--- Log opened Mon Apr 11 00:00:19 2011 | ||
--- Day changed Mon Apr 11 2011 | ||
@sonney2k | serialhex, you should at least test this - e.g. sample 100k times and then plot a histogram | 00:00 |
---|---|---|
@sonney2k | if it looks gaussian well then it is good | 00:00 |
serialhex | sonney2k: ok... sounds good | 00:00 |
* serialhex looks into how to do that :-/ | 00:01 | |
blackburn | I think there is no problem with it, it will produce a gaussian random value | 00:03 |
@sonney2k | serialhex, easy: generate the numbers and then use octave or python + matplotlib and do hist(numbers) | 00:10 |
@sonney2k | potentially with 100 bins | 00:10 |
serialhex | sonney2k: ok, sounds good | 00:10 |
serialhex | is there anyway to just make the Math.h file? or do i have to re- ./configure; make; make install ??? | 00:14 |
blackburn | why do get_feature_vector(..) returns matrix?! oh my | 00:15 |
-!- Mengyun [~dorothyk@S0106001a803d4363.vc.shawcable.net] has quit [Quit: Leaving] | 00:17 | |
@sonney2k | blackburn, it shouldn't | 00:24 |
blackburn | sonney2k: it seems that doc isn't very good | 00:25 |
blackburn | /** get a copy of the feature matrix (in feature space) | 00:25 |
blackburn | * num_feat,num_vectors are returned by reference | 00:25 |
@sonney2k | serialhex, almost all modules use math - so you have to redo them all | 00:25 |
@sonney2k | for get_feature_matrix right? | 00:25 |
blackburn | iit is for virtual void get_feature_vector(float64_t** dst, int32_t* len, int32_t num); | 00:25 |
serialhex | ok... i was compiling them all anyway and DistanceKernel.h gave me an error: | 00:26 |
serialhex | In file included from base/class_list.cpp:23: | 00:26 |
serialhex | ./kernel/DistanceKernel.h:115: error: ‘virtual void shogun::CDistanceKernel::set_width(float64_t)’ cannot be overloaded | 00:26 |
serialhex | ./kernel/DistanceKernel.h:96: error: with ‘virtual void shogun::CDistanceKernel::set_width(float64_t)’ | 00:26 |
serialhex | ./kernel/DistanceKernel.h:124: error: ‘virtual float64_t shogun::CDistanceKernel::get_width() const’ cannot be overloaded | 00:26 |
serialhex | ./kernel/DistanceKernel.h:105: error: with ‘virtual float64_t shogun::CDistanceKernel::get_width() const’ | 00:26 |
serialhex | In file included from ./kernel/StringKernel.h:15, | 00:26 |
serialhex | from ./kernel/MatchWordStringKernel.h:15, | 00:26 |
serialhex | from base/class_list.cpp:24: | 00:26 |
blackburn | sonney2k: oh, I see now, doc for both get_feature_matrix() and get_feature_vector() is identical | 00:26 |
blackburn | it is in DotFeatures.h | 00:26 |
serialhex | ...i'm taking a look now, it might be from someone elses commit cause i've picked up other ppl's commits somehow :-/ | 00:27 |
@sonney2k | blackburn, copy&paste bug | 00:27 |
@sonney2k | serialhex, weird. compiles here | 00:28 |
* sonney2k does a make distclean | 00:28 | |
blackburn | I haven't any problems with upstream master | 00:28 |
serialhex | i see what happened (i think) | 00:28 |
serialhex | the same functions: get & set width, were being created twice... :-/ | 00:29 |
blackburn | they not | 00:29 |
blackburn | https://github.com/SergeyLisitsyn/shogun/blob/master/src/libshogun/kernel/DistanceKernel.h | 00:30 |
serialhex | idfk how that happened... | 00:30 |
dvevre | i think theres a problem with serialhex's copy only | 00:30 |
serialhex | yeah, probably | 00:30 |
blackburn | serialhex: do 'git reset HEAD libshogun/kernel/DistanceKernel.h | 00:30 |
blackburn | serialhex: sudo do .. :) | 00:31 |
@sonney2k | alright - off to bed - cu tomorrow | 00:33 |
serialhex | ...i think i might have to figure out whats going on... i seem to get a BUNCH of other peoples commits into my fork somehow... i'm nto sure what i'm doing wrong | 00:33 |
dvevre | sonney2k: see ya | 00:33 |
serialhex | alright, nite... hopefully i'll have that histogram for ya tomorrow! | 00:33 |
@bettyboo | bye | 00:33 |
blackburn | serialhex: git remote add upstream https://github.com/shogun-toolbox/shogun.git | 00:34 |
serialhex | i already have that... and that's the only thing i really pull from | 00:35 |
blackburn | git fetch upstream | 00:35 |
blackburn | git merge upstream/master | 00:35 |
blackburn | git push | 00:36 |
blackburn | git make money | 00:36 |
blackburn | git get car | 00:36 |
dvevre | we wish! | 00:36 |
serialhex | :P i wish for that last one!! | 00:36 |
@bettyboo | hihi!?! serialhex | 00:36 |
serialhex | hihi! bettyboo!!! | 00:36 |
@bettyboo | serialhex: and in turn this is like a mentor's duty to his institution type thing | 00:37 |
@bettyboo | serialhex, yeah | 00:37 |
blackburn | serialhex: unfortunately you have to do only first 3 | 00:37 |
blackburn | git get car is unimplemented yet :( | 00:37 |
serialhex | i think bettyboo loves me :P | 00:38 |
@bettyboo | serialhex: i was thinking of doing that anyways, but got us involved in a totally different direction :P | 00:38 |
@bettyboo | he | 00:38 |
serialhex | yeah, and git make money | 00:38 |
blackburn | now will 'git go sleep' for a while :) | 00:40 |
serialhex | nite blackburn | 00:44 |
blackburn | see you | 00:44 |
dvevre | g.nite | 00:45 |
-!- blackburn [~qdrgsm@188.122.237.148] has quit [Quit: Leaving.] | 00:45 | |
-!- Gagan [~brocolli@nat-fw.cs.ualberta.ca] has joined #shogun | 00:55 | |
-!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 03:22 | |
-!- siddharth [~siddharth@117.211.88.150] has joined #shogun | 05:29 | |
-!- Stephen [dbea5187@gateway/web/freenode/ip.219.234.81.135] has joined #shogun | 06:22 | |
-!- Stephen is now known as Guest98570 | 06:23 | |
Guest98570 | I wonder can I still participate in the projects posed on the GSoC after the deadline, which I mean not be parts of the GSoC program. | 06:25 |
-!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has joined #shogun | 06:58 | |
warpy | anyone here ? | 06:59 |
serialhex | not really... | 07:04 |
serialhex | whats up? | 07:05 |
warpy | was going over shogun just now, need to consult with someone | 07:06 |
warpy | i need to figure out what needs to be done in order to add a wrapper /interface for c# | 07:07 |
serialhex | afaik... swig | 07:07 |
