:08
blackburnI don't know whether it should be implemented in edrt itself22:08
blackburnwhat do you think?22:08
blackburnsonney2k: ok I can't promise still but I'll try to add testing there22:10
blackburnas soon as I am finished with libedrt22:10
blackburnsonney2k: good thing is LLE/LTSA/KLTSA/KLLE/NPE/LLTSA are all the same22:10
blackburnso 6 methods are almost ready22:11
@sonney2k...waste of time22:11
blackburnyou will see it is not22:11
-!- n4nd0 [] has joined #shogun22:12
blackburnsonney2k: there was no way to embed 100K vectors before22:12
blackburnit would take a hour but it would fit to memory this time22:12
blackburnit was all matrix based22:12
blackburnnow it is more sparse and compact22:13
blackburnyes I could do that in the way it was done before but it would take *more* time because of not flexible design22:13
blackburnI am 500% sure with it22:13
-!- romi__ [] has joined #shogun22:14
@sonney2kblackburn, but you don't need that weird libedrt / compute callbacks to do that - that really is a different issue22:15
blackburnsonney2k: yes exactly what I need to make it more flexible22:15
blackburnsonney2k: the only alternative is a class structure with some virtual compute kernel methods22:16
blackburnsonney2k: it was kernel matrix based22:16
@sonney2kbut why was it kernel matrix based before? kernel(i,j) did always exist22:17
blackburnsonney2k: because of wrong design!22:17
blackburnand this way fixes it22:18
@sonney2kI thought you did this because it was easier to call whatever inversion/eigenvalue lapack functions22:18
blackburnsonney2k: well things changed a great deal there :D22:18
blackburnsonney2k: now there is less code and more flexibility22:19
-!- Marty28 [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120310010446]]22:19
blackburnfor new methods / complex cachine one wouldn't have to create new kernel22:19
@sonney2kkernel caching should be in CKernel anyways22:20
blackburnok someday I'll convince you22:21
@sonney2knot this century :D22:22
blackburnok then next one22:22
blackburnI have to go22:22
blackburnsee you tomorrow22:22
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]22:23
@sonney2kn4nd0, very nice application...22:24
shogun-buildbotbuild #192 of nightly_none is complete: Success [build successful]  Build details are at
@sonney2kn4nd0, and thanks for helping harshit22:24
@sonney2kI hope he manages to use DotFeatures...22:24
gsomixsonney2k, ok.22:25
CIA-64shogun: Soeren Sonnenburg master * r0ff05d8 / testsuite/python_modular/ : show case / exception in tester -
CIA-64shogun: Soeren Sonnenburg master * r9d9bf6e / src/shogun/features/BinnedDotFeatures.cpp : initializ m_features with NULL / fixes hang -
@sonney2kn4nd0, ahh could you please file a bug - such that we remember to implement SGNDArray serialization support?22:28
n4nd0sonney2k: hey!22:31
n4nd0sonney2k: so where can I file that bug?22:31
@sonney2kn4nd0, github -> create new issue22:32
n4nd0sonney2k: ok, I am going to do that now22:32
n4nd0sonney2k: apart from that, I have been using this StreamingFile and StreamingSimpleFeatures classes22:33
n4nd0sonney2k: I am detecting some memory leaks with valgrind22:33
n4nd0sonney2k: do you know if they are supposed to be completely ok?22:33
n4nd0sonney2k: or could it be the case that they leak?22:33
@sonney2kthey should be yes22:33
@sonney2kif not -> bug22:33
n4nd0sonney2k: ok, I will take it a closer look before doing it22:34
n4nd0sonney2k: about the application, something I should complete or change then?22:34
-!- romi_ [] has quit [Quit: Leaving]22:45
shogun-buildbotbuild #595 of cmdline_static is complete: Success [build successful]  Build details are at
-!- gsomix [~gsomix@] has quit [Read error: Operation timed out]22:46
@sonney2kyay! of the 313 test configurations 'just' 49 fail22:49
shogun-buildbotbuild #573 of r_static is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #574 of octave_static is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #562 of python_static is complete: Success [build successful]  Build details are at
-!- gsomix [~gsomix@] has joined #shogun23:02
-!- lamer13329602986 [] has quit [Quit: jIRCii -]23:10
gsomixsonney2k, can you help me?23:13
@sonney2kgsomix, ?23:13
gsomixI always have problems with git repository. =___=23:15
@sonney2kwhat do you  want to do?23:15
@sonney2ke.g. to drop your changes you could do git checkout -f src/shogun/base/Parameter.cpp23:16
@sonney2kand then the pull23:16
gsomixsonney2k, I just want to get the latest changes.23:17
@sonney2kgsomix, yeah but you changed sth in files someone else changed so now you have a conflict23:17
@sonney2kquestion is what you want now23:17
@sonney2kget their version or keep yours23:18
@sonney2kor manually resolve changes23:18
@sonney2kfor example you can specify git pull -s STRATEGY23:19
@sonney2kwhere STRATEGY is their or our23:19
@sonney2ktheirs / ours23:19
@sonney2kso the remote / your version will be used23:20
@sonney2kor you edit the files to resolve the conflict23:21
@sonney2kand then do git add FILE_WITH_CONFLICT23:21
@sonney2kand git commit when done23:21
gsomixsonney2k, but, I did not change these files.23:23
shogun-buildbotbuild #424 of octave_modular is complete: Failure [failed compile]  Build details are at  blamelist: heiko.strathmann@gmail.com23:24
@sonney2kI would say you did23:25
@sonney2kso do the git rebase master -s theirs23:25
@sonney2kif you are sure23:25
gsomixsonney2k, thanks.23:46
shogun-buildbotbuild #399 of python_modular is complete: Success [build successful]  Build details are at
-!- OgreMagi [~vinaygarg@] has joined #shogun23:57
OgreMagi@sonney2k: I want to apply for the project " Built Generic Multiclass Learning Framework ". How should I proceed and are there any potential candidates for this particular project ?23:59
--- Log closed Thu Mar 29 00:00:19 2012