Open in new window / Try shogun cloud
--- Log opened Wed Mar 21 00:00:19 2012
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:23
-!- in3xes_ [~in3xes@] has joined #shogun00:42
-!- in3xes [~in3xes@] has quit [Ping timeout: 246 seconds]00:46
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Quit: This computer (or maybe me) has gone to sleep]02:32
-!- GenX [~Sambhav@] has quit [Ping timeout: 246 seconds]02:42
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 252 seconds]03:28
-!- GenX [~Sambhav@] has joined #shogun03:33
-!- GenX__ [~Sambhav@] has joined #shogun03:41
-!- GenX [~Sambhav@] has quit [Ping timeout: 250 seconds]03:44
-!- Sambhav__ [~Sambhav@] has joined #shogun03:55
-!- GenX__ [~Sambhav@] has quit [Ping timeout: 252 seconds]03:58
-!- in3xes [~in3xes@] has joined #shogun04:02
-!- Sambhav__ [~Sambhav@] has quit [Ping timeout: 246 seconds]04:13
-!- in3xes [~in3xes@] has quit [Ping timeout: 265 seconds]04:14
-!- in3xes [~in3xes@] has joined #shogun04:42
-!- in3xes [~in3xes@] has quit [Ping timeout: 260 seconds]04:48
-!- in3xes [~in3xes@] has joined #shogun05:45
-!- in3xes [~in3xes@] has quit [Ping timeout: 244 seconds]05:50
CIA-64shogun: Sergey Lisitsyn master * rdf1b927 / (2 files): Updated set_weights method -
CIA-64shogun: Sergey Lisitsyn master * r6b510fc / src/interfaces/modular/Library.i : Restored plif array things -
CIA-64shogun: Sergey Lisitsyn master * rebe175c / (6 files in 3 dirs): Fixes for applications -
CIA-64shogun: Sergey Lisitsyn master * r0e3adb0 / .gitignore : Updated git ignore -
-!- GenX [~Sambhav@] has joined #shogun07:45
-!- GenX [~Sambhav@] has quit [Ping timeout: 245 seconds]08:01
CIA-64shogun: Soeren Sonnenburg master * rba918e6 / (src/Makefile src/Makefile.template): add ctags target to upper makefile -
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun08:13
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]08:30
-!- n4nd0 [] has joined #shogun08:48
-!- shogun-buildbot [] has joined #shogun09:03
-!- GenX [~Sambhav@] has joined #shogun09:37
n4nd0sonne|work: hi! I am thinking of applying for the project "various usability improvements" apart from SO to raise the possibilities of success a bit09:42
n4nd0sonne|work: I have not previous coding experience with swig though, do you think that should prevent me from applying for that project then?09:42
-!- Miggy [~piggy@] has joined #shogun10:26
-!- Miggy [~piggy@] has quit [Ping timeout: 240 seconds]10:30
-!- l0nr4n [] has joined #shogun10:52
-!- l0nr4n [] has quit [Changing host]10:52
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun10:52
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:09
-!- puneetgoyal [~chatzilla@] has joined #shogun11:14
-!- harshit_ [~harshit@] has joined #shogun11:14
harshit_hello everyone11:17
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]11:17
harshit_@n4nd0 are you in the team of shogun ?11:20
n4nd0harshit_: I have made some contributions to the code, but I am just an student11:22
harshit_okay, Just had a few questions related to shogun's framework can you help me out11:23
n4nd0of course, as long as I have the answers :P11:24
harshit_if i want to do some operations on labels can i use CMath functions on it ?11:26
-!- GenX [~Sambhav@] has quit [Ping timeout: 250 seconds]11:27
n4nd0harshit_: so CLabels have a member that is a SGVector called labels11:28
harshit_operations as in scalar multiplication, or dot multiplication with any other vector11:28
n4nd0probably you can use that member to do the operations11:28
harshit_that class : SGVector contains all the functions for these opns within it ?11:29
n4nd0that class contains some functions but is not very exhaustive11:29
harshit_wait i'll hav a look at its signature11:30
n4nd0however, it has a public member11:30
n4nd0T * where T is likely to be int32_t or float64_t11:30
n4nd0so you can use CMath operations on that member11:30
harshit_okay thanks bro.11:31
harshit_and also i sent a mail on shogun's mailing list, But didnt get any reply11:31
harshit_if you knw some thing related to those questions11:31
