Open in new window / Try shogun cloud
--- Log opened Sat May 28 00:00:38 2011
CIA-32shogun: Soeren Sonnenburg master * r8381b60 / src/configure : at least on debian based systems ruby libs are named libruby<VER>*.so* - http://bit.ly/lqKxF400:06
serialhexsonney2k, i'm updating my shogun now, will test the ruby_modular interface when it's built00:59
serialhexfscking slow-ass computer!!!02:02
serialhexsonney2k, my ruby doesn't work with your change, so we'll have to find a medium somewhere in between, it gives me a '/usr/bin/ld cant find -lruby1.8' error (i'm trying different configs & stuff and lost the exact msg, but i'll post it when i try that again)06:04
serialhexalso, i was having trouble loading modules, and i came across this: http://goo.gl/sHdbK and i went into each of the modules.  like it says, %module names should be lowercase, and since all of the modules (classifying, clustering, distance, etc...) are capitalized, we should do some non-break-backwards-compatibility thing like add either a test to see whether  we're compiling ruby or not in the Foo.i file, or in the ruby06:10
serialhexcompile opt include the "-module foo" the first is probably easier & more direct06:10
serialhexmy days have recently gotten very busy, as i *just* got a job (yay me!) so while i have some ideas on how to fix this stuff i might not get the time to actually do so...06:11
serialhex...and i just noticed sploving is on!06:12
serialhex...well, afk actually :P06:12
serialhexand i *really* wish my computer was faster, 20+ min to compile shogun is just sad!  but i got to sleep, so i'll see what i can find out tomorrow, and let sploving & sonney2k know what i find... take care!06:16
@sonney2kserialhex, congrats on the job!07:30
@sonney2kserialhex, I am doing the conditional rename for ruby now07:35
@sonney2kserialhex, regarding configure - we will need a test to compile a minimal ruby extension? do you have an example for that?07:39
-!- blackburn [~qdrgsm@188.168.2.233] has joined #shogun12:49
blackburnehh.. CMath::display_matrix is column major routine13:37
blackburnsonney2k: why it is?13:37
blackburnomg these matrices is rotating in LLE all the time13:41
blackburndamn it13:41
-!- blackburn [~qdrgsm@188.168.2.233] has quit [Ping timeout: 258 seconds]17:45
-!- blackburn [~qdrgsm@188.168.4.77] has joined #shogun19:45
blackburnsonney2k: reported ;)20:15
@bettyboo;>20:15
@sonney2kblackburn, all matrices are assumed to be column by column20:39
@sonney2kthat is the fortran way (tm)20:39
@sonney2kblackburn, thnanks20:39
blackburnsonney2k: I had a problem with set_feature_matrix20:39
blackburnwell I have eigenvectors as ROWS of matrix20:40
blackburnthen I have to choose m of them20:40
blackburnand columns of that matrix is feature matrix20:40
blackburneverything seems to be ok, right?20:40
blackburnbut when I set this feature matrix and print it in python - I have eh.. something like transposed20:41
blackburnright sizes - but not that order I think20:41
@sonney2kblackburn, it could of course be that I introduced bugs in the new typemaps20:44
@sonney2kit should be like 1 column == 1 vector20:44
blackburnfor the moment I just i<->j20:44
blackburnI don't know exactly20:44
blackburnthese matrix issues makes me mad20:44
@sonney2kand matrix == mulltiple columns concatenated20:44
@sonney2kdoes that match your assumption20:45
@sonney2k?20:45
@sonney2kif so try to set a feature matrix and get it afterwards20:45
blackburnin python?20:45
@sonney2kthen you will see if it is still the same20:46
blackburnsonney2k: have you seen pics in my report? looks like it doing the things right, but again I'm not confident with it20:47
blackburnanyway the possible 'error' could be only in this issue - changing feature matrix20:47
@sonney2kblackburn, I suggest to sleep one night over this and then check again21:07
@sonney2kI don't know...21:07
alesis-noviksonney2k, I assume code maintainability is more important than slight increase in performance.21:11
@sonney2kalesis-novik, depends...21:11
alesis-novikAs in using Gaussian class with GMM instead of having means and covariances in the GMM class itself21:11
@sonney2kusually yes though21:11
@sonney2kalesis-novik, how does having them in the gaussian class possibly slow down things?21:12
@sonney2kI would keep them there...21:13
