Open in new window / Try shogun cloud
--- Log opened Mon Oct 19 00:00:09 2015
-!- PirosB3 [~pirosb3@] has joined #shogun00:46
-!- PirosB3 [~pirosb3@] has quit [Ping timeout: 260 seconds]03:03
-!- PirosB3 [] has joined #shogun03:12
-!- shaochuan [] has quit [Read error: Connection reset by peer]03:15
-!- shaochuan [] has joined #shogun03:15
-!- PirosB3 [] has quit [Quit: PirosB3]04:33
-!- PirosB3 [] has joined #shogun05:53
-!- PirosB3 [] has quit [Quit: PirosB3]06:34
-!- PirosB3 [] has joined #shogun06:35
@wikinglisitsyn: hey here?07:14
-!- PirosB3 [] has quit [Quit: PirosB3]11:07
-!- sonne|work [] has joined #shogun13:30
-!- shaochuan [] has quit [Read error: Connection reset by peer]14:52
-!- shaochuan [~shaochuan@2601:647:4603:2a70:35a4:52ce:a5bd:a690] has joined #shogun14:53
-!- shaochuan [~shaochuan@2601:647:4603:2a70:35a4:52ce:a5bd:a690] has quit [Remote host closed the connection]18:52
-!- shaochuan [] has joined #shogun18:52
-!- shaochuan [] has quit [Ping timeout: 250 seconds]18:57
-!- sonney2k_ is now known as sonney2k20:48
-!- mode/#shogun [+o sonney2k] by ChanServ20:48
@sonney2klisitsyn, was wiking ever awake when you were?20:54
lisitsynsonney2k: yeah we catched up a few times20:55
@sonney2klisitsyn, around when?20:55
lisitsynsonney2k: closer to 'our' morning actually20:55
* sonney2k needs chocolate20:55
@sonney2knothing in the whole #&%*YfhDF house20:55
@sonney2kI guess I could eat 1kg of it20:56
@sonney2kjsut now20:56
lisitsynsonney2k: is he in singapore now? I have no idea to be honest20:56
lisitsynI guess so..20:56
lisitsynI remember we chatting around 1-3 pm CET20:57
lisitsynso I guess if you want to catch up you gotta try tomorrow heh20:57
@sonney2klisitsyn, will try20:58
@sonney2kbut I don't actually recall how to write c++ code ;)20:58
lisitsynsonney2k: is it about license switch?20:58
lisitsynsonney2k: ask me I do that for living haha20:59
lisitsynwhy C++?20:59
@sonney2khe wanted it badly but IMHO didn't really spent time pushing it20:59
lisitsynwant to write some code?20:59
@sonney2kwell C/C++ all the same to me20:59
@sonney2kshogun wise you mean20:59
@sonney2knah I am so java polluted now21:00
lisitsynyeah I mean why bother about C++21:00
@sonney2kI leak memory and write code w/o thinking21:00
@sonney2kwell shogun!~21:00
@sonney2kor what do you program21:00
@sonney2kand which $LANG?21:00
lisitsynwell you know shogun is really slow21:01
lisitsynif you're asking about me I am doing C++ and python21:01
@sonney2klisitsyn, shogun & slow you mean development currently?21:02
@sonney2kor what?21:02
lisitsynsonney2k: yes basically no real development21:03
lisitsynwell it is actually good if we were satisfied21:03
@sonney2klisitsyn, I am cleaning my fish tank ATM so I am not quick21:03
lisitsynbut everyone has ideas and we don't move21:03
@sonney2klisitsyn, heh well I thought heiko gave up?21:03
lisitsynthey sum to zero21:03
@sonney2kwell I guess no one has time right?21:04
@sonney2kor everyone wants to do something different :)21:04
@sonney2klisitsyn, so you use scikits now or what?21:05
@sonney2klisitsyn, and what would you improve if you had the time?21:05
lisitsynsonney2k: sorry, back21:06
lisitsynsonney2k: I wouldn't say its about time21:06
lisitsynsome kind of passion I'd say21:06
@sonney2kyeah if you don't use it why do it...21:06
@sonney2klisitsyn, what do you use now?21:06
lisitsynI am doing some neural nets now (fuuu)21:06
lisitsynwith in-house stuff21:06
lisitsynbig company has their own tools21:07
lisitsynbut I used shogun a few times21:07
lisitsynwe need to improve how we handle data and modularize21:07
@sonney2klisitsyn, handle data as in?21:07
lisitsynwell lacks some magic21:07
lisitsynhave you used vw?21:08
@sonney2klisitsyn, never21:08
lisitsynit has strict format but you can do basically anything21:08
lisitsynthere is no such thing as dense features or sparse features in real life21:08
lisitsynthey are always mixed together21:08