harshit_please help me out, I am currently stuck in middle11:31
n4nd0I read your mail yesterday11:32
n4nd0but I couldn't answer most of the stuff11:32
n4nd0let me read again and refresh, maybe we can discuss about it anyway11:32
n4nd0so what does matlab minres exactly do?11:33
harshit_Thanks, In advance11:33
harshit_it calculates minimum resolution of a matrix11:33
harshit_ i also dont knw much abt it, Maybe11:33
harshit_some linear algebra guy will have more knowledge abt it11:34
n4nd0I don't know what is the minimum resolution of a matrix :P11:34
harshit_and second question: hav you come across any function in shogun for issparse()11:34
n4nd0so for the issparse thing11:35
n4nd0I know that in shogun there are structures prepared to hold dense data like CDotFeatures11:35
n4nd0and sparse, CSparseFeatures11:35
n4nd0so what is exactly the use that they make in the code of issparse?11:35
n4nd0is it just to test if the structure is itself sth sparse?11:36
n4nd0or do they actually check the content of the matrix to check whether it should be store as sparse?11:36
harshit_it checks the matrix if it is sparse or dense11:37
n4nd0but do you know how they make this check?11:37
harshit_exactly, Same was my reaction when i came across this function11:38
harshit_i can only say one thing MATLAB is gr8, it provides so complex thing in so much easy form11:38
n4nd0take a look here11:38
n4nd0is from the matlab doc11:39
n4nd0that makes me believe that they just check the type of container11:39
n4nd0let me see if I get to the actual code11:39
harshit_thats weird ?11:40
n4nd0why do you think so?11:41
harshit_how can they decide whether it is sparse or not on basis of class being used ?11:42
n4nd0I think matlab can store matrices in two (at least) different ways11:43
n4nd0dense representation and sparse11:44
n4nd0if you don't know what I mean, do like this in matlab11:44
n4nd0A = [1 2; 3 4]11:44
n4nd0you'll notice the different representation11:44
n4nd0so I guess that internally they have something to check if the matrix is stored in one or the other way11:45
n4nd0probably that issparse is just checking the value of that "something"11:45
harshit_what value ?11:46
harshit_i didnt get that last point11:46
harshit_i think they must have some algo11:46
n4nd0the value of the type of object11:46
harshit_to check whether to store the matrix in sparse or dense format11:46
n4nd0not whether TO store but whether IT is stored in sparse or dense11:47
n4nd0but again, I'm not sure about this11:47
n4nd0is just how I understand it from the doc of the file I pasted before11:48
harshit_but we dont make any explicit call to store our matrices in dense or sparse format while creating it !11:48
n4nd0what do you mean with our? in shogun?11:48
harshit_no in matlab . when we create any matrix11:49
n4nd0if you just define it like A = [....]11:49
n4nd0it will be dense11:49
n4nd0if you do sparse(A)11:49
n4nd0the return value is sparse11:49
n4nd0check it with11:49
n4nd0issparse(A) ( returns 0 )11:50
n4nd0issparse( sparse(A) ) ( returns 1 )11:50
harshit_yeah i saw that11:50
harshit_, you made a good point with it11:50
harshit_thanks i'll explore more about it11:50
n4nd0I'm not sure about it though, so check it always just in case11:50
harshit_yeah i know.. :)11:51
n4nd0in order to translate this to shogun I'd say that11:51
n4nd0if you deal with CFeatures in your function11:51
n4nd0you should check if it is CDotFeatures or CSparseFeatures11:51
n4nd0I am checking minres now11:52
n4nd0it is for the minimum residual, I have heard that before at least :)11:52
harshit_I think that wont be right .. but i'll hav a look on it11:52
harshit_oh yeah, i wrote it wrong last time11:53
n4nd0probably there is a method in blas for the minimum residual11:53
harshit_but also min res is of special type ..11:56
harshit_its Hessian11:57
harshit_i dont knw what it is11:57
harshit_but it is given in the comment with minres() in .m file11:57
n4nd0the second derivative of a matrix, like the Jacobian but one order more11:57