serialhex | swig is what they use for all of their modular interfaces, and it works with c# | 07:08 |
warpy | ouch what is swig | 07:08 |
serialhex | you can check out the code by looking at the swig_typemaps.i files in the src/*_modular dirs | 07:09 |
serialhex | swig.org it's pretty simple actually, just grab the headers, write some specialized C and you're done (for the most part, then comes the debugging :D ) | 07:10 |
@bettyboo | ;> | 07:10 |
serialhex | hey bettyboo, how are you? | 07:10 |
@bettyboo | serialhex: hey, how are you? | 07:10 |
serialhex | i'm good thank you, and yourself bettyboo?? | 07:10 |
@bettyboo | serialhex: why are you repeating yourself over and over? | 07:10 |
serialhex | because i like to bettyboo! | 07:11 |
@bettyboo | serialhex: aint one hir? | 07:11 |
serialhex | no one is here bettyboo | 07:11 |
@bettyboo | serialhex: it's because I'm novice in git | 07:11 |
warpy | make the bot shutup :) | 07:11 |
serialhex | ok... i'll stop talking to her | 07:11 |
warpy | should i use a specific swig download | 07:14 |
warpy | or just grab the latest ? | 07:14 |
Guest98570 | anyone can answer my question just now? | 07:16 |
Guest98570 | I wonder can I still participate in the projects posed on the GSoC after the deadline, which I mean not be parts of the GSoC program. | 07:17 |
warpy | doesnt sound like it | 07:18 |
serialhex | the latest warpy | 07:19 |
serialhex | ...i think if you simply want to contribute code you can, though you wont be getting anything except good vibes from the shogun guys if you do that. so it's not a total loss :D | 07:20 |
@bettyboo | ;> serialhex | 07:20 |
Guest98570 | I mean can I get help from mentors as the students joint in the GSOC program? | 07:31 |
Guest98570 | ............nobody here? | 07:34 |
-!- Guest98570 [dbea5187@gateway/web/freenode/ip.219.234.81.135] has quit [Ping timeout: 252 seconds] | 07:47 | |
-!- Stephen [dbea5187@gateway/web/freenode/ip.219.234.81.135] has joined #shogun | 08:44 | |
-!- Stephen is now known as Guest30851 | 08:44 | |
-!- Guest30851 [dbea5187@gateway/web/freenode/ip.219.234.81.135] has left #shogun [] | 08:47 | |
-!- yayo3 [~jakub@ip-78-45-113-245.net.upcbroadband.cz] has joined #shogun | 09:00 | |
-!- sploving [~root@124.16.139.196] has joined #shogun | 09:02 | |
sploving | hello sonney2k, I tested it. and it worked well! | 09:03 |
sploving | I want to give a patch. I use "git add swig_typemaps.i", "git commit", and tried to use "git push" but failed | 09:06 |
sploving | I think I do not have the permmission to git push. | 09:07 |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 260 seconds] | 09:08 | |
@sonney2k | sploving, you need to fork shogun on github | 09:12 |
@sonney2k | then commit to your forked project | 09:12 |
@sonney2k | and then send a git push request | 09:12 |
@sonney2k | sploving, I sent an HOWTO to the ml | 09:13 |
@sonney2k | sploving, I forgot *great job*!!! | 09:13 |
sploving | I saw your email | 09:13 |
sploving | i clone it before | 09:13 |
sploving | if i use github, i need clone it again | 09:14 |
sploving | which waste a lot of time | 09:14 |
sploving | i mean first i need fork it, then clone | 09:14 |
warpy | hey sonney2k, i understand that i need to download the swig library. i am looking at lib/csharp/ | 09:15 |
@sonney2k | sploving, yes but where is the problem? | 09:15 |
@bettyboo | hi | 09:15 |
@sonney2k | warpy, for c# (like any other lang) you need to write typemaps (file swig_typemaps.i) and point me to an example of how to get c# to compile - then I add this to configure | 09:16 |
warpy | sonney2k, what i dont get if i should just use typemaps.i or actually gather some more code from the other .i files in there | 09:16 |
warpy | i need to know where to start, which is what i am asking | 09:17 |
warpy | so many files, just need to figure out what to do at the beginning | 09:17 |
sploving | sonney2k, it shows the following error: Permission denied . fatal: the remote end hung up unexpectedly | 09:17 |
@sonney2k | warpy, if you intend to work on c# I can provide you with a ./configure --interfaces=csharp | 09:18 |
sploving | i use git clone git@github:com:sploving/shogun.git | 09:18 |
@sonney2k | warpy, then all you need to do is write the typemap | 09:18 |
@sonney2k | sploving, did you upload your ssh pub key? | 09:18 |
warpy | where do you use ./configure --interface=csharp ? | 09:19 |
sploving | not yet. I just set github token | 09:19 |
sploving | and use git config --global github.token ... | 09:19 |
sploving | warpy, ./configure --interfaces=csharp, is used when configure our shogun | 09:20 |
warpy | gotcha, okay so i am a litttle newbish to swig so i am hoping to understand a few things | 09:22 |
@sonney2k | warpy, well I would add this feature if you intend to do the wrappers | 09:24 |
@sonney2k | but I need some pointer on how to do csharp extension | 09:25 |
@sonney2k | s | 09:25 |
warpy | one of my students will do that in the next few weeks | 09:25 |
sploving | sonney2k, as my patch is a new file, not to change other file. could you test and update it for me if the patch works well | 09:26 |
@sonney2k | warpy, well then I can prepare the infrastructure in the same way as I did for sploving for lua/ruby/java | 09:27 |
warpy | okay. | 09:27 |
@sonney2k | sploving, just upload your ssh public key | 09:28 |
warpy | i still need a few more clues to where to start. | 09:28 |
warpy | i have looked in lib/csharp/ | 09:28 |
warpy | should i take one of those files as a template ? | 09:28 |
@sonney2k | sploving, here is a troubleshooting guide http://help.github.com/troubleshooting-ssh/ | 09:28 |
@sonney2k | warpy, all you have to do is write a file called swig_typemaps.i in (the upcoming) csharp directory of shogun | 09:29 |
@sonney2k | these typemaps are needed to translate e.g. matrices to doulbe* x, int d1, int d2 | 09:29 |
@sonney2k | to become accessible from C/C++ | 09:30 |
@sonney2k | that is all | 09:30 |
warpy | just matrices to double* ? | 09:30 |
warpy | but i looked at the python_modular/swig_typemaps.i and it seems like there is a whole world of things going on in there | 09:30 |
