Open in new window / Try shogun cloud
--- Log opened Sat Feb 15 00:00:59 2014
wikinglisitsyn: hey00:02
thoralfHey wiking00:06
-!- HeikoS [] has joined #shogun00:14
-!- mode/#shogun [+o HeikoS] by ChanServ00:14
-!- HeikoS [] has quit [Ping timeout: 245 seconds]00:18
wikingmorning :)00:53
-!- thoralf [] has quit [Quit: Konversation terminated!]01:08
-!- bumblebee [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun06:09
-!- bumblebee is now known as Guest3310406:10
Guest33104hey!! is anyone around?06:14
Guest33104wanted to know the openings regarding opencv integration in shogun?06:16
-!- bumblebee_ [~quassel@] has joined #shogun06:31
-!- bumblebee_ [~quassel@] has quit [Client Quit]06:35
-!- bumblebee_ [~quassel@] has joined #shogun06:35
@sonney2kbumblebee_, ask pickle27 here on monday06:38
@sonney2kbumblebee_, and no people are usually asleep atm06:38
bumblebee_i guess it was listed in the ideas page? Is is under maintainance ?06:43
-!- Guest33104 [ca4eaca2@gateway/web/freenode/ip.] has quit [Quit: Page closed]06:44
-!- dsockwell [~twgs@] has left #shogun []06:44
-!- bumblebee_ [~quassel@] has quit [Remote host closed the connection]07:24
-!- pranet [73f91219@gateway/web/freenode/ip.] has joined #shogun10:31
pranetHi. Am getting a "no module named shogun" error when doing "import shogun" in python. Any tips?10:32
-!- ajgar [7aa7f31e@gateway/web/freenode/ip.] has joined #shogun10:33
besser82pranet:  are you using python_modular from shogun?10:35
ajgarI installed shogun cloning from github but I'm having trouble with the library path.. can anyone help me..?10:36
besser82pranet:  then you should import "modshogun"  ;)10:36
besser82ajgar:  currently no, unfortunately.  But I'll be back in ~one and a half hour to help you  ;)10:36
besser82ajgar:  I think we can work stuff out then  ;)10:37
ajgarbesser82 no problem.. thanks a lot.. :)10:37
pranetajgar. what problem with library path?10:41
pranetbesser: umn nop. il try that10:41
pranetajgar: is it the missing .so files?10:42
ajgarpranet: after "make all" , i did a sudo make install10:42
ajgarthis is supposed to add libshogun to usr/lib right..?10:42
pranetu might have to run ldconfig explicitly. that worked for me10:43
ajgarok thanks. can you pls elaborate more on that..?10:44
pranethave you tried compiling a cpp program yet?10:45
ajgari did. compilation failed saying10:46
ajgarfatal error: shogun/base/init.h: No such file or directory10:46
pranethmn ya i got that too. leme remember what i did10:47
pranetfirst try "ldconfig" in terminal10:47
pranetsee if that helps10:48
ajgarit compiles if i explicity pass paths to lib and include folders.10:48
ajgarlike this. g++ -I/path_to_include _L/path_to_lib test.cpp -lshogun10:49
-!- Saurabh7 [~Saurabh7@] has joined #shogun10:50
ajgarpranet ldconfig did not work :(10:54
pranethmn . sec10:55
pranetneed to find out how to add the libs to default path10:56
pranetdoes ur usr/local/include have shogun?10:58
ajgarlet me check10:58
ajgarno it does not.10:58
pranethmn ya thats the problem. now lets see how to add it10:59
pranetthis is how u installed right?11:05
pranet1. go to the shogun repository root 2. do `mkdir build` 3. do `cmake [options] ..` (or `ccmake ..` if available). It is very recommended to use any of CMake GUIs (such as ccmake) if you feel unsure about possible parameters and configurations. 4. do `make` (and `sudo make install` if needed)11:05
pranetwell if u just want a working fix i could mail u the folder ur supposed to paste in usr/local/include. that may work11:05
pranetor just try pasting ur src/shogun folder there11:06
ajgari read that copy pasting into usr/local/include could be dangerous and should not be done.. anyhow its working if I pass the path in arguments..11:08
ajgarI have already spent a lot of time setting up all the dependencies. I dont want to break anything.. :D11:08
pranetoh didnt know about that.11:08
pranetwell technically since the locatin is empty how could it be dangerous11:09
pranetworst case it doesnt work, u just delete it11:09
pranetagain , just an opinion :P11:09