n4nd0but I guess that if the function in matlab computes the minimum residual11:58
n4nd0a function that does the same should work11:58
harshit_is that Hessian you are talkin about11:58
harshit_yeah i'll have a look at clbas function list11:59
harshit_i could not find minres in cblas .12:01
harshit_checked all level 1,2,312:01
harshit_btw why are cblas functions divided into level 1,2,312:02
harshit_do you hav any idea ?12:02
n4nd0opeartions between vectors, vectors and matrices and matrices; that's for the levels12:02
harshit_okay, so what i am looking for should be in level 312:03
harshit_but there is nothing like that12:03
harshit_are you there /12:07
n4nd0but I don't know how to compute that either so I am trying to find sth on the web12:07
harshit_okay , thanks for your time12:08
n4nd0for me it looks like this function you do the job12:11
n4nd0it's a LAPACK funtion so I'd say it can be used from shogun12:12
harshit_i dont think whole lapack is included in shogun12:13
harshit_is it?12:13
harshit_coz last time i searched shogun's CMath library i didn't find any thing like this12:14
n4nd0things in CMath are things that are implemented in shogun12:14
n4nd0but shogun uses external libraries and packages12:14
n4nd0so I think you can use that functions from shogun including the appropriate header12:15
harshit_and while linking do i need to include appropriate library file and library path too ?12:16
harshit_or every thing is in -lshogun itself12:16
n4nd0what do you normally do to compile?12:17
n4nd0I just run make and I think that with that it should be enough12:17
harshit_okay i'll have a look at shogun's Makefile, That will probably tell me what all external libraries are included12:18
harshit_damm shogun's makefile is really huge i havnt dealt with any thing like that before12:21
harshit_and the function that you just told me gmres.f12:27
harshit_it is written in fortran12:27
harshit_how can i use it in C++12:27
n4nd0in lapack.cpp you can see how other lapack functions are used from shogun12:29
harshit_okay ,thanks.12:29
harshit_got it, Thanks for your help12:31
n4nd0no problem12:31
n4nd0have to go now12:31
-!- n4nd0 [] has quit [Quit: leaving]12:31
-!- harshit_ [~harshit@] has quit [Quit: Leaving]12:32
-!- puneetgoyal_ [~chatzilla@] has joined #shogun12:33
-!- puneetgoyal [~chatzilla@] has quit [Ping timeout: 252 seconds]12:35
-!- puneetgoyal_ is now known as puneetgoyal12:35
-!- PhilTillet [] has joined #shogun12:40
-!- gsomix [~gsomix@] has joined #shogun12:47
-!- blackburn [~qdrgsm@] has joined #shogun12:55
-!- blackburn [~qdrgsm@] has quit [Client Quit]12:56
-!- puneetgoyal [~chatzilla@] has quit [Remote host closed the connection]12:59
-!- blackburn [~qdrgsm@] has joined #shogun13:01
gsomixblackburn, yo13:01
blackburngsomix: this time feel free to contact me via jabber :)13:02
blackburngsomix: or personally13:02
gsomixblackburn, 5 minute13:05
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]13:06
-!- puneetgoyal [~puneet@] has joined #shogun13:16
-!- menonnik [~chatzilla@] has joined #shogun13:51
-!- mayanks43 [~Marcus@] has joined #shogun13:52
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 272 seconds]14:16
-!- blackburn [~qdrgsm@] has joined #shogun14:16
-!- PhilTillet [] has quit [Ping timeout: 272 seconds]14:38
-!- gsomix [~gsomix@] has joined #shogun15:01
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds]15:05
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun15:08
-!- in3xes [~in3xes@] has joined #shogun15:09
-!- vikram360 [~vikram360@] has joined #shogun15:10
-!- gsomix [~gsomix@] has quit [Ping timeout: 246 seconds]15:33
blackburnsonney2k: sonne|work: are any of you guys around? ;)15:35
-!- gsomix [~gsomix@] has joined #shogun15:45
-!- in3xes [~in3xes@] has quit [Ping timeout: 246 seconds]16:08
-!- menonnik [~chatzilla@] has quit [Quit: ChatZilla 0.9.88 [Firefox 10.0.2/20120215223356]]16:16
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]16:18
-!- puneetgoyal [~puneet@] has quit [Ping timeout: 252 seconds]16:18