warpy | so to make things short, if there is a list of things to do inorder to make a complete type map that would be a help. | 09:38 |
sploving | warpy,python_modular/swig_typemaps.i mainly deal with matrix | 09:39 |
sploving | like other modular | 09:40 |
@sonney2k | warpy, vectors, matrices, sparse matrices, strings | 09:41 |
warpy | of any type i assume | 09:42 |
warpy | is there a small doc/tutorial about this that relates to shogun. something that i can read | 09:43 |
-!- blackburn [~qdrgsm@188.122.237.148] has joined #shogun | 09:43 | |
@sonney2k | sploving, did you finally manage to get access to github? | 09:44 |
blackburn | hi there | 09:44 |
sploving | I am working it | 09:44 |
blackburn | yew! 18 forks | 09:48 |
sploving | hello blckburn, what does that mean? | 09:48 |
@bettyboo | hiho | 09:48 |
sploving | blackburn, does that mean 18 students clone the project? | 09:49 |
blackburn | sploving: I guess it is ~12 students' forks | 09:49 |
warpy | sonney2k, so basically i need to do vec/mat/sparse/strings and for each make sure there is functionality for byte/char/int/float/double etc ? | 09:49 |
yayo3 | so there are already more people with forks than can be accepted? | 09:51 |
blackburn | oh, I think I really get wrong | 09:52 |
blackburn | all forks exclude ameerkat is students' | 09:52 |
blackburn | so seems that 17 students forked so far | 09:52 |
blackburn | yayo3: of course, <=10 will be accepted | 09:52 |
@sonney2k | warpy, yes | 09:53 |
@sonney2k | warpy, of course if sth is not supported by the language then just some features in shogun are unavailble | 09:53 |
yayo3 | now ain't that fun | 09:53 |
warpy | sth ? | 09:54 |
blackburn | sth=something | 09:54 |
warpy | oh right | 09:54 |
sploving | warpy, for example, if a lange A does not support char, then you donot need wrap it when binding to A | 09:55 |
-!- skydiver [4deac315@gateway/web/freenode/ip.77.234.195.21] has joined #shogun | 09:55 | |
sploving | when I clone it is so slow | 09:56 |
warpy | okay but how do i know if shogun has uspport for all of these variable types for each data type | 09:56 |
sploving | of course they do | 09:56 |
sploving | as it is written in C++ | 09:57 |
warpy | i see, so from what i understand swig is connected to that part of shogun that does all the data manipulation. can you guys tell me which files has those, just to make sure that i have all the reference i need. | 09:59 |
sploving | all the .i files | 10:00 |
warpy | directory ? | 10:01 |
sploving | some base .i file which is generic to all module is in src/modular | 10:01 |
sploving | the specific .i file is in every modular | 10:02 |
-!- yayo3 [~jakub@ip-78-45-113-245.net.upcbroadband.cz] has left #shogun [] | 10:02 | |
-!- sploving [~root@124.16.139.196] has left #shogun [] | 10:03 | |
blackburn | sonney2k: one performance issue: is get_feature_vector a 'slow' operation and should I work with feature matrix? | 10:04 |
warpy | how come matlab doesnt have a .i file ? | 10:06 |
@sonney2k | blackburn, it is as fast as accessign feature matrix if it is there | 10:13 |
blackburn | sonney2k: I mean I have to get each feature vector, is it bad to do it N times? | 10:14 |
@sonney2k | warpy, matlab is closed source - so has no swig support | 10:14 |
@sonney2k | blackburn, not at all | 10:14 |
@sonney2k | just transverse the vecotrs in order | 10:14 |
blackburn | what do you mean? | 10:15 |
@sonney2k | warpy, all in modular/ .... all classes are wrapped btw | 10:15 |
@sonney2k | blackburn, calling get_feature_vector is as cheap as getting a ptr | 10:15 |
@sonney2k | (if the matrix is there) | 10:15 |
blackburn | sonney2k: yeap it is CDotFeatures | 10:16 |
@sonney2k | (there can be computes like preproc on top of it though) | 10:16 |
blackburn | sonney2k: in that case, will get_feature_vector return initial vector with no preproc? | 10:17 |
blackburn | sonney2k: shortly, will I have any problems with it? :) | 10:18 |
@sonney2k | no always the preproc'd one | 10:18 |
@sonney2k | just use it :) | 10:18 |
@bettyboo | rahaha!?! | 10:18 |
blackburn | okay | 10:18 |
blackburn | sonney2k: and last one question, may I ask you? | 10:19 |
@sonney2k | dont ask to ask - ask! | 10:19 |
blackburn | sonney2k: so, in CKNN there are min_class and max_class | 10:19 |
blackburn | for example, if labels are -9 -4 4 9 | 10:19 |
blackburn | min_class will be -9, max_class will be 9 | 10:20 |
blackburn | and num_classes 4 | 10:20 |
CIA-31 | shogun: Siddharth Kherada master * r9c534b9 / (15 files in 3 dirs): Merge commit 'upstream/master' - http://bit.ly/h6R8LM | 10:20 |
CIA-31 | shogun: Soeren Sonnenburg master * r412d643 / src/libshogun/kernel/DotKernel.h : whitespace fix - http://bit.ly/e7NBYv | 10:20 |
@sonney2k | well we assume mc labels to start with 0... | 10:20 |
@sonney2k | for binary case we assume -1/+1 | 10:21 |
blackburn | ah | 10:21 |
blackburn | and can I sure it will not be 0 9 18 27 ...? | 10:21 |
skydiver | sonney2k: do i correctly understand that preproc now can't do dimension reduction? | 10:22 |
@sonney2k | skydiver, they can but they cannot change feature type | 10:22 |
@sonney2k | skydiver, look at e.g. PCACut | 10:23 |
skydiver | sonney2k: ok ,thanks | 10:23 |
* blackburn gets tangled with that labels numbering :D | 10:24 | |
blackburn | sonney2k: look, in CKNN we use 'output->set_label(i, out_idx+min_label);' but if we have '1 4 5 7' there will be a problem | 10:25 |
blackburn | min_label for that case is 1, so then out_idx will be, for example, 2, 2+1=3 and there is no '3' label | 10:26 |
@sonney2k | blackburn, argh - I only ever used this with labels 0...(nr_classes-1) - never ever with such cases | 10:27 |
@sonney2k | I agree that this is a problem then | 10:28 |
blackburn | so, may I do it just like in CKNN? | 10:28 |
blackburn | or better to check labels? | 10:28 |
warpy | sonney2k, what is the closest typemap example to a higher level language that i use as reference | 10:29 |
@sonney2k | warpy, ? | 10:30 |
@sonney2k | andy in *_modular/swig_typemaps.i can be used as examples | 10:30 |
warpy | if i had a java typemap i could easily convert it to csharp | 10:30 |