ajgarI am not sure how it is dangerous..11:10
ajgarits working for now..let it be like this until a proper fix can be found. pls do tell me if you figure it out..11:11
ajgarand thanks a lot for the help.. :)11:12
-!- pranet_ [73f91219@gateway/web/freenode/ip.] has joined #shogun11:13
-!- pranet [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]11:13
-!- pranet_ [73f91219@gateway/web/freenode/ip.] has quit [Client Quit]11:14
-!- pranet [73f91219@gateway/web/freenode/ip.] has joined #shogun11:16
pranetnpz :D11:20
pranetanyone knows whats python_modular?11:20
-!- prams [~prams@] has joined #shogun11:23
-!- ajgar [7aa7f31e@gateway/web/freenode/ip.] has left #shogun []11:24
-!- suryak [0e8b9b42@gateway/web/freenode/ip.] has joined #shogun11:39
suryakHello there, I just had a look around the GSoC 2014 project ideas on your site and got a bit of interest to look deeply.11:41
suryakI have hands on experience in Machine Learning.. in which I know algorithms like logistic regression, k-means clustering, svm11:43
suryakand regarding programming background - I know Python, and little bit of MATLAB; I have previously done some Web programming in Django11:44
suryakso, basically, is there anyone I can talk to on the ideas, I appreciate; thanks11:44
-!- pranet [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]11:47
-!- lisitsyn [~lisitsyn@] has quit [Ping timeout: 252 seconds]11:47
suryakanyone there?11:55
-!- iglesiasg [] has joined #shogun11:57
-!- mode/#shogun [+o iglesiasg] by ChanServ11:57
-!- suryak [0e8b9b42@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]12:10
-!- prams [~prams@] has quit []12:17
-!- pranet [ca4eafc7@gateway/web/freenode/ip.] has joined #shogun12:33
-!- pranet [ca4eafc7@gateway/web/freenode/ip.] has quit [Quit: Page closed]12:46
@iglesiasgI could really be using some help while I am building the website with the project ideas13:09
@iglesiasgsomething as simple as proof-reading could help me a lot13:09
@iglesiasgso please if anyone can, let me know13:09
-!- HeikoS [] has joined #shogun13:19
-!- mode/#shogun [+o HeikoS] by ChanServ13:19
-!- Saurabh7 [~Saurabh7@] has quit [Ping timeout: 248 seconds]13:24
@iglesiasghey HeikoS13:26
-!- lisitsyn [~lisitsyn@] has joined #shogun13:29
lisitsyniglesiasg: hey13:30
@iglesiasglisitsyn, hello hello13:31
lisitsyniglesiasg: how do you handle this stuff?13:33
@iglesiasglisitsyn, yes13:33
@HeikoSiglesiasg:  hi!13:35
lisitsyniglesiasg: what repo? ;)13:36
@iglesiasglisitsyn, aah the one in the mail13:37
lisitsyniglesiasg: what do you move to the webpage right now?13:39
@iglesiasglisitsyn, I just pushed the last part I have stable13:42
@iglesiasglisitsyn, so pretty much, now the ideas that are really small are left, so very little work13:42
@iglesiasglisitsyn, I think I can do that all right, but someone should proof read13:42
lisitsyniglesiasg: let me add a few more words to kevin's opencv thing13:42
@iglesiasgand also, I guess the ideas that are small should be improved13:42
@iglesiasglisitsyn, mmmm13:42
lisitsyncould you push it to the page then?13:42
@iglesiasghow should we do it?13:43
@iglesiasgsoon we will start having things in two places13:43
@iglesiasgso I guess it is not good to update the drive more13:43
lisitsyniglesiasg: well true for ideas that are online13:43
lisitsynbut these are not13:43
@iglesiasglisitsyn,  I am touching the HTML for these ones right now :)13:43
lisitsyniglesiasg: ahh13:43
@iglesiasglisitsyn, let me push the opencv right now for you13:44
lisitsynso you have them in the html right now?13:44
@iglesiasgand you can go on with the html, all right?13:44
-!- HeikoS [] has quit [Quit: Leaving.]13:44
lisitsynyeah okay13:44
lisitsynlets do it13:44
lisitsyniglesiasg: have you received any mails yet?13:45
-!- HeikoS [] has joined #shogun13:46
-!- mode/#shogun [+o HeikoS] by ChanServ13:46
@iglesiasglisitsyn, yes13:47
@iglesiasglisitsyn, and you?13:47
lisitsyniglesiasg: I have one with no notion of project :D13:47
lisitsynI guess this doesn't matter13:47
@iglesiasglisitsyn, ok! I just pushed everything13:49