-!- in3xes [~in3xes@] has joined #shogun16:20
-!- vikram360 [~vikram360@] has joined #shogun16:27
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 252 seconds]16:28
-!- gsomix [~gsomix@] has quit [Ping timeout: 246 seconds]16:52
-!- vikram360 [~vikram360@] has quit [Ping timeout: 276 seconds]16:53
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun16:56
-!- cwidmer [] has joined #shogun16:57
-!- cwidmer is now known as ckwidmer16:57
-!- ckwidmer [] has quit [Client Quit]16:59
-!- vikram360 [~vikram360@] has joined #shogun17:02
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]17:15
-!- vikram360 [~vikram360@] has joined #shogun17:15
-!- wiking [] has joined #shogun17:17
-!- wiking [] has quit [Changing host]17:17
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun17:17
-!- cwidmer [] has joined #shogun17:21
-!- cwidmer [] has quit [Client Quit]17:21
-!- ckwidmer [] has joined #shogun17:21
-!- in3xes [~in3xes@] has quit [Ping timeout: 252 seconds]17:50
-!- harshit_ [~harshit@] has joined #shogun17:51
-!- ckwidmer [] has quit [Quit: - Chat comfortably. Anywhere.]17:56
-!- mayanks43 [~Marcus@] has quit [Quit: Leaving.]18:02
-!- in3xes [~in3xes@] has joined #shogun18:06
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:17
-!- in3xes [~in3xes@] has quit [Ping timeout: 245 seconds]18:21
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun18:33
-!- in3xes [~in3xes@] has joined #shogun18:33
-!- in3xes [~in3xes@] has quit [Remote host closed the connection]18:37
-!- in3xes [~in3xes@] has joined #shogun18:39
-!- romi_ [~mizobe@] has quit [Ping timeout: 246 seconds]18:41
-!- PhilTillet [] has joined #shogun18:43
-!- romi_ [~mizobe@] has joined #shogun18:54
-!- harshit_ [~harshit@] has quit [Ping timeout: 252 seconds]19:01
-!- romi__ [~mizobe@] has joined #shogun19:05
-!- romi_ [~mizobe@] has quit [Ping timeout: 260 seconds]19:05
-!- harshit_ [~harshit@] has joined #shogun19:12
-!- emrecelikten [~emre@] has joined #shogun19:16
-!- gsomix [~gsomix@] has joined #shogun19:21
-!- emrecelikten [~emre@] has quit [Quit: Leaving.]19:23
-!- harshit_ [~harshit@] has quit [Ping timeout: 240 seconds]19:28
PhilTilletHey :)19:33
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:50
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]19:55
-!- gsomix [~gsomix@] has joined #shogun19:55
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]20:01
-!- gsomix [~gsomix@] has joined #shogun20:01
-!- harshit_ [~harshit@] has joined #shogun20:01
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun20:06
-!- in3xes [~in3xes@] has quit [Remote host closed the connection]20:32
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Quit: This computer (or maybe me) has gone to sleep]20:42
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:47
-!- romi__ [~mizobe@] has quit [Ping timeout: 246 seconds]20:47
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun20:59
-!- romi__ [~mizobe@] has joined #shogun21:00
@sonney2kPhilTillet, any progress on opencl'ing linear combinations of kernels?21:03
PhilTillethello sonney2k21:04
PhilTilletyes :)21:04
PhilTilletI have implemented the formula with a gaussian kernel and random samples21:04
PhilTilletbut I don't really know where to start to integrate it21:04
@sonney2kwell did you benchmark it?21:04
@sonney2kI mean that is what is most interesting for now or?21:05
PhilTilletyes :) I wanted to benchmark it but needed a dataset21:05
shogun-buildbotbuild #185 of nightly_none is complete: Failure [failed compile]  Build details are at
@sonney2kwell random data again21:05
@sonney2krandom alphas...21:05
PhilTilletis it normal to get k(x,xi) = 0 for random data of very high values?21:06
PhilTillet(between 0 and 1)21:06
@sonney2kgaussian kernel is between 0 and 121:06
PhilTilleti think it is because the norm of the difference becomes infinity21:06
@sonney2k1 on diagonal21:06
PhilTilletbut the opencl computation is not optimized at all for now, just wanted to make it work21:07
PhilTilleti'll benchmark it right away21:08