warpy | i am just asking which one of those typemaps would you suggest as the closest most readable example to use as a base | 10:31 |
warpy | if i dont want to rewrite everyline from scratch | 10:31 |
@sonney2k | warpy, how so? Is java jni so similar to csharps native interface | 10:31 |
warpy | yeah i always go back and forth with java and c# | 10:32 |
@bettyboo | morning | 10:32 |
warpy | very similar | 10:32 |
@sonney2k | warpy, look at octave_modular then... | 10:32 |
-!- yayo3 [~jakub@ip-78-45-113-245.net.upcbroadband.cz] has joined #shogun | 10:35 | |
warpy | so r,octave,python have a typemap. matlab is closed and it doesnt. then how do you interface with it? | 10:37 |
yayo3 | so there's a log of black magic know as linear algebra in shogun, right? | 10:38 |
yayo3 | it might be interesting project to make interface to some GPGPU linear algebra libraries | 10:39 |
yayo3 | so anyway if I have a application update (new info that wasn't there last week), can I use the "comment" on melange, or do I need to send an email or something? | 10:41 |
@sonney2k | warpy, not using any of the cooler modular interfaces (see static interfaces in the doc) | 10:43 |
@sonney2k | yayo3, yes use the comment box | 10:43 |
@sonney2k | yayo3, shogun is not really a linear algebra library - we are only using atlas for such cases | 10:44 |
@sonney2k | yayo3, there are many such libraries though (just now don't recall any :-[ | 10:45 |
@sonney2k | but gpu based speedups for the functions that we have in there would be cool - but I think it is more the whole algorithm that needs porting | 10:46 |
@sonney2k | (features need to be in GPU mem etc) | 10:46 |
warpy | site down ? | 10:48 |
blackburn | when porting to vector processors we could better delete shogun and write a new one :D | 10:48 |
@bettyboo | he ;D | 10:48 |
@sonney2k | warpy, which? | 10:48 |
warpy | shogun-toolbox | 10:48 |
yayo3 | sonney2k: sure, I know shogun is not a LA library :) but it uses a lot of it, right? | 10:48 |
@sonney2k | warpy, works here | 10:48 |
@sonney2k | yayo3, not that much admittedly | 10:48 |
@sonney2k | these string related algorithms / svms / kernels very often cannot gain anything from them | 10:49 |
yayo3 | I see. nevermin then | 10:49 |
serialhex | wow... alot has been going on while i wasnt paying attention!! :D | 10:50 |
@bettyboo | :> | 10:50 |
yayo3 | hmm, I can't find Regression interface. I'm I looking wrong, or should regression methods just extend Classifier? | 10:57 |
yayo3 | KRR is Classifier, SVR is Classifier. I guess that answers that | 10:58 |
@sonney2k | yayo3, they currently do | 10:58 |
@sonney2k | yes... classify is a stupid name for applying them though | 10:58 |
@sonney2k | I guess at some point we have to do add another layer: SupervisedMethod / UnsuperVisedMethod or so and then call the classify() function apply() or so | 10:59 |
@sonney2k | warpy, ok, doing c# in configure seems easy. I will try to do it on my train ride to work. | 11:00 |
@sonney2k | I am off now | 11:00 |
@sonney2k | l8r | 11:00 |
yayo3 | see ya | 11:00 |
-!- yayo3 [~jakub@ip-78-45-113-245.net.upcbroadband.cz] has quit [Quit: Lost terminal] | 11:01 | |
CIA-31 | shogun: Soeren Sonnenburg master * r9985a76 / (4 files in 3 dirs): Add csharp support (for mono) - http://bit.ly/idHHyh | 11:51 |
CIA-31 | shogun: Soeren Sonnenburg master * r549dde2 / src/lua_modular/swig_typemaps.i : Merge branch 'master' of https://github.com/sploving/shogun - http://bit.ly/gHojmc | 11:53 |
@sonney2k | warpy, now I am back | 11:53 |
warpy | hey, i saw that you added for mono | 11:53 |
@sonney2k | so csharp interfaces seem to be generated | 11:53 |
warpy | i am using vs.net | 11:54 |
@sonney2k | warpy, shouldnt that be the same thing then? | 11:54 |
warpy | i dont think so | 11:54 |
@sonney2k | someone has to try out if the generated .cs files work | 11:54 |
warpy | i looked at the swig site i think i saw csharp for non mono | 11:54 |
@sonney2k | you? | 11:54 |
@sonney2k | warpy, it says c++ -> csharp | 11:54 |
@sonney2k | so I suspect it will work with any csharp | 11:55 |
warpy | http://www.swig.org/compat.html#SupportedLanguages | 11:55 |
warpy | look at the site | 11:55 |
warpy | they distinct 2 language | 11:55 |
warpy | i mean compilers | 11:55 |
@sonney2k | warpy, but these are just links to mono and dotnet | 11:56 |
warpy | ah alright | 11:56 |
warpy | my mistake | 11:56 |
CIA-31 | shogun: Soeren Sonnenburg master * r4b3721d / (2 files): Merge branch 'master' of https://github.com/karlnapf/shogun - http://bit.ly/eDQtHq | 11:58 |
CIA-31 | shogun: Soeren Sonnenburg master * rf981d87 / (2 files): | 12:06 |
CIA-31 | shogun: remove SG_UNSTABLE & whitespace cleanup | 12:06 |
CIA-31 | shogun: Serialization support is now there SG_UNSTABLE is no longer necessary. | 12:06 |
CIA-31 | shogun: In addition remove trailing whitespaces. - http://bit.ly/fjla80 | 12:06 |
* serialhex finally goes to sleep! and will catch up on whats been going on later!!! | 12:20 | |
blackburn | sonney2k: why do CDotFeatures re-order vectors? :) | 12:21 |
blackburn | oh, may be my fault | 12:24 |
@sonney2k | blackburn, it doesn't | 12:29 |
blackburn | sonney2k: my fault, load_numbers from examples read file in a strange way | 12:30 |
@sonney2k | blackburn, not really anything near optimzied | 12:30 |
blackburn | for some reason, if I give it, for example | 12:30 |
blackburn | 1 2 | 12:30 |
blackburn | 2 3 | 12:30 |
blackburn | it will produce vectors (1 2) , (2 3) | 12:31 |
@sonney2k | yes one by column | 12:31 |
blackburn | thought that one string is one vector | 12:31 |
@sonney2k | blackburn, there is this confusion we have no solution for: matlab, octave, R matrices are column major so one column== one vector | 12:32 |
blackburn | so, if I want to test it on 5 vectors | 12:32 |
@sonney2k | in ascii formats it is supposedly the opposite but - no idea to make this cleanly | 12:32 |
blackburn | i should do sth like | 12:32 |
blackburn | 1 2 3 4 5 | 12:33 |
blackburn | 1 2 3 4 5 | 12:33 |
blackburn | for (1,1) , ... (5,5) | 12:33 |
blackburn | right? | 12:33 |
blackburn | in file, I mean | 12:33 |