@sonney2kwith the size of data I had to deal with I always had to write the basic i/o from scratch...21:08
@sonney2kwell I manually de-mixed the features back in the days21:09
@sonney2kbut regarding modularize21:09
lisitsynbut things speed up21:09
lisitsynand you can't spend your time doing that21:09
@sonney2kI see that this could in principle be done for libshogun / c++21:09
@sonney2kbut I don't see it for *any* interface21:09
@sonney2kat least not with swig21:10
lisitsynhave you seen my mail?21:10
lisitsynI had some solution21:10
@sonney2knah I don't read email21:10
lisitsynpretty ugly..21:10
lisitsynquite similar to static21:10
@sonney2kI am only scanning things...21:10
lisitsynwell if you have parameter just set it with set("parameter", 3);21:10
lisitsynlike that21:10
lisitsynor other thing, tags21:11
lisitsynlike you have tags for parameters21:11
@sonney2kyeah sounds like static21:11
lisitsyntag = kernel.parameter("width");21:11
lisitsynkernel.set(tag, 3.0);21:11
lisitsynthis would work in swig21:11
@sonney2kahh you are talking about this still being wrapped in swig21:12
@sonney2kI see21:12
lisitsynw/o swig doesn't make sense21:12
@sonney2kbut still you would need to expose the whole object .h tree to swig21:12
@sonney2kand you cannot modularize that21:12
lisitsynno, you can avoid that21:12
lisitsynyou can expose only basic objects21:12
lisitsynclassifier, kernel, etc21:12
lisitsynand call kernel("gaussian")21:12
@sonney2kyou would need to feed any object into another21:12
lisitsynwell that's the limitation, you would have to use only base classes21:13
lisitsynno crazy relationship between two really different modules21:13
@sonney2kcould you please make an example for kernel and features21:13
@sonney2kso I know how it could work?21:14
lisitsynwell, features are base classes21:14
lisitsynfeatures = shogun.DenseFeatures would stay I guess21:14
lisitsynas for kernel you just do21:14
lisitsynkernel = shogun.kernel("gaussian")21:14
lisitsynkernel.init(features, features)21:15
lisitsynkernel.set("width", 10.0)21:15
lisitsynsonney2k: most checks would go to runtime then21:15
lisitsynbut I don't see any need to expose .h of plugins this way21:15
lisitsynclassifier = shogun.classifier("liblinear")21:16
@sonney2kwell ok features make up 50% of all swig wrapped code21:18
lisitsynsonney2k: I would actually drop features..21:19
lisitsynand write adapters to use pandas dataframe21:19
lisitsynspark rdd21:19
lisitsynand stuff like that21:19
@sonney2klisitsyn, so you would have a classifier21:20
@sonney2ksure you can init is just by some 'ascii' name21:20
@sonney2kso you would just have a stand alone classifier module21:20
lisitsynyeah it would be easier to write/load plugins then21:21
lisitsynsonney2k: I am also obsessed by other thing21:21
@sonney2kbut how do classifiers underneath work with features w/o knowing about e.g. CDenseFeatures?21:21
lisitsynit knows about dense features21:21
lisitsynthey would have to be in the base shogun21:21
lisitsynit doesn't know about gaussian kernel21:21
@sonney2klisitsyn, OK21:21
lisitsynor oligowdstringkernelwhatever21:21
lisitsynsonney2k: I am obsessed about having shogun.classifier('some_matlab_classifier_from_fancy_researcher')21:22
@sonney2kbut then kernelmachines would need to know about kernels right?21:22
lisitsynyes about base class, CKernel21:22
@sonney2kok makes sense21:22
@sonney2kso basically you modularize out the machine part right?21:22
lisitsynwe would just move all the concrete kernels21:22
lisitsynconcrete machines21:23
lisitsynand stuff like that21:23
lisitsynthis would be a step21:23
@sonney2kthis sounds very doable21:23
lisitsynyeah probably21:23
lisitsynbut this string-based thing makes me cry and I see no other way21:23
lisitsynwhen it comes to something dynamic all you have is some string21:23
@sonney2klisitsyn, yeah even enums would make it non-dynamic21:24