PhilTillet(for now it is really really really not optimized though)21:09
PhilTilletone second for 1000*10000 data matrix21:15
PhilTilletbut it is really not optimized21:15
shogun-buildbotbuild #184 of nightly_all is complete: Failure [failed compile]  Build details are at
gsomixsonney2k, hi21:20
@sonney2khi gsomix21:20
gsomixit seems that python_static works21:22
@sonney2kgsomix, nice21:24
@sonney2kdid you need to do lots of fixes for that?21:24
PhilTilletsonney2k, do you think I should try to optimize my OpenCL code or try to somewhat integrate it ? :)21:24
@sonney2kPhilTillet, how many alphas did you have in your setup?21:25
@sonney2kand 1000 test examples of which dimensionality?21:25
PhilTilletI think I am confused21:25
@sonney2kyou should compare this to the non opencl variant21:25
@sonney2kwhen you do k(x, x_i)21:26
@sonney2kwhat is the dimension of x?21:26
@sonney2kand how many examples did you test this with? 1 then?21:26
PhilTilletOkay I think there is some point I didn't understand21:26
PhilTilletin the formula here21:27
PhilTilletI have : N = 10 000 and x.size() = 1 00021:27
@sonney2k(have to leave train soon)21:27
PhilTilletand computed one f(x)21:27
@sonney2kyeah that is evaluated for 1 example21:27
@sonney2kbut that is not what you want to optimize for21:27
@sonney2kyou should assume that this will be evaluated for thousands of x21:28
PhilTilletoh okay21:28
@sonney2kso the overhead of transfering alpha / x_i into GPU mem pays off21:28
PhilTilletwell, the transfer is at 16 GB/s , so it is more or less instant :p21:28
-!- n4nd0 [] has joined #shogun21:29
PhilTilletI think I get it21:29
PhilTilletThere is a big matrix A, and f is computed for each column of A, using the other columns as x_i21:29
PhilTillet(I should watch again my courses on SVM :p)21:30
@sonney2kPhilTillet, otherwise you cannot hope to be faster at all because all there is is reading over x_i / alpha_i x once21:34
-!- l0nr4n [] has joined #shogun21:35
-!- l0nr4n [] has quit [Changing host]21:35
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun21:35
PhilTilletah wait, I said 1 sec but i made a huge mistake benchmarking21:35
gsomixsonney2k, no, not much21:36
gsomixi learned how to do git branches (tnx blackburn).21:39
harshit_hello sonney2k, can you please help me on minimum residual function . i found one in lapack (GMRES.f), but it is not in the lapack library in <shogun/mathematics/lapack.h>21:39
gsomixi can try to make small commits now, instead of one large :)21:39
@sonney2kgsomix, very good :)21:44
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]21:44
-!- gsomix [~gsomix@] has joined #shogun21:45
-!- romi__ [~mizobe@] has quit [Ping timeout: 246 seconds]21:46
n4nd0sonney2k: hi there! I am sorry to insist on the same but do you think it could be possible to apply to "various usability impromevents" even if I have not experience with swig?21:48
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Ping timeout: 265 seconds]21:48
-!- l0nr4n [] has joined #shogun21:48
-!- l0nr4n [] has quit [Changing host]21:48
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun21:48
@sonney2kn4nd0, well you should attempt to show that you can deal with swig then21:48
n4nd0sonney2k: my first option is SO but I think it might be better if I apply to another project as well, just to get better odds21:48
@sonney2kn4nd0, do you know of anyone else applying for SO?21:49
n4nd0sonney2k: I have not heard of anyone, neither here nor in the mailing list21:49
@sonney2kn4nd0, regarding how to show - write some small typemap - e.g. the one missing in octave modular...21:50
@sonney2k...have to leave train.21:50
n4nd0sonney2k: ok, see you later21:50
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:52
n4nd0harshit_: hey!21:59
n4nd0harshit_: I read your question, have you in any case tried to use it?21:59
harshit_hey n4nd022:00
n4nd0harshit_: I think it is not necessary for it to be defined in shogun/mathematics/lapack.h, you may use the function from the lapack library22:00
harshit_but lapack library is in fortran, how can i use it in c++22:00