lisitsyniglesiasg: cool thanks13:49
@iglesiasglisitsyn, so I have put the output kernel learning project as placeholder13:49
@iglesiasglisitsyn, you can find the project you want to modify by having a look at the comments13:50
@iglesiasglisitsyn, for instance, <!-- OPENCV PROJECT IDEA -->13:50
lisitsyniglesiasg: alright13:50
@iglesiasglisitsyn, I am working on the remaining of ML tasks now, so we try to avoid conflicts when merging later13:51
lisitsyniglesiasg: PR-113:51
-!- prams [~prams@] has joined #shogun13:52
lisitsyniglesiasg: I think they should be pretty mergeable13:52
@iglesiasglisitsyn, I hope so13:52
@iglesiasglisitsyn, I have never really done anything with html and version control13:52
@iglesiasgcomplete surprise for me what will happen :D13:52
-!- HeikoS [] has quit [Quit: Leaving.]13:53
-!- prams [~prams@] has left #shogun []13:53
-!- prams [~prams@] has joined #shogun13:55
-!- pranet [ca4eafc7@gateway/web/freenode/ip.] has joined #shogun13:58
@iglesiasglisitsyn, is it ok for you if I port the other ideas in infrastructure except from OpenCV one?14:04
alcedineI think your GSoC ideas page is a bit broken. The three ideas after "Dictionary Learning" all show the description for "Output Kernel Learning" instead.14:05
-!- Saurabh7 [~Saurabh7@] has joined #shogun14:06
@iglesiasgalcedine, if you read a bit above, you will see that we are currently adding the ideas14:13
alcedineAh, indeed. Sorry. (:14:14
@iglesiasglisitsyn, still there?14:27
lisitsyniglesiasg: yeah14:27
@iglesiasglisitsyn, just opencv left to remove all placeholders14:28
@iglesiasglisitsyn, maybe we could push something even if it is not completed?14:28
lisitsyniglesiasg: yeah sure14:28
@iglesiasglisitsyn, PR-me baby :)14:30
lisitsyniglesiasg: will be back in a few mins14:31
@iglesiasglisitsyn, ok, then let me push the current version in the google docs14:31
@iglesiasglisitsyn, I have to go soon, and want to get rid of placeholder first14:31
-!- Saurabh7 [~Saurabh7@] has quit [Read error: Connection reset by peer]14:35
-!- prams [~prams@] has left #shogun []14:39
-!- bumblebee [~quassel@] has joined #shogun14:42
-!- bumblebee is now known as Guest7032214:42
-!- Guest70322 [~quassel@] has quit [Remote host closed the connection]14:43
-!- bumblebee_ [~quassel@] has joined #shogun14:44
-!- pranet [ca4eafc7@gateway/web/freenode/ip.] has quit [Quit: Page closed]14:51
-!- Saurabh7 [~Saurabh7@] has joined #shogun14:53
-!- Saurabh7 [~Saurabh7@] has left #shogun []14:53
lisitsyniglesiasg: ok back14:54
@iglesiasglisitsyn, I think the ideas page is in good shape now14:56
lisitsyniglesiasg: looks kool indeed14:57
@iglesiasgit is likely that there will be errata, but we will polish those with a little bit more of time14:58
-!- HeikoS [] has joined #shogun15:04
-!- mode/#shogun [+o HeikoS] by ChanServ15:05
-!- iglesiasg [] has quit [Quit: Leaving]15:09
-!- votjakovr [~votjakovr@] has joined #shogun15:24
-!- bumblebee_ [~quassel@] has quit [Ping timeout: 260 seconds]15:27
@sonney2kHeikoS, hey - did you find the time to write this 'how to do gsoc with us' document?15:54
@HeikoSsonney2k: not yet, but will start tonight15:55
@sonney2kHeikoS, I am just now looking at the ideas page. Looks like the creating packages & win32 port is missing and also shogun & teaching.15:57
@sonney2kHeikoS, I am not sure about the RBM stuff - I think a NN framework that is trained based on standard algorithms would be more valuable15:59
@sonney2k(standard algorithms == already available implementations)15:59
@HeikoSsonney2k: all this deep learning stuff is all massively depending on implementation details, so I would rather interface to existing ones16:00
@HeikoSyeah the packageing is important, and teaching .....16:00
@sonney2kHeikoS, we have 2 labs supporting us with the teaching thing at least partially16:01
@sonney2kHeikoS, and this would 'just' mean write good notebooks and demos16:02
@HeikoSits a bit hard with mentors16:02
@HeikoSbut since this stuff is not so much about Shogun, it might be easier16:02
* sonney2k is reading the python3.3.4 source code to understand pickling16:16