@sonney2k | meh | 12:33 |
@sonney2k | I don't know (I mix this up myself). | 12:33 |
blackburn | I'm confused with it pretty much :) | 12:33 |
warpy | i am trying to understand the difference between static and modular, statis is basically adding a wrapper function for every shogun function and modular you already have the infrastructure, but you just add support for moving the data in and out of it. and swig talks to shogun and does everything for you? | 12:34 |
blackburn | may be Chris would help, but he is not answering at all :) | 12:34 |
@bettyboo | :> | 12:34 |
blackburn | seems that he is working on examples, if I understand it right | 12:34 |
@sonney2k | blackburn, some people have to do the work they are paid for | 12:35 |
@sonney2k | :D | 12:35 |
blackburn | sonney2k: what do you mean? | 12:35 |
@sonney2k | blackburn, well we are paid for doing research ... shogun is 'just' a side project :D | 12:36 |
blackburn | I know | 12:36 |
@sonney2k | blackburn, so it is very difficult to do research and answer questions all the time. | 12:36 |
@sonney2k | that is why | 12:36 |
blackburn | sonney2k: ah, I mean he is not answering in irc :) | 12:37 |
blackburn | at least during last week | 12:37 |
blackburn | but answering mail | 12:37 |
@sonney2k | warpy, the static interface is what we had there in the beginning: one global variable for e.g. a classifier and some ascii like control functions | 12:37 |
@sonney2k | they work with almost all languages | 12:38 |
blackburn | I did | 12:38 |
blackburn | 1 1 2 2 3 3 | 12:38 |
blackburn | 4 4 5 5 6 6 | 12:38 |
blackburn | and it produced (1,4) , (1,4) , (2,5), (2,5), (3,6), (3,6) | 12:38 |
@sonney2k | but you cannot have e.g. multiple classifiers interacting etc | 12:38 |
blackburn | so, seems to be right | 12:38 |
warpy | can you elaborate a little more sonney2k | 12:38 |
@sonney2k | looks like | 12:38 |
@sonney2k | warpy, I don't know in which way - have a look at the examples for static / modular then you see the difference | 12:39 |
warpy | i actually did look | 12:39 |
@sonney2k | warpy, modular interfaces are just the C++ classes with some functions removed/renamed/parameter translated | 12:39 |
blackburn | seems that my GNB is training now, hooray | 12:41 |
-!- sploving [~sploving@124.16.139.196] has joined #shogun | 12:41 | |
blackburn | have to play on balalayka now | 12:41 |
sploving | hello sonney2k. I saw you email. thx. I still have a question. | 12:42 |
@bettyboo | hey | 12:42 |
@sonney2k | sploving, ask... | 12:42 |
@sonney2k | blackburn, heh :) | 12:42 |
@bettyboo | HA | 12:42 |
sploving | every time we need require("Kernel"), require("Feartures"), are there way we just require("shogun") | 12:43 |
@sonney2k | blackburn, try to beat an SVM ;-) | 12:43 |
@bettyboo | sonney2k, HA :> | 12:43 |
blackburn | sonney2k: with gaussian naive bayes? oh, no way | 12:43 |
@sonney2k | sploving, I mean we could have a simple shogun.lua or so that loads all the modules. | 12:43 |
sploving | That is a good idea. | 12:44 |
sploving | Let me have a try | 12:44 |
@sonney2k | (like the init_shogun script I generate for octave) | 12:45 |
sploving | I have not notice it. I will take a look at it | 12:45 |
blackburn | sonney2k: should set_features() in classifiers set _test_ features or _train_ features? | 12:46 |
blackburn | or may be I should make "set_test_features()" and "set_train_features()"? | 12:47 |
blackburn | https://github.com/shogun-toolbox/shogun/network is impressive :D | 12:49 |
@sonney2k | blackburn, we only have one set of features | 12:50 |
blackburn | sonney2k: so, what should set_features() set? features to be classified, or train ones? | 12:50 |
@sonney2k | well both | 12:51 |
blackburn | both?! | 12:52 |
@sonney2k | well these objects have just one feature set | 12:52 |
@sonney2k | when you clall train() they train on this | 12:52 |
@sonney2k | and if you call test then it will test on them | 12:53 |
@sonney2k | err classify() | 12:53 |
blackburn | oh | 12:53 |
blackburn | I see now, thank you | 12:53 |
blackburn | it's a miracle! it's working! | 12:55 |
blackburn | hehehe | 12:56 |
blackburn | now will have fun with nuclear missles and drink vodka | 12:56 |
-!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 13:02 | |
blackburn | sonney2k: I'll make a pull request for your comments on CGNB, okay? | 13:12 |
-!- sploving [~sploving@124.16.139.196] has quit [] | 13:19 | |
-!- sploving [~root@124.16.139.196] has joined #shogun | 13:23 | |
-!- blackburn [~qdrgsm@188.122.237.148] has quit [Read error: Operation timed out] | 13:53 | |
-!- blackburn [~qdrgsm@188.122.237.148] has joined #shogun | 13:55 | |
sploving | hello sonney2k | 13:56 |
sploving | i had a try. it works | 13:57 |
sploving | I define a file shogun.lua, and write all the require. | 13:58 |
sploving | I aso use "Labels = Features.Labels", so that we just need call "x = Labels()", not need call "x = Features.Labels()" | 13:59 |
-!- josip [~josip@unaffiliated/josip] has joined #shogun | 14:11 | |
-!- skydiver [4deac315@gateway/web/freenode/ip.77.234.195.21] has quit [Ping timeout: 252 seconds] | 14:21 | |
@sonney2k | sploving, great | 14:57 |
sploving | there are a lot of subkernel in kernel file. I need rename them all | 14:58 |
sploving | DotKernel = Kenel.DotKernel | 14:58 |
sploving | could we use the swig to rename? | 14:58 |
blackburn | sonney2k: GNB is here, if you will have time, please take a look on it | 14:59 |
-!- blackburn [~qdrgsm@188.122.237.148] has quit [Quit: Leaving.] | 15:35 | |
-!- siddharth [~siddharth@117.211.88.150] has quit [Remote host closed the connection] | 15:44 | |
-!- siddharth [~siddharth@117.211.88.150] has joined #shogun | 15:47 | |
-!- sploving [~root@124.16.139.196] has quit [Ping timeout: 246 seconds] | 16:06 | |
-!- sploving [~root@124.16.139.196] has joined #shogun | 16:07 | |
-!- sploving [~root@124.16.139.196] has left #shogun [] | 16:10 | |
CIA-31 | shogun: Baozeng Ding master * rd03da92 / src/lua_modular/shogun.lua : init shogun in lua - http://bit.ly/fChGQS | 16:15 |