n4nd0harshit_: one thing is the language the code is written and another the library22:01
n4nd0harshit_: our C++ code in shogun is used from python, java, matlab, octave etc using swig for example22:01
harshit_okay, havnt done that before22:02
n4nd0harshit_: if you look at lapack.cpp22:02
n4nd0harshit_: that's C++ code, but is it calling the functions from lapack22:02
harshit_so in that sense i can add new22:03
harshit_functions in lapack.cpp22:03
n4nd0in any case, I am not completely sure about this22:03
harshit_and then use it in my program22:03
n4nd0I recommend you to try22:03
n4nd0yes, you could do that22:03
harshit_i'll try that22:04
harshit_btw which project are you going for in gsoc22:05
n4nd0structured output22:05
-!- akarshanSK [~akarshan@] has joined #shogun22:06
harshit_newtonSVM and integration of new versions of liblinear22:07
harshit_also i was thinking to give another proposal for C5.022:07
harshit_decision trees22:08
harshit_is that possible, To give 2 proposals to same organization22:08
n4nd0the deep learning stuff didn't succeed finally?22:08
n4nd0yes, no problem with that22:08
n4nd0the only limitation is 20 proposals22:08
n4nd0you can distribute them as you wish22:09
harshit_nah, didn't get any update on that22:09
harshit_also there are not much libraries for deep learning22:10
harshit_that can be integrated into shogun22:10
harshit_Any other organization you applying for ?22:11
-!- harshit_ [~harshit@] has quit [Ping timeout: 260 seconds]22:17
-!- harshit_ [~harshit@] has joined #shogun22:22
harshit_@n4nd0 no, thought of applying to orange, but then dropped the idea22:24
-!- harshit_ [~harshit@] has quit [Quit: Leaving]22:33
-!- puffin444 [230bf329@gateway/web/freenode/ip.] has joined #shogun22:37
-!- blackburn [~qdrgsm@] has joined #shogun22:38
gsomixsonney2k, hey?22:40
-!- akarshanSK [~akarshan@] has quit [Quit: Leaving]22:40
@sonney2kgsomix, yes?22:42
blackburnsonney2k: I need to discuss my issue ;)22:43
blackburnn4nd0: hey I'm here finally22:43
blackburnn4nd0: why do you want to abandon SO idea?22:44
gsomixsonney2k, i have problem in "./configure" code.22:44
-!- puffin444 [230bf329@gateway/web/freenode/ip.] has left #shogun []22:45
-!- Miggy [~piggy@] has joined #shogun22:45
gsomixthis is related to get flags for linking python's library22:46
@sonney2kgsomix, can you be more specific?22:46
-!- romi_ [~mizobe@] has joined #shogun22:47
gsomixsonney2k, distutils.sysconfig.get_config_vars('SO') for python3 gives a result different from python222:48
Miggysonney2k:i am interested in working on the project of implementing decision how should i proceed with it?22:48
gsomixsonney2k, see ". / configure" on line 4302. the results are "?" for python3 and ".so" for python222:50
CIA-64shogun: Sergey Lisitsyn master * r2002a08 / src/interfaces/modular/Library.i : Late fix for java serialization dummies -
blackburnshogun-buildbot: work!22:51
shogun-buildbotWhat you say!22:51
Miggyblackburn: which project are u choosing this time? :P22:51
@sonney2kgsomix, maybe you can workaround this by doing distutils.sysconfig.get_config_vars('SO')[0].rsplit('.')[-1]22:51
blackburnMiggy: I want to work on domain adaptation and MT learning22:51
blackburnMiggy: you are asking in a way that makes me think you know who I am and what is I was working on before ;)22:52
Miggyyes i know u worked on dimensionality reduction before :)22:53
gsomixsonney2k, may be necessary to use distutils.sysconfig.get_config_vars('BLDLIBRARY')? it returns "-L. -lpython3.2"22:54
@sonney2kgsomix, -L. is not what we want though22:54
Miggysonney2k: waiting for ur reply22:55
blackburnMiggy: while Soeren is in ignore mode I would suggest you to contribute with something small22:55
Miggyhehe :)22:55
Miggyrelated to decision tress or in general?22:55
blackburnit is mainly up to you22:56
blackburndecision trees in C++ can take a while22:56
@sonney2kgsomix, look - scipy people had the same problem22:57
blackburnMiggy: were you applying last year?22:57
blackburnaha I see22:58
Miggyonly 5 slots :(22:58