alesis-novikWell, moving stuff around will take a bit of time, but like I said, nothing major. Will need to completely redo the Gaussian code though.21:35
alesis-novikBut since Gaussian isn't used anywhere else as far as I know, it won't cause any problems21:35
@sonney2kalesis-novik, don't you have covariance matrix / mean vector in gaussian too? what else is missing?21:47
alesis-novikyeah, but I need to represent the covariance depending on the type used (as svd or diagonal or spherical).21:50
alesis-novikSo the covariance field is effectively gone21:51
blackburnsonney2k: could you remind me the name of a tool for compile caching?22:31
serialhexhey blackburn, i think it's cmake23:06
blackburnhi, serialhex23:06
serialhex(i know, i'm like a half hour after the Q but whatev! i just got homw :P )23:06
serialhexhow are ya?23:06
blackburnfine, but sitting at home all day long and having a sexual time with some dim reduction things :D23:08
blackburnand you?23:08
serialhexreally?  you're having sex with some dim reduction stuff?? :P23:08
serialhexi'm doing well, just got done with my last day of training for my new job!!!23:09
* serialhex sees blackburn's spiffy 3d graph!!!23:09
blackburnwhat is your new job?23:09
blackburnserialhex: yeah it could show that method is working23:10
blackburnor not23:10
blackburnI don't know :D23:10
blackburnI have to check if neighborhood is 'saved' in 2d23:10
blackburndamn borg book about MDS have 612 pages23:10
serialhexyeah, that souds like it might work23:10
serialhexDAMN!!!23:11
blackburnseems they have wrote almost everything that they know23:11
serialhexwell my new job is really kind of cool, i'm selling really spiffy high quality knives!!23:11
serialhexyes it does!23:11
blackburnselling knives sounds very profitable23:13
serialhexi sometimes dont like books like that because theres no "this is what you *REALLY REALLY* need to know about this stuff" section in the beginning... so you dont have to grok the whole book23:13
blackburnat least you can threat someone with your goods :D23:13
serialhexbut still, it looks impressive on a shelf!! :D23:13
serialhexoh yeah!!23:13
blackburn'damn man i'll show you how sharp it is' :D23:14
blackburnhrshsh23:14
serialhexdude, you've probably never dealt with a cocanut in russia, but while i was in jamaica i had one and it took the dude 3 whacks with a machete to break open the sucker!23:14
blackburnhehe23:14
serialhexwith the cleaver that my boss (whos really hot and really awesome) let me use, i split that thing in HALF with one swing!!!23:15
serialhexit was freaking awesome!!!23:15
serialhexso yeah, i think it's going to be a lot of fun and probably very profitable!23:16
blackburnnice to you :)23:16
serialhextho i'm still working on some of the ruby things, but i gotta make it work first!! :(23:17
@sonney2kserialhex, I hope you can recommend me a good knive now :)23:17
serialhexhellz yeah!!!  i can probably sell em to ya if you like, they're really spiffy!!!23:18
@sonney2kall the ones I have here are just blunt23:18
@sonney2kserialhex, made in germany ;-)23:18
serialhexwell i'm not 100% sure if it'll work outside the US but they have a forever guarantee so you can get them sharpened whenever for the price of shipping23:19
serialhextheres a lot more to em, but i'll talk to my manager and see what i can do :D23:19
blackburnsonney2k: at least you have good cars :D23:19
@bettyboololomat23:20
serialhexyes indeed!!  german cars rock!!!23:20
serialhexanyway, i'm going to work on a ruby example tonight, and some more probably sunday a bit, tho it's still not compiling for me23:21
@sonney2kblackburn, I am riding a bike carrying to kids now23:21
@sonney2kin some kind of trailer23:21
serialhexHAHAHA!!!23:21
@sonney2kserialhex, yeah we need a test23:21
serialhexhaving fun sonney2k??23:21
@sonney2kserialhex, sure23:21
@sonney2kserialhex, some minimal ruby extension test...23:21
serialhexdo you want something that loads up all the modules?23:21
@sonney2kif you have that we could really check if things compile23:22
serialhexok, i'll see what i can cook up23:22
@sonney2kserialhex, ahh and btw, I tried lowercasing all module names but then the swig wrapped stuff stopped to work completely23:22
@sonney2kI couldn't import e.g. features nor Features23:22
