Open in new window / Try shogun cloud
--- Log opened Thu May 19 00:00:11 2011
@sonney2kserialhex, hmmhh but this is w/ make distclean before?00:20
serialhexwhat? the java error?  no, it happened when i just said 'make'  i can try that in a bit, but i'm still compiling (this time with just ruby & python modular)00:22
@sonney2kthat shouldn't happen00:23
@sonney2kI would suggest to run git clean -dfx00:23
@sonney2kto start off fresh00:23
@sonney2k(will remove all files not in git)00:23
CIA-31shogun: Soeren Sonnenburg master * rddf2879 / src/libshogun/base/SGObject.h :00:24
CIA-31shogun: Revert "potentially fix a gcc 4.6 compilation error"00:24
CIA-31shogun: This reverts commit 0fe924eca47c3ff5f7c88cac489c2850011d4aec. -
CIA-31shogun: Soeren Sonnenburg master * r4c16c19 / (2 files in 2 dirs): introduce SG_UNREF_NOT_NULL to compilation error -
@sonney2kserialhex, ^^ btw00:24
@sonney2kI hope this fixes it00:24
serialhexooh, ok, will cancel & re-try!00:25
@sonney2kanyway I will have to sleep now00:25
serialhexlater, i'll e-mail you with the results sonney2k00:26
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]00:34
-!- ThePing [~phycho@] has joined #shogun02:27
-!- ThePing [~phycho@] has left #shogun []02:27
-!- sploving [~root@] has quit [Remote host closed the connection]04:51
-!- sonney2k [~shogun@] has quit [Ping timeout: 258 seconds]05:31
-!- sonney2k [] has joined #shogun05:32
-!- mode/#shogun [+o sonney2k] by ChanServ05:32
-!- blackburn [~qdrgsm@] has joined #shogun09:21
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]10:41
-!- heiko [] has joined #shogun12:08
-!- ameerkat [] has quit [Ping timeout: 260 seconds]13:17
-!- sploving [~root@] has joined #shogun13:54
splovingsonney2k, aroud?13:54
splovingI could map the jblas,  DoubleMatrix13:55
splovingI will push it right now13:55
-!- heiko [] has quit [Ping timeout: 258 seconds]13:58
@sonney2ksploving, great!14:23
@sonney2ksploving, I suggest to continue with a list of strings14:23
splovingsonney2k, I pushed it14:23
splovingsonney2k, does SGstring work now?14:24
@sonney2ksploving, it will be SGStringList14:24
@sonney2klook at DataType.h I introduced this new type14:24
splovingI think you could give me python example. like typemap and how to use it14:25
@sonney2kwill implement some example for it tonight - so would be great if you could start drafting the typemap14:25
splovingthen I could speed up14:25
@sonney2ksploving, just look at the data type and what is currently there from the python side14:25
splovingsonney2k, when considering support string, we should not consider jblas, as it does not have it14:29
@sonney2ksploving, correct14:29
splovingwe just consider the java side.14:29
splovingOK. then it will be easy14:29
@sonney2ksploving, I didn't do python yet but it is easy given that we now know how things work14:43
splovingsonney2k, could you give me a test example14:44
splovingso I could have a try14:44
@sonney2ksploving, not yet14:44
@sonney2khopefully tonight14:44
@sonney2kso draft the code only for now and tomorrow you can test14:45
splovingsonney2k, it does not matter. I have staring the computer for a whole day. I need a rest14:46
splovingsonney2k, SGString<T>, does it mean the string could be double14:48
splovingint, and other type?14:48
splovingnot only char?14:48
splovingchar *14:48
-!- sploving [~root@] has left #shogun []15:07
serialhexsonney2k: the fix seems to work, now i have to figure out why it's not finding my ruby.h file :-/17:32
@mlsecsonney2k: Alive?17:37
serialhexi think he's gone the way of the dodo mlsec :P17:40
-!- bettyboo [] has quit [Remote host closed the connection]17:41
serialhexwtf??  bettyboo left?!!?!?17:41
@mlsecjust some internal check17:41
serialhexahh, i see17:41
@mlsecbetty, now has captured roughly 10k sentences.17:42
@mlsecso i need to filter stuff a little.17:42
serialhexbtw, she's pretty spiffy, i dont know if yo've read some of the logs but she's eerily accurate in her responses17:42
-!- bettyboo [] has joined #shogun17:44
-!- mode/#shogun [+o bettyboo] by ChanServ17:44
@mlsecit's no black magic. just randomization and simple pattern matching17:44
serialhexcool cool17:44