blackburnyeah it was tight22:58
PhilTilletsonney2k, I made a mistake on my benchmark, the result was 1ms and not 1second :p22:58
Miggythis time expecting more?22:58
blackburnthis year we expect a little more22:58
blackburnno reason to really expect more although :D but makes sense cause we finished with glance22:59
gsomixsonney2k, tnx22:59
blackburnshogun-buildbot: !help22:59
Miggythe main problem is with the priority of the projects...some project u like to work on does not get selected :(23:00
blackburnMiggy: I hope most interesting projects would be covered this year23:01
@sonney2kPhilTillet, what would be really interesting is to see a comparison to CPU version and for multiple x ...23:02
blackburnwhat can I say is some file formats idea/mloss idea/ update ocas have lower priority23:02
blackburnactually ocas is up-to-date here23:02
PhilTilletsonney2k, yes, sure, but I think the computation of a single x with the parallel algorithm is way faster23:02
@sonney2kI don't think so23:03
@sonney2kwhy should it?23:03
PhilTilletbecause everything is done in a parallel maner23:03
Miggyblackburn: ok so i want to know one thing...the previous 5 students can also apply this how can a new student get chance?23:03
PhilTilletwith a looot of threads23:03
blackburnMiggy: only 2 of us are applying23:03
@sonney2kPhilTillet, yes but bottleneck should be memory accesses23:04
Miggyu and?23:04
@sonney2kso it doesn't matter if GPU or CPU is waiting23:04
blackburnMiggy: actually we have equal chances :)23:04
blackburnI mean each should proceed in standart way23:04
PhilTilletsonney2k, Oh I understand your point23:04
blackburnMiggy: rest have dissappeared :D23:04
@sonney2kPhilTillet, that is why I am saying try for multiple x (say 10k or more)23:05
Miggyi mean they will definitely pick you because u have worked once with them23:05
@sonney2kand dim(x) = 100 or so23:05
blackburnMiggy: even in this case we would have 6-7 slots and there is a room23:05
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]23:06
PhilTilletsonney2k, are you talking about the kernel function?23:06
Miggyi see23:06
PhilTillet . I thought on this formula the number of examples was N23:07
PhilTilletthis is why I am confused23:07
Miggyi hope its not too late to contribute23:07
@sonney2kPhilTillet, the number of support vectors is N23:07
@sonney2kthe number of vectors this formula is applied to is just 123:07
@sonney2k(just one x)23:07
PhilTilletOh, I get it :p23:07
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]23:07
blackburnMiggy: sure, not late23:08
Miggyso how many students are currently active on shogun?23:09
Miggythis would give me an idea of my standing :P23:09
blackburnMiggy: we've got contributions from 3 students already23:10
blackburn(w/o me and Heiko)23:10
blackburnlast year there was a simple task which I come up with23:10
blackburnkernel functions23:10
PhilTilletI understand sonney2k, it's like the difference between a matrix-vector and a matrix-matrix product :)23:11
Miggyok so can u suggest me something?23:11
Miggywhat type of contributions have they done?23:11
blackburnMiggy: ok let me try23:11
blackburnwiking has done HomogeneousKernelMap preprocessor23:11
blackburnn4nd0 was working on qda23:12
blackburngsomix are working on python323:12
blackburnMiggy: how familiar you are with shogun's internals?23:12
Miggyhad a look last year23:13
Miggybut was able to code comfortably23:13
blackburnI am pretty interested with least angle regression but no idea if it would be appropriate to suggest23:14
Miggysmthing in classification or clustering?23:15
blackburnMiggy: some dataset downloader doing classification/regression and printing out results would be nice23:15
PhilTilletsonney2k, support vectors and examples are stored columnwise, right ?23:16
Miggyok will look into that23:16
@sonney2kPhilTillet, yes23:16
PhilTilletokay :)23:17
blackburnMiggy: some simple extension like new kernel/distance are welcome as well23:17
Miggyall the kernel functions have been already done :P23:17