@HeikoSwhoooo :)16:22
votjakovrHeikoS: hi! how is it going with the kernel->clone() issue? I've fixed parameter's hashing framework a little bit. Now GPs are training one time.16:23
@HeikoSvotjakovr: great work!16:23
@HeikoSvotjakovr: yeah I looked into that, it has to do with SGVectors of CSGObjects16:23
votjakovrHeikoS: But it depends on kernel->clone()16:24
@HeikoSI think this case not not handled properly16:24
@HeikoSso adding a unit test for that would help (it would fail for now)16:24
@HeikoSand then trying to find out the issue16:24
@HeikoSI am not sure I even thought about that when I wrote the clone stuff (the vector of sgobjects case)16:24
@HeikoSvotjakovr: I will get back on that later today and/or tomorrow16:24
@HeikoSbut cool that the GPs are faster now!16:24
@HeikoSI am using them quite a bit for my research btw16:25
votjakovrHeikoS: yep, I'll send a PR after kernel->clone() problem will be fixed16:25
@HeikoSvotjakovr: thanks!16:26
votjakovrHeikoS: Also i thought about adding numerical derivation to SHOGUN to find kernel derivatives wrt parameters, for which we can't do that analytically.16:28
votjakovrHeikoS: what do you think about that?16:33
@HeikoSvotjakovr: I agree on that, should be general purpose and a fall back option16:33
@HeikoSyes, good idea1!16:33
@HeikoSif you want to work on that16:33
@HeikoSthats even a whole gsoc project, and very useful for various things16:33
@HeikoSin general automagic differentiation btw16:33
@HeikoSalso non-numerical variants16:33
@HeikoSI gotta go now, so lets discuss later16:33
@HeikoSbut yeah +1 from my side :)16:34
votjakovrHeikoS: ok, thanks :) i'll think about that16:34
-!- votjakovr [~votjakovr@] has quit [Quit: WeeChat 0.4.0]16:51
@sonney2kI think torch, theano and caf* all have nice impl's written in C underneath that we might be able to use17:06
@sonney2kHeikoS, ^17:06
@sonney2kHeikoS, also it seems a bit easier than interfacing SVMs since they don't share a common kernel17:07
@sonney2kHeikoS, just some weights / architecture we have to pass to them and which training algo to use17:07
@sonney2kHeikoS, and then fetch all the results back in17:07
@HeikoSsonney2k: yeah exactly, much better, and we can plug things together with the rest of shogun17:10
@HeikoSso no need to implement those things17:11
@HeikoSwhich is hard and messy17:11
lisitsynsonney2k: sounds like my shogun vision (tm) :D17:32
lisitsynjust a thing that ties various things together17:32
lisitsynsonney2k: what's with fastfood? is it pt.2?17:55
-!- prams [~androirc@] has joined #shogun17:55
-!- iglesiasg [] has joined #shogun18:51
-!- mode/#shogun [+o iglesiasg] by ChanServ18:51
-!- zxtx [] has quit [Ping timeout: 260 seconds]18:52
@iglesiasgsonney2k, is it possible to make a setup so that we don't need to use the admin interface of the website to update article pages, e.g. gsoc ideas?19:05
-!- HeikoS [] has quit [Ping timeout: 246 seconds]19:11
-!- parijat [671b082a@gateway/web/freenode/ip.] has joined #shogun19:12
parijatsonney2k: hi, I know you guys are busy with gsoc. I had a doubt regarding PCA. Is it a good time to discuss?19:20
parijatsonney2k: the file was probably written by you so thats why I am reaching out to you19:20
@iglesiasgparijat, I am testing your file soon, let me see19:39
parijatiglesiasg: alright! thanks.19:43
@iglesiasgparijat, so there are a couple of things here19:44
@iglesiasgparijat, first, you should set the target  dimension before calling init19:44
parijatiglesiasg:  tried that as well. doesnt rectify the error19:45
@iglesiasgparijat, that will still give you a 3x4 matrix19:45
@iglesiasgparijat, you can check that it does different things by looking at the dimension of the matrix outputted by get_transformation_matrix19:46
parijatiglesiasg: yup but my target dimension is 2 right? so i am expecting 2x4 matrix19:46
@iglesiasgparijat, it would make sense if it does the operation in place, but I am not sure19:47
parijatiglesiasg: if you run kernelPCA instead with say gaussian kernel, it works perfect. I get a 2x4 matrix as output19:48