@mlsechowever, i had time ... well, there is plenty of room for extensions17:44
@mlsecif i had17:45
serialhexare you going to open source her?17:45
serialhexhey bettyboo, how are you??17:45
@bettybooserialhex: and assume that cov and mean (together with their sizes) are set as member varialbes17:45
serialhexoh, ok17:45
@mlsecit's just 240 lines of tcl code17:45
* serialhex thought it would be more complex than that17:46
@mlsectake 240 lines of code + 9000 captured messages17:46
@mlsecand you have a bot17:46
@mlsecone great addition that is currently missing in betty, is the ability to track conversations17:47
@bettyboomlsec: how peopleless it is..17:47
serialhexyeah, that would make her even more spooky :D17:47
@mlsecin irc it is often not difficult to detect and observe conversations due to nick names in the messages17:47
@mlsecusing conversations could help to get a better answer/response behavior17:48
serialhexthat makes sense17:48
@mlsecso an obvious test for the lack of this ability is this:17:48
@mlsecbetty: how are you?17:48
@bettyboomlsec: hey how are you??17:48
@mlsechehe. it's more matching17:49
serialhexyeah, you could do something that matches common responses to Q's like that17:49
@mlsecyeah. in principle, you can also automatically detect questions; even if the question mark is missing or the structure is not correct, e.g. you are an atom?17:50
@mlseci have been doing this with a colleague some years ago. tools for parsing and processing natural language text are pretty advanced nowadays17:51
serialhexcool cool17:51
serialhexyeah i saw a internet chatbot (i forget the url) that was kind of interesting17:51
serialhexthe one thing that seems to be missing in all the ones i've seen/heard of is the ability to start a conversation on it's own, i'm sure it's easy to do random strings, but something more advanced is probably much harder17:53
@mlsecbrings us back to conversations. if we would track them, we could easily pick a first message to start a chat17:54
-!- alesis-novik [~alesis@] has quit [Quit: I'll be Bach]18:32
-!- blackburn [~qdrgsm@] has joined #shogun19:34
serialhexoi blackburn!!19:48
blackburnhow are ya?19:49
serialhexpretty good, yourself?19:49
blackburnfine, just waked up :D19:49
serialhexwtf??  it's got to be like 11pm or something there?  [my cool clocks arent set up :( ]19:50
blackburnaha, fell asleep about 7pm and waked up 9pm :D19:51
serialhexahh, cool catnap :D19:53
@sonney2kserialhex, did my patch fix the compile issue?20:44
@sonney2kahh ok good20:46
blackburnsonney2k: added examples for both prc and roc20:48
serialhexsonney2k: i guess you saw, i'm going to try and fix the problem with ruby not being compiled but shogun itself seems to compile fine20:51
serialhexbbl, doin some house work20:52
@sonney2kblackburn, nice :)20:58
blackburnsonney2k: thinked a lot about which classifiers to choose :D LDA+SVM seems to be ok?20:59
@sonney2kblackburn, you could illustrate that even w/o classifiers20:59
@sonney2ke.g. just use some random data21:00
blackburnI know, but this way could be more 'real'21:00
blackburnit shows that svm is better than lda :D21:00
@sonney2kblackburn, well depends21:20
@sonney2ksingle hand typing sucks...21:21
blackburnwhy single?21:21
@sonney2kone arm baby21:22
blackburnsuch a kung-fu :D21:22
@bettybooblackburn, hoho!21:22
@sonney2kblackburn, now 2 arms :)21:24
@sonney2kI looked at your examples21:25
@sonney2kyou should create one without figures21:25
@sonney2kotherwise we have no testsuite for this21:25
@sonney2kand not just roc/prc21:25
@sonney2kbut all the measures21:25
@sonney2kyou could do the for random data for this purpose21:26
blackburni'll do it a bit later21:26
@sonney2kslighly correlated with label ...21:26
blackburnah, i'd rather do it now hehe21:28
CIA-31shogun: Sergey Lisitsyn master * r3bec041 / (2 files): Added graphical examples for PRC and ROC -
CIA-31shogun: Soeren Sonnenburg master * r064cdc5 / (2 files): Merge branch 'master' of git:// -
@sonney2kblackburn, go ahead21:28
-!- ameerkat [] has joined #shogun21:28
@sonney2kblackburn, figures look cool. one way to improve would be to plot both rocs in one figure and make one red-ish transparent and one blueish21:29