blackburnyeah even mahalanobis (which I forgot)23:18
blackburnwas done by n4nd023:18
Miggyso prvious year students who g23:18
Miggyprevious year students also applying this time?23:18
Miggythe one who contributed previously?23:19
blackburnno, only 223:19
blackburnwe both have been contributing sinch last pre-gsoc period23:19
Miggyi mean those who did smthing but were not selected23:19
blackburnI think you are the first one of that kind23:19
Miggywell i was late so i was not abe to give any patch that time23:20
blackburnI see23:20
Miggyi got to know abt shogun a little late :/23:20
blackburnsonney2k: sudo answer me23:20
blackburndamn is he kidding23:21
wikingblackburn: don't forget the great jensen kernel :)))23:21
blackburnoh yeah, Miggy: wiking has also done Jensen-Shannon kernel23:21
blackburnI have not been aware of it for whole life23:22
Miggyerr need some good patch23:23
blackburnwiking: in fact I did not manage to use J-S explicitly23:23
Miggyall have been done :P23:23
blackburnit takes %INFINITY% time23:23
blackburnbut marvellous homogay :D map is awesome23:24
wikingbut then i guess with the homomap was good ;)23:24
blackburnbest results are with intersection23:24
blackburnno idea why23:24
wikingmmm i've seen somebody today sending an email wanting to do the latent thingy23:24
blackburnwiking: well you are a strong candidate on latent svms :D23:25
wikinghehehhe :>23:25
wikingi have to do some work with it23:25
wikingbut now i have so much stupid distractions23:25
wikingi might be able to do it tomorrow23:25
n4nd0blackburn: not abandoning SO man, no way :)23:25
wikingat least class definitions23:25
n4nd0blackburn: I am really investing time on there23:25
blackburnn4nd0: so what is the problem?23:26
n4nd0blackburn: none, I thought it could be better to apply to a second project just in case SO doesn't get selected finally or another studen takes it23:26
n4nd0blackburn: what do you think?23:26
blackburnn4nd0: ah sure23:26
n4nd0blackburn: I want to do SO for sure, but I'd like to get better chances23:26
blackburnn4nd0: then you may also apply for ECOC/MC idea23:27
n4nd0blackburn: let me check23:27
blackburnwiking: btw in this vedaldi paper there are map for gaussian kernel, right?23:28
wikingblackburn: not that i know of23:28
wikingor i don't remember it having23:28
n4nd0blackburn: which one is that? ECOC/MC23:28
blackburnare there maps only for additive kernels?23:28
blackburnn4nd0: multiclass23:28
blackburnmentored by Soeren23:29
n4nd0blackburn: isn't it that your target?23:29
blackburnn4nd0: no23:29
n4nd0blackburn: aham ... I thought that!23:29
-!- PhilTillet_ [] has joined #shogun23:29
blackburnn4nd0:  noo, I apply for mt/da23:29
n4nd0blackburn: I have not even read that because of that :P23:29
blackburnI apply for multitask learning with christian23:30
-!- PhilTillet_ [] has quit [Client Quit]23:30
n4nd0blackburn: all right, I will check that other then before starting to see how swig works23:30
n4nd0blackburn: anyhow, time to fix the repo ...23:31
blackburnoh damn yeah23:31
n4nd0blackburn: so do you think the best is to kill the fork and do it again?23:31
blackburnn4nd0: yeah the easiest way for me23:31
blackburnn4nd0: swig would be competetive cause there is a student wants to work on that already23:32
blackburnbut multitask is pretty free :)23:32
n4nd0blackburn: ok good tip man, thank you very much :)23:32
n4nd0blackburn: you ok :P?23:32
blackburnok late here -> sleep time23:34
n4nd0good night then23:34
blackburngood night23:34
blackburnsonney2k: sudo ignore-mode-off!23:34
blackburnn4nd0: kick him when he comes23:35
blackburnsee you23:35
n4nd0blackburn: we'll see23:35
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 276 seconds]23:40
shogun-buildbotbuild #384 of java_modular is complete: Success [build successful]  Build details are at
PhilTillethello, shogun-buildbot ! tired to be a bot? :/23:49
n4nd0shogun-buildbot: work!23:50
shogun-buildbotWhat you say!23:50
n4nd0he has worked enought today maybe ...23:50
PhilTilletyes :)23:51
--- Log closed Thu Mar 22 00:00:19 2012