lisitsynsonney2k: and we should employ your COFFIN much more21:26
lisitsynsonney2k: recently I realized nobody use that approach for neural nets21:26
lisitsynalthough it is quite natural21:26
@sonney2klisitsyn, so if I get it right you would have a dynamic runtime open for the classifier21:26
@sonney2kand this way you can just have ton's of .so's hanging around21:27
@sonney2kand nice and modular and you only need the libshogun core installed21:27
@sonney2klisitsyn, and if you want to use more of shogun's helper stuff linalg etc?21:28
lisitsynnot sure21:28
lisitsynsonney2k: I am not sure anybody needs it being exposed via swig21:28
lisitsynall major langs have their own linalg21:28
@sonney2klisitsyn, well not swig but from the c++ side21:29
@sonney2kyou would still need to link against that part21:29
@sonney2klisitsyn, and yes about COFFIN... if I use stuff I only do it linearly and the coffin way21:30
lisitsynsonney2k: it is a very good idea, essentially employing principle of limiting the interface21:31
lisitsynlike you don't need all the stuff so don't ask for that broad interface21:31
@sonney2klisitsyn, thanks ;)21:34
@sonney2klisitsyn, the one thing that feels hard with c++ to me is I have to use my brain to not leak21:35
@sonney2klisitsyn, I really cannot do that (by default)21:35
lisitsynsonney2k: ah yes that's another thing to change21:36
lisitsynI'd like to not see any raw pointer in shogun21:36
lisitsynsonney2k: you know it is quite natural for modern C++ to use unique and shared pointer semantics21:38
lisitsyn*sometimes* weak21:38
lisitsynbut that's all to not think a lot21:38
@sonney2klisitsyn, alright and how do we get it in then?22:11
@sonney2klisitsyn, this is really a blocker to me. the modularization part sounds rather easy to do and could even be done in a compatible way in *addition* to what we have now until it is funcitonal22:12
@sonney2kwiking, hey - lets chew bubble gum and kick ass...22:16
@sonney2kwiking, well I guess I will have to chew it for you :P22:16
@sonney2kwiking, so where the heck is the license thing?!22:16
lisitsynsonney2k: that's true22:23
lisitsynwe can start right away heh22:23
lisitsynno idea what blocks me22:23
@wikingwith one of the website deploys22:25
@wikingthat thing got wiped out22:25
@wikingand as much as i wanted couldnt find any copy of it anywhere22:25
@wikingand i have no idea what i put on that page :D22:25
@wikingi mean text wise22:25
@sonney2kwiking, and I guess no idea who signed it either?22:26
@sonney2kwiking, man!22:26
@sonney2kwiking, doesn't sound like you pushed very hard didn't you?22:26
@wikingthat i know22:26
lisitsynlets do that once again lool22:26
@wikinghahaha you think of me too little ;)22:26
@wikingthat's in a db table22:27
@wikingi just need the stupid static page22:27
lisitsynso you have a list of signers22:27
lisitsynbut no text22:27
@wikingwith one webform in it22:27
lisitsynthat's funny22:27
@wikingi just need to recover the webform basically22:27
@wikingthat'll send some POST params to a php script22:28
@wikingthat'll update this table22:28
@wikingso stop being cocky and help me with the content of the webform :)22:28
@sonney2kwiking, didn't you send out an emal to everyone?22:31
@sonney2kI mean that content should be in everyones inbox22:32
@wikingyeah partly in some way yes22:32
@sonney2kwiking, do you remember the sender22:34
lisitsynDear root,22:35
lisitsynWe are contacting you as You are one of the developers who has been committing to Shogun machine learning toolbox in the past.22:35
lisitsynlike that?22:36
lisitsynwiking: we your inbox22:37
@sonney2klisitsyn, subject?22:44
lisitsynShogun Licensing22:44
@sonney2klisitsyn, found ...22:45
@sonney2kwiking, was the webform on's servers?22:46
@sonney2kbecause the url I got was22:46
@sonney2kwiking, well we have (or had) backups22:50
@sonney2kwiking, but probably not dating back 2 years22:50
* sonney2k Zzzzz22:52
--- Log closed Tue Oct 20 00:00:10 2015