blackburnsonney2k: they are almost equal sometimes, will it be informative?21:30
@sonney2kblackburn, I usually plot roc curves just as lines21:31
@sonney2kand for many classifiers in one figure21:31
@sonney2kthis way I can see when which classifier is better21:31
blackburnsonney2k: I see21:33
@sonney2kanyway nothing mandatory - figures look nice21:34
CIA-31shogun: Soeren Sonnenburg master * rc852a03 / (2 files in 2 dirs):21:44
CIA-31shogun: Merge branch 'master' of git://
CIA-31shogun: Conflicts:21:44
CIA-31shogun:  src/java_modular/swig_typemaps.i -
blackburnsonney2k: Labels(numpy.random.randn(1,N)) caused a memory corruption :D21:45
blackburnbecause it is [[....]] but not [...]21:46
blackburnevaluator = ContingencyTableEvaluation()22:06
blackburnassert evaluator.evaluate(predicted_labels,ground_truth_labels) == base_evaluator.get_accuracy()22:06
blackburngood for test?22:06
blackburnevaluator = AccuracyMeasure()22:07
blackburnassert evaluator.evaluate(predicted_labels,ground_truth_labels) == base_evaluator.get_accuracy()22:07
blackburnor just compute it and nothing?22:07
@sonney2kjust compute it and return it22:08
@sonney2klike the example you have written about a month ago22:08
blackburnsonney2k: there are helper classes: AccuracyMeasure, etc22:09
@sonney2kblackburn, why don't you use rand(10) ?22:09
blackburnsonney2k: where?22:09
@sonney2kwhy the 1, there?22:09
blackburni fixed it already22:09
blackburnsonney2k: so about helpers22:10
blackburnshould I test it in one .py?22:10
blackburnbecause all of them are similar, it could be better to test it at once22:10
@sonney2kblackburn, I would do it like you did in C++22:14
@sonney2kall the contingency table based ones in one file22:14
@sonney2kthe others roc / prc in different22:14
blackburnyeap I did it this way for now22:14
blackburnsonney2k: done22:28
@sonney2kblackburn, I am currently writing some helper functions for strings / sparse matrix22:28
@sonney2kto use our new datatypes22:29
@sonney2kblackburn, please seed the random number generator such that this is reproducible23:04
blackburnsonney2k: do you think it is needed for numpy.random?23:07
@sonney2ksonney2k, yes it is needed23:07
@sonney2knumpy.random.seed(17) :)23:08
blackburnsonney2k: done23:09
CIA-31shogun: Soeren Sonnenburg master * r81c627f / (5 files): Merge branch 'master' of git:// (+6 more commits...) -
blackburnsonney2k: have you seen this scheme?
blackburnwe could use something similar23:13
@sonney2kblackburn, you mean a development branch?23:14
serialhexi suggested that weeks ago blackburn and no one listened to me!!! :P  but if you have more luck convincing ppl then thats awesome!  (i thinkit's pretty spiffy myself!)23:14
blackburnhotfixes + develop + features23:14
blackburnI'm already using branches for lle and mds23:14
serialhex is the main link & is a spiffy video23:15
@sonney2kblackburn, sure but the problem is how to do that across people - I mean e.g. the SGVector business is something that sploving, mikio and I need23:17
@sonney2kand heiko contributed to23:17
blackburnsonney2k: just the naming issue - i.e I should use mds,lle for my project work, hotfixes for fixes :) and some other branch to small additions23:19
blackburne.g. sploving could do it in java branch or so23:19
@sonney2kI guess we could have created a branch for that and only merged this back to master once everything is done23:19
@bettybooho sonney2k23:19
@sonney2kblackburn, you can name your branches as you like... workflowwise the only thing I prefer is to merge small changes often23:21
@sonney2kwhat is different in this release model is that we do (joint) development in master only23:25
@sonney2kbigger features in some feature branches yes (smaller ones directly in master)23:26
@sonney2kwe only tag releases so far23:26
@sonney2kno branch for releases + hotfixes23:26
blackburnmay be some more branches could be better to comprehend, well idk if it would be better23:26
@sonney2kbtw, I used to do this back in the days when we uses cvs for shogun23:27
@sonney2kbut then we never needed the release branch23:27
@sonney2kI guess now it makes more sense considering shogun's size23