parijatiglesiasg: thats why i think there might be something wrong going on in PCA19:49
@iglesiasgparijat, in PCA::apply_to_feature_matrix for some reason it is fixed that the output has the same dimensions of the input19:49
@iglesiasgparijat, I don't really understand why though. I think you are right and it should output a 2x419:50
parijatiglesiasg: ya saw that! but that doesn't make sense19:50
@iglesiasgparijat, I mean that code is really crazy for what it actually has to do...19:50
@iglesiasgwhich is just a matrix multiplication19:51
parijatiglesiasg: :D19:51
@iglesiasgparijat, that is probably one of the things that Heiko meant when he said that (K)PCA could be pretty much rewritten19:51
lisitsyniglesiasg: remember we have it in tapkee? ;)19:52
@iglesiasgparijat, anyhow I would say you can do like this. For your example now, work with the matrix transformation given by PCA. You can do the product in Python easily19:52
@iglesiasglisitsyn, I don't actually!19:52
parijatiglesiasg: i tried printing the matrix m at the end of each iteration. 2 of 3 dimensions are updated per iteration. Its crazy! you should try that19:53
parijatiglesiasg: ok alright! let me try that19:53
@iglesiasgparijat, I used PCA for LMNN just using the outputted matrix transformation, so that should be ok19:55
parijatiglesiasg: alright I will have a look at your code for reference as well.19:56
parijatiglesiasg: but we should also try fixing this PCA bug (if there is one) sometime soon :)19:57
@iglesiasgparijat, yeah sure. It would be nice to remove the operations that use cblas there and use eigen to start with19:57
@iglesiasgparijat, take it as another entrance related to the fundamental ML project if you want ;)19:58
parijatiglesiasg: :D ya sure! my pleasure.19:58
parijatiglesiasg: thanks for looking into this.19:59
@iglesiasgno problem!20:00
@iglesiasgparijat, I am just going to answer the mail in the mailing list so there is an answer there20:10
parijatiglesiasg: ya sure go ahead. I was just going through the pca transform used in LMNN, I am mostly just translating the code to python :)20:14
-!- HeikoS [] has joined #shogun20:15
-!- mode/#shogun [+o HeikoS] by ChanServ20:15
@iglesiasgparijat, all right! Probably you just need to do get_transformation_matrix to get the PCA matrix and then to perform matrix multiplication between the feature matrix and the transformation matrix20:20
-!- zxtx [] has joined #shogun20:20
parijatiglesiasg: ya, but before that subtracting out the mean is also necessary maybe?20:21
@iglesiasgparijat, that is an additional step you may or may not do20:22
@iglesiasgparijat, I see that the PCA code does it, but it is not a compulsory step in standard PCA (I believe)20:22
parijatiglesiasg: ya thats why I planned on doing it. Maybe I will do it anyways. It isn't much of a work20:23
@iglesiasgindeed, just another 2-3 lines at most it must be :)20:24
-!- parijat [671b082a@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:36
-!- huska1 [] has quit [Quit: WeeChat 0.4.3]20:38
lisitsyniglesiasg: I think mean is important here ;)20:41
@iglesiasglisitsyn, I saw tapkee has it too so I started wondering hehe20:43
lisitsyniglesiasg: I think mean is changing the angle20:43
lisitsynspeaking of geometry20:43
-!- huska [] has joined #shogun20:43
lisitsynany prinicipal axis has different angle when you don't subtract mean and when you do20:44
-!- prams [~androirc@] has quit [Ping timeout: 252 seconds]20:48
-!- lisitsyn [~lisitsyn@] has quit [Read error: Connection reset by peer]20:50
-!- lisitsyn [~lisitsyn@] has joined #shogun20:50
-!- iglesiasg [] has quit [Ping timeout: 252 seconds]20:54
-!- iglesiasg [] has joined #shogun21:07
-!- mode/#shogun [+o iglesiasg] by ChanServ21:07
-!- iglesiasg [] has quit [Client Quit]21:07
-!- zxtx [] has quit [Ping timeout: 245 seconds]22:05
-!- zxtx [] has joined #shogun22:19
-!- prams [~androirc@] has joined #shogun22:28
-!- thoralf [] has joined #shogun23:35
-!- thoralf [] has quit [Quit: Konversation terminated!]23:44
--- Log closed Sun Feb 16 00:00:01 2014