CIA-31 | shogun: Soeren Sonnenburg master * r85fa746 / src/lua_modular/shogun.lua : Merge branch 'master' of https://github.com/sploving/shogun - http://bit.ly/e7J1i6 | 16:15 |
CIA-31 | shogun: Soeren Sonnenburg master * r6a817ec / (5 files in 3 dirs): Merge branch 'master' of https://github.com/SergeyLisitsyn/shogun - http://bit.ly/h9nCOR | 16:17 |
-!- yayo3 [c371f1a6@gateway/web/freenode/ip.195.113.241.166] has joined #shogun | 16:24 | |
CIA-31 | shogun: Soeren Sonnenburg master * r4dd2b60 / (3 files): Merge branch 'master' of https://github.com/Bautista/shogun - http://bit.ly/fYoGlH | 16:46 |
CIA-31 | shogun: Konrad Rieck master * r4f69e19 / README : fixed typos in README - http://bit.ly/dPWSY4 | 16:51 |
-!- yayo3 [c371f1a6@gateway/web/freenode/ip.195.113.241.166] has quit [Quit: Page closed] | 17:00 | |
CIA-31 | shogun: Soeren Sonnenburg master * rfcc7455 / (9 files in 2 dirs): Merge branch 'master' of https://github.com/ahmaurya/shogun - http://bit.ly/dTuYlt | 17:04 |
@knrrrd | baaaack. | 17:09 |
@knrrrd | bettyboo: did you record all the conversations? so that you have something to keep in mind ;) | 17:10 |
@bettyboo | knrrrd: I have implemented a fast nearest neighbor classification (in case of k=1) in CKNN class. what should I do after cleaning and commenting to get it commited? | 17:10 |
@knrrrd | bettyboo: nearest neightbor is similar to remembering stuff, but I was just asking if you kept stuff in mind | 17:11 |
@bettyboo | knrrrd: i am just asking which one of those typemaps would you suggest as the closest most readable example to use as a base | 17:11 |
-!- akhilvij [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 17:13 | |
@knrrrd | bettboo: i am confused | 17:13 |
@knrrrd | i mean "bettyboo" | 17:13 |
@bettyboo | knrrrd: I mean should I do this? | 17:13 |
@knrrrd | yes, you should ;) | 17:14 |
@bettyboo | yeah!?! | 17:14 |
-!- alesis-novik [~alesis@188.74.87.84] has quit [Quit: I'll be Bach] | 17:17 | |
-!- knrrrd is now known as mlsec | 17:18 | |
-!- mlsec [~knrrrd@bane.ml.tu-berlin.de] has quit [Quit: ZNC - http://znc.sourceforge.net] | 17:19 | |
-!- mlsec [~mlsec@bane.ml.tu-berlin.de] has joined #shogun | 17:20 | |
-!- mode/#shogun [+o mlsec] by ChanServ | 17:20 | |
@mlsec | we received almost 70 applications! | 17:24 |
@mlsec | that's incredible. | 17:24 |
@mlsec | thanks for your support | 17:24 |
-!- akhilvij [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 17:28 | |
CIA-31 | shogun: Soeren Sonnenburg master * r3bb5ffe / (2 files): minor coding style changes in gaussiankernel - http://bit.ly/fLSeQB | 17:50 |
CIA-31 | shogun: Soeren Sonnenburg master * r7ba79c1 / .gitignore : ignore generated *.cs in c# interface - http://bit.ly/eV09zs | 17:50 |
dvevre | hey sonney2k. how useful do you think it will be to have a standard serialization format (like google's protocol buffers) for shogun? | 18:29 |
@bettyboo | morning | 18:29 |
dvevre | bettyboo: g'morning to ya! | 18:29 |
@bettyboo | dvevre: in kpca, you have a set of vectors and your learn a new basis in the kernel-induced feature space. | 18:29 |
dvevre | because i'm going to work on that for vw | 18:29 |
dvevre | so i thought why not try getting that into shogun as well | 18:29 |
-!- blackburn [~qdrgsm@188.168.14.95] has joined #shogun | 19:11 | |
-!- blackburn [~qdrgsm@188.168.14.95] has quit [Quit: Leaving.] | 19:44 | |
-!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 19:45 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 19:50 | |
@sonney2k | dvevre, I don't even know googles protocol buffers | 20:16 |
dvevre | sonney2k: its google's way of serializing data.. and is becoming industry standard, apparently | 20:17 |
dvevre | they say it's optimized for storing large datasets too | 20:17 |
dvevre | the vw team is planning to replace their custom cache format with this.. and i was suggested to look into it | 20:18 |
@sonney2k | dvevre, so you propose to do this for C++ objects too | 20:18 |
@sonney2k | ? | 20:18 |
dvevre | http://code.google.com/apis/protocolbuffers/docs/cpptutorial.html | 20:18 |
@sonney2k | or just for data | 20:18 |
dvevre | yeah | 20:18 |
dvevre | for shogun features objects | 20:18 |
@sonney2k | sorry have to leave | 20:19 |
@sonney2k | will be back in 1-2hrs | 20:19 |
@bettyboo | morning | 20:19 |
dvevre | bye | 20:19 |
@bettyboo | nite | 20:19 |
-!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Quit: Page closed] | 20:21 | |
-!- blackburn [~qdrgsm@188.168.14.95] has joined #shogun | 20:34 | |
blackburn | https://github.com/shogun-toolbox/shogun/network looks like underground scheme :D | 20:39 |
@bettyboo | yeah | 20:39 |
-!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has quit [Quit: Try HydraIRC -> http://www.hydrairc.com <-] | 20:43 | |
-!- Tanmoy_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 20:52 | |
-!- blackburn1 [~qdrgsm@188.168.5.156] has joined #shogun | 21:08 | |
-!- blackburn [~qdrgsm@188.168.14.95] has quit [Ping timeout: 240 seconds] | 21:09 | |
-!- blackburn1 is now known as blackburn | 21:09 | |
-!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Ping timeout: 252 seconds] | 21:09 | |
CIA-31 | shogun: Rafał Surowiecki master * r47bc5b4 / (3 files in 2 dirs): SparsePolyFeatures and RationalQuadraticKernel post-review fixes - http://bit.ly/dSfERr | 21:45 |
CIA-31 | shogun: Rafał Surowiecki master * r520f75c / (14 files in 4 dirs): Merge - http://bit.ly/gaAPvz | 21:45 |
CIA-31 | shogun: Heiko Strathmann master * r7f0b001 / (3 files): Added implementation of the 'Distant Segments'-string-kernel - http://bit.ly/hwY3zs | 21:45 |
CIA-31 | shogun: Heiko Strathmann master * r34b078e / src/libshogun/kernel/GaussianKernel.h : renamed member variable compact to m_compact. fixes a compile error - http://bit.ly/hlKmAc | 21:45 |
CIA-31 | shogun: Soeren Sonnenburg master * rdd3033b / (5 files in 3 dirs): Merge branch 'master' of https://github.com/vzxzcz/shogun into klaus - http://bit.ly/h8mZNN | 21:45 |
blackburn | sonney2k: one more pull for CGNB, forgot some type check | 21:46 |
blackburn | really forgot it :) | 21:51 |
@bettyboo | lol ;D | 21:51 |
CIA-31 | shogun: Soeren Sonnenburg master * rabe1f10 / (2 files): Merge branch 'master' of https://github.com/SergeyLisitsyn/shogun - http://bit.ly/ijMPSn | 21:51 |