serialhexhmm... thats odd...23:22
@sonney2kerr require23:22
@sonney2kserialhex, not sure if this is related to a pre 2.0 swig version here23:23
serialhexwell i couldnt do that before either23:23
serialhexidk...23:23
* serialhex should probably find his old copy of vmware and install it23:23
@sonney2kalesis-novik, I am not an expert in GMM - EM so ask vojtech what he thinks what is better23:24
serialhexi also can't link it with the -lruby1.8 thing, i'll gist my error msg now...23:24
@sonney2kserialhex, just edit .config an replace the -lruby1.8 with -lruby23:25
@sonney2k(occurs twice)23:25
alesis-noviksonney2k, I think this is more an aesthetic issue than anything else23:25
serialhexok... i'll do that for now then23:25
@sonney2kalesis-novik, I just don't know what EM for GMM will involve23:25
blackburnsonney2k: have you heard about some dim reduction for strings? just recalled that shogun is highly biased to it :)23:26
alesis-novikok, I'll ask Vojtech for his opinion23:26
@sonney2kalesis-novik, in the end vojtech and you decide what is reasonable - only if the code later looks really weird I will give comments :)23:27
@sonney2kblackburn, nope ... well if they are kernel based then I guess you can use kpca23:27
@sonney2kblackburn, and shogun is not *that* biased towards strings23:28
blackburnI just thinking about what to add to my project23:28
alesis-noviksonney2k, well, that's one point of writing this software: making it readable and manageable :)23:28
@sonney2kheh23:28
@sonney2kblackburn, whatever you like :)23:28
@sonney2kblackburn, btw I would love to do the CClassifier -> CMachine rename23:29
@sonney2kany objections?23:29
blackburnit takes no difference for me23:29
blackburnbut sounds better, yes23:30
blackburnmy worries is related to preproc 'framework'23:30
@sonney2kblackburn, in which sense?23:30
@sonney2kbtw, serialhex we have one module name PreProc23:31
serialhexok23:31
blackburnas I said earlier, mds and lle doesn't make sense at 'learning'23:31
@sonney2kI guess that should then at least be called just Preprocessors23:31
serialhexok, sounds good23:31
blackburnso init() in preproc and apply_to_feature_matrix() doesn't make sense at all23:31
blackburncalling init for one feature matrix and apply to another - makes no useful embedding23:32
blackburn*be back in 10 minutes23:32
@sonney2kblackburn, I don't see a problem23:32
@sonney2kblackburn, you just don't do anything in init()23:32
@sonney2kand always do the embedding in apply*23:32
@sonney2kto the current fm23:33
@sonney2kserialhex, ok will rename it then23:33
serialhexoh, sonney2k, how were you doing the %module renaming?  something like #ifdef SWIGRUBY; %module foo; #endif; ??23:33
@sonney2kserialhex, yes23:33
serialhexok, i'll test that out then sonney2k, and see how it works on my end23:33
@sonney2kserialhex, this is the patch https://gist.github.com/99725323:35
serialhexsweet!!!  thanks sonney2k, i'm compiling now but i'll get this applied once it's done (and recompile :P )23:36
@sonney2kserialhex, blackburn: shall I name the PreProc module Preprocessor or Preprocessors?23:37
@sonney2kI tend towards Preprocessor23:37
serialhexthat works for me23:37
blackburnsonney2k: hehe i don't understand why I didn't choose this way (with do nothing in init)23:39
@sonney2kbrain twists...23:39
blackburnsonney2k: do we really need 'pre'?23:39
blackburnPreprocessor is better for me23:40
@sonney2kblackburn, you mean Processor?23:41
@sonney2knah23:41
blackburnmay be23:41
blackburnbut it sounds not so 'sensefully', I agree now :)23:41
@sonney2kshould I also rename the preproc dir to preprocessor ?23:41
@sonney2kand the PreProc class to Preprocessor?23:42
blackburnmakes sense23:42
@sonney2kdoing ...23:42
blackburnPreProc is strange camel case23:42
blackburncaught an error a few times because of Preproc<->PreProc23:43
blackburnsonney2k: I think in preproc there should be a class for dim.reduction :)23:43
blackburnmay be I'll add one later23:43
blackburndoes it makes sense to you?23:44
blackburnanyway I'm going to bed now, tomorrow will be learning some shity game theory and so on, damn exams23:45
blackburnsee you23:45
-!- blackburn [~qdrgsm@188.168.4.77] has quit [Quit: Leaving.]23:45
@sonney2kbest of luck...23:46
--- Log closed Sun May 29 00:00:41 2011