Open in new window / Try shogun cloud
--- Log opened Sat Oct 29 00:00:10 2011
CIA-3shogun: Soeren Sonnenburg master * re9c8ecd / src/interfaces/lua_modular/swig_typemaps.i : fix an error in the lua typemaps (checking for the typecode without using strcmp) -
-!- BrianC_ [43dc92a2@gateway/web/freenode/ip.] has quit [Ping timeout: 265 seconds]00:39
-!- blackburn [~blackburn@] has joined #shogun12:32
blackburnsonney2k: wow you did a lot of fixes12:34
-!- blackburn [~blackburn@] has quit [Read error: Operation timed out]18:32
-!- blackburn [~blackburn@] has joined #shogun21:46
-!- blackburn1 [~blackburn@] has joined #shogun22:03
-!- blackburn [~blackburn@] has quit [Ping timeout: 258 seconds]22:04
supernaughtsonney2k: When you say "use the signum of these numbers, e.g. if the sign is negative then it is a letter from group A, if positve then from group B"22:05
supernaughthow do I feed in a number for which training has not been done?22:06
supernaughte.g. In the program, one creates points {0,1,2} with labels {-1,1,-1}22:06
supernaughtbut when I attempt to extrapolate from my inputs and determine which group {3} belongs to, I pull an exception "terminate called after throwing an instance of 'shogun::ShogunException'"22:11
blackburn1supernaught: are you doing multi-class classification?22:11
supernaughtI believe not22:12
blackburn1how many letter groups you have?22:12
supernaughtI'm working with third one from top at
supernaughtI'm not sure what a letter group is22:12
blackburn1well you said group A, group B, ..?22:13
supernaughtoh, sure22:13
blackburn1and there are only A and B, right?22:13
supernaught+1, -122:13
blackburn1and you set up labels with +1 and -122:13
blackburn1and got exception?22:14
supernaughtthe example is nearly verbatim what I have22:14
supernaughtexcept when I go to "classify over training variables"22:14
supernaughtI loop from 0 to 4 instead of 0 to 322:15
blackburn1can you send the example reproducing the error to me?22:15
blackburn1I would check it and fix then22:15
blackburn1or say what is wrong22:15
supernaughtlet me paste it somewhere22:15
blackburn1just send it by email22:16
blackburn1to blackburn91@gmail.com22:16
-!- blackburn1 is now known as blackburn22:16
blackburnsupernaught: btw what are the features you do use?22:17
blackburnah sure gist works too22:17
supernaughtI'm a complete newb to machine learning22:18
supernaughtI'm not sure what you mean by "features" ?22:18
blackburnhmm for example simple features is just dense vectors22:18
supernaughtwhoops, I forgot the error22:19
blackburnand feature is 1 and 2 and 3 and 422:19
blackburnthen [1,2,3,4] is a feature vector22:19
blackburnand matrix containing feature vector is a feature matrix22:19
blackburnbut sometimes feature vector is just a string ("some string")22:20
blackburnand there is no feature matrix22:20
supernaughtOk: so a feature vector is an input which is used for training?22:20
blackburnyes, it is vector representing some object22:21
supernaughtok. Lingo :)22:21
blackburnsomething like that :)22:21
supernaughtso in this case, I am using a feature vector of [0,1,2,3,4,5]22:21
blackburnso do you understand what are train set and test set?22:21
blackburn [1,3,5]]22:22
blackburnit is a feature matrix22:22
supernaughtoh, right22:22
blackburn[0,1] [2,3] [4,5] is a feature vectors22:22
supernaughtand the fortran-style indexing22:22
blackburnyeah old-school hah22:22
blackburnokay I got it22:22
blackburnabout train features and test features22:23
blackburnyou are trying to classify 4th object/vector22:23
blackburnthat are not here anyhow22:23
supernaughtperhaps I am misunderstanding a training and a test22:23
supernaughtyou say22:23
supernaughtthis makes sense22:24
supernaughthow do I input some test data then ?22:24
blackburnsupernaught: you could use set_features22:25
blackburnor apply to whole features22:25
supernaughtI will check out the documentation now22:25
blackburnwell we have pretty bad doc22:25
supernaughtjust so I am sure,22:25
blackburnwill return CLabels instance22:26
blackburnwhen you can get classified labels22:26
blackburnsupernaught: btw why are you using C++ for doing that?22:27
supernaughtso what is happening is I will get:     <interpretted/classified data>   =  svm->apply(  <test data>  );22:27
supernaughtI am tinkering with embedding shogun in a set of medical physics libraries which are written in c++22:27
blackburnI see22:28
blackburnapply(features) returns CLabels22:28
blackburnwhich are containing labels (-1,+1,..)22:28
supernaughtAwesome. Thanks for the clarification. I'll check out the documentation and come back when I get stuck again :)22:28
blackburni.e. the results of classification22:28
blackburnyou could just ask here :)22:29
supernaughtunfortunately I never seem to learn anything unless I struggle through it a little22:29
supernaughtmaybe it is 'starving artist' syndrome22:29
supernaught: /22:29
blackburneveryone do the same I guess22:29
blackburngood thing is you understand my broken english :D22:30
supernaughtnot a problem22:30
supernaughtthanks again - this is some unbelieveably cool software22:31
supernaughtI'm grateful, thanks22:31
blackburnsupernaught: in the example I sent to you22:31
blackburnthere are features and features_test22:31
blackburnah I did it wrong22:31
blackburnokay when there are features and features_test22:31
blackburnand if you do svm->train with features22:32
blackburnand then svm->apply with features_test22:32
blackburnyou will get classification results22:32
blackburnsupernaught: it is nice to hear our software is any cool :)22:33
supernaughtI understand22:33
supernaughtone more question, though22:34
supernaughthow stable is the c++ api ?22:34
blackburnsupernaught: in what means?22:34
supernaughtin a few years, will I be able to compile my code?22:35
supernaughte.g. shogun is 1.0.0 right now. How much will things change at 8.0.0 ?22:35
blackburnoh, I hope so22:35
blackburnare you doing eternal software? :D22:35
supernaughtin the medical world, everything is supposed to be 'eternal' and 'error-free'22:36
blackburnyeah sure22:36
blackburnI'm not sure about every method but I think apply/train/CLabels/etc things won't be changed22:36
supernaughtWell, I'm off for now. I'll catch you later blackburn22:37
blackburnsupernaught: okay see you22:37
blackburnsupernaught: I'm really interested with attaching shogun to any other software (as Soeren does too) - so don't hesitate to ask22:38
CIA-3shogun: Sergey Lisitsyn master * r5ced67a / (src/shogun/base/SGObject.cpp src/shogun/base/SGObject.h): Fixed spacings at SGObject -
supernaughtblackburn: the application will be a compatibility layer between raw medical data and a uniform-surface application layer23:47
blackburnsupernaught: I see, I hope that will work23:47
supernaughtit is a for a master's thesis, but the (public) organization I work for covers a large part of Canada23:48
supernaughtand they are eager to adopt it23:48
supernaughtso I would _love_ to put shogun to work23:48
blackburnsupernaught: what are the tasks that software specialize with?23:54
blackburnsupernaught: anyway as I said before don't hesitate to ask me or Soeren and if there is no one of us here - ask on the mailing list or by email23:59
--- Log closed Sun Oct 30 00:00:17 2011