blackburn | thank you | 21:52 |
@sonney2k | serialhex, OK now I believe you finally :) | 21:54 |
@sonney2k | blackburn, thanks for the patch | 21:54 |
@sonney2k | really hard to find problems in you patch now - so good that you critically revised it on your own :) | 21:55 |
@bettyboo | ;> | 21:55 |
-!- blackburn [~qdrgsm@188.168.5.156] has quit [Ping timeout: 246 seconds] | 21:56 | |
-!- blackburn [~qdrgsm@188.168.4.84] has joined #shogun | 21:57 | |
blackburn | damn cable! | 21:57 |
@sonney2k | <sonney2k> serialhex, OK now I believe you finally :) | 21:57 |
@sonney2k | <sonney2k> blackburn, thanks for the patch | 21:57 |
@sonney2k | <sonney2k> really hard to find problems in you patch now - so good that you critically revised it on your own :) | 21:57 |
@bettyboo | ;D | 21:57 |
blackburn | sorry for getting disconnected often | 21:58 |
blackburn | sonney2k: so, is it okay? glad to hear it! :) | 21:59 |
serialhex | sonney2k: no problem! i'm going to work on getting the float64_t version workin in swig tonight | 22:01 |
@sonney2k | serialhex, what is the problem with it? | 22:02 |
blackburn | I'm lost with what to do next :D | 22:02 |
@bettyboo | blackburn, rahaha | 22:02 |
serialhex | i can't remember _exatly_ the exception that it threw, but it was something to the extent of: "i dont know which version to use float32 or 64, clarify or i'll just sit here and do nothing" | 22:03 |
serialhex | that's WAY paraphrasing, but you get the idea, if you want to see it uncomment the float64 version & compile... you'll see it | 22:04 |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 240 seconds] | 22:04 | |
@sonney2k | blackburn, russian doc :-) | 22:04 |
@bettyboo | sonney2k, he | 22:05 |
@sonney2k | serialhex, hmmhh tthat could be | 22:05 |
blackburn | sonney2k: oh, I could get myself 'behind' with not writing code :) | 22:06 |
@bettyboo | ;> | 22:06 |
blackburn | if I will start translating | 22:06 |
@sonney2k | heh | 22:06 |
blackburn | as you see, I hope to get into GSoC with shogun :)) | 22:06 |
blackburn | anyway I really don't know what to do next and will translate it a little | 22:07 |
@sonney2k | blackburn, if you want you can have a look at PCACut.h ... | 22:09 |
@sonney2k | that is a preproc that does pca and then removes dimensions if the eigenvalues are below some threshhold | 22:10 |
blackburn | :) are you kidding? I know what is PCA :) | 22:10 |
@bettyboo | hihi | 22:10 |
blackburn | I'm applying to dim.reduction idea! | 22:11 |
blackburn | :) | 22:11 |
@bettyboo | 22:11 | |
@sonney2k | blackburn, no kidding - I don't know if you know shogun's preproc framework | 22:11 |
blackburn | ah, you mind shogun-specific ;) | 22:12 |
@bettyboo | blackburn, ;D | 22:12 |
@sonney2k | you could extend PCACut to either reduce to a fixed number of dimensions or to eigenvalue<eps | 22:12 |
blackburn | sonney2k: I don't know, seems that people who will realize KPCA during GSoC will fix it up | 22:13 |
blackburn | if that idea will take place | 22:13 |
@sonney2k | blackburn, if you want to do more advanced stuff - same for KPCACut. | 22:13 |
blackburn | sonney2k: anyway, thank you for 'idea', I'll look it a bit later :) | 22:14 |
@sonney2k | blackburn, well you could of course write python examples for all the new kernels if you like | 22:14 |
blackburn | sonney2k: btw, that is the matter with it? I mean there is no graphics or etc, how could one realize that classifier works? | 22:15 |
blackburn | what*, sorry | 22:15 |
@sonney2k | I don't understand | 22:16 |
blackburn | sorry :) | 22:16 |
@bettyboo | lol ;D | 22:16 |
blackburn | I wrote a test in python for CGNB and it doing almost nothing | 22:16 |
blackburn | and don't representing itself's results | 22:16 |
blackburn | same thing with other classifiers | 22:16 |
@sonney2k | blackburn, CGNB was multiclass classifier but needs linearly separable data? | 22:18 |
blackburn | sonney2k: it don't | 22:18 |
@sonney2k | I mean visualize the multiclass data | 22:19 |
@sonney2k | you will see it is a few gauss blobs | 22:19 |
blackburn | yeap, but there is no that visualization, that's it | 22:19 |
@sonney2k | blackburn, well it has a train / test data set and NB should perfom not too bad on this... | 22:20 |
blackburn | sorry for bad english :) trying to translate intro and switching languages in my mind | 22:20 |
blackburn | sonney2k: I'm sure it will, but when we launch that test we see nothing, it classifies something and then exit | 22:21 |
blackburn | oh, translating machine learning text is difficult :D understand it right but thinking about proper words in russian verrrrry slow | 22:25 |
blackburn | sonney2k: oh, it seems that i will have troubles with some terms | 22:33 |
@sonney2k | blackburn, just read the chinese docuemntation then 8-) | 22:33 |
blackburn | may I jump from a window first? :) | 22:33 |
@bettyboo | rahaha ;D | 22:33 |
blackburn | sonney2k: for example, there is no russian term for 'multiple kernel learning' | 22:35 |
@sonney2k | blackburn, I can imagine | 22:35 |
blackburn | will write множественное ядровое обучение :D | 22:35 |
@bettyboo | funny | 22:35 |
blackburn | I have same troubles when lecturing machine learning at my university | 22:36 |
blackburn | sometimes used english terms | 22:36 |
blackburn | there is no machine learning teachers so I tried to be one :D | 22:38 |
@bettyboo | ;D | 22:38 |
blackburn | sonney2k: is linear programming machine = machine of linear programming (looks ugly, but will help to translate) | 22:42 |
@sonney2k | blackburn, it is a svm with l1-regularization | 22:42 |
blackburn | so, SVM is машины опорных векторов :) it could be машины линейного программирования | 22:43 |
@bettyboo | ;D | 22:43 |
blackburn | there is an inversed order, I think you know, so have to do some separation | 22:44 |
blackburn | sonney2k: pre-pre-beta of introduction http://pastebin.com/DUtXyYLF :) | 22:51 |
@sonney2k | blackburn, I can even reas some things: what does обучения mean? | 22:52 |
blackburn | learning :) | 22:52 |
@sonney2k | heh | 22:53 |
@sonney2k | very nice | 22:53 |
CIA-31 | shogun: Soeren Sonnenburg master * r979e303 / src/Makefile.template : fix make doc / make install targets for csharp,lua,... - http://bit.ly/g1QPBi | 22:54 |
blackburn | sonney2k: will make it more 'smoothly', now it looks like translation | 22:54 |
@sonney2k | well you can use google translate :) | 22:54 |
blackburn | it will make some terrible thing | 22:55 |
@sonney2k | heh | 22:55 |
siddharth | sonney2k, what should we do next to get to know shogun?...have been busy lately with my research paper | 22:56 |
siddharth | just want to start with the basics...so where should I look first? | 22:57 |
@sonney2k | blackburn, I have some comment for your CGNB - maybe you rename it to Naive Bayes ... and btw, there is CMath::sq(x) that you can use instead of pow(x,2) | 22:57 |
blackburn | sonney2k: there is a difference between GNB and NB (with no G).. | 22:58 |
blackburn | as for sq, will do it now | 22:58 |
@sonney2k | blackburn, then GaussianNaiveBayes - more readable | 22:58 |
blackburn | ah, I see now | 22:58 |
@sonney2k | blackburn, we already have too many abbrv's in shogun that even I forget what they stand for, GMNPSVM, etc | 22:59 |
blackburn | :) okay, no problem | 22:59 |
@bettyboo | yeah | 22:59 |
siddharth | sonney2k, any comments? | 22:59 |
@sonney2k | siddharth, anything related to your project or enjoy doing. if you are desperately looking for something to work on - well look for SG_NOTIMPLEMENTED or SG_UNSTABLE and fix serialization or implement some (simple algorithm) | 23:00 |
siddharth | sonney2k, I was thinking of introducing conceptual learning.. | 23:01 |
siddharth | candidate elimination algorithm | 23:02 |
* sonney2k does not know what this does / is | 23:03 | |
blackburn | candidate elimination could be good algorithm for you now, sonney2k :) | 23:03 |
@bettyboo | :> | 23:03 |
blackburn | with these 60 candidates :D | 23:03 |
@bettyboo | yep | 23:04 |
@sonney2k | :) | 23:04 |
-!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 23:05 | |
siddharth | sonney2k, Its basically learning from examples | 23:07 |
siddharth | http://www.google.co.in/url?sa=t&source=web&cd=3&ved=0CCIQFjAC&url=http%3A%2F%2Fwww.site.uottawa.ca%2F~nat%2FCourses%2FCSI5387%2FML_Lecture_2.ppt&ei=Hm2jTfSQMszHrQfgjYGNAw&usg=AFQjCNE9h6daJchjGXhSbhiYjj6V9cWBAA&sig2=IsbBHRdR9lbv-XR0LPXcTA | 23:07 |
siddharth | here is a ppt | 23:07 |
blackburn | that link is damn awful :) | 23:07 |
siddharth | :P | 23:08 |
siddharth | or I can start with my SVM work | 23:09 |
blackburn | sonney2k: | 23:09 |
@sonney2k | siddharth, I see - when does one use this nowadays? | 23:09 |
blackburn | ./kernel/GaussianKernel.h:118: error: ‘m_compact’ was not declared in this scope | 23:09 |
@sonney2k | blackburn, git pull I hope | 23:09 |
@sonney2k | (my bad) | 23:10 |
blackburn | ah | 23:10 |
blackburn | okay ) | 23:10 |
siddharth | sonney2k, dont know for sure...btw did you read my application | 23:10 |
blackburn | now working, thanks | 23:11 |
blackburn | forgot to merge with upstream :) | 23:11 |
@bettyboo | :> | 23:11 |
-!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 23:16 | |
siddharth | sonney2k, whenever you read my application,can you write some comments/suggestions so that I can work on that? | 23:17 |
blackburn | sonney2k: pull request created | 23:21 |
blackburn | sonney2k: renamed and tested it | 23:21 |
blackburn | have to leave now, see you later | 23:22 |
@sonney2k | siddharth, I read both of them - it is very hard for me to ask specific questions that are not a lot of work | 23:22 |
-!- blackburn [~qdrgsm@188.168.4.84] has quit [Quit: Leaving.] | 23:23 | |
@sonney2k | siddharth, I mean if you could e.g. show that you can wrap sgd-qn or newton svm already you would certainly be in - but that is a lot to ask for and I really don't see anything in between *sigh* | 23:24 |
siddharth | sonney2k, hmm I see...well then I will start working on this and give you update | 23:25 |
@sonney2k | siddharth, I think integrating SGD-QN should be very easy | 23:26 |
@sonney2k | so I would start with that | 23:26 |
siddharth | sonney2k, ok I will look into that...thanks for the help | 23:26 |
siddharth | :) | 23:26 |
CIA-31 | shogun: Sergey Lisitsyn master * re5fa152 / (8 files in 3 dirs): Renamed GNB to GaussianNaiveBayes - http://bit.ly/dLX7ca | 23:27 |
siddharth | sonney2k, one more question...shogun does not have kernel k means right? | 23:28 |
@sonney2k | siddharth, you probably have had a look at http://www-etud.iro.umontreal.ca/~bordesa/mywiki/doku.php?id=sgdqn already | 23:29 |
siddharth | yes | 23:29 |
@sonney2k | the file svmsgd2.cpp in there is what shogun's sgd is based on (I just converted it to use the operations from CDotFeatures - hopefully correclty) | 23:29 |
@sonney2k | so svmsgdqn.cpp is the one to translate ... | 23:34 |
siddharth | I was not clear about the calibrate function in SVMSGD....I will again have a look at it | 23:34 |
@sonney2k | siddharth, me neither - I just tried to translate what was there | 23:34 |
siddharth | ok then I will start working upon it | 23:34 |
@sonney2k | siddharth, no guarantees that I haven't made a mistake though.... | 23:34 |
@sonney2k | siddharth, no we don't have kernel k-means | 23:34 |
siddharth | sonney2k, I will let you know if I find any...also will see about the kernel k means | 23:34 |
@sonney2k | siddharth, ok thanks! | 23:35 |
siddharth | thank u too :) | 23:36 |
@bettyboo | rahaha ;> siddharth | 23:36 |
siddharth | hmm hi bettyboo..long time | 23:37 |
@bettyboo | siddharth: then you can mess around in your git and push to your repository when you are ready | 23:37 |
@bettyboo | siddharth, hiho | 23:37 |
siddharth | as u say bettyboo | 23:37 |
@bettyboo | siddharth: (features need to be in GPU mem etc) | 23:37 |
siddharth | ok i wil do that bettyboo | 23:38 |
@bettyboo | siddharth: I see, will do a fix in a minute | 23:38 |
@sonney2k | good night | 23:41 |
dvevre | good night. see you in 2 days | 23:42 |
-!- Tanmoy_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 23:57 | |
--- Log closed Tue Apr 12 00:00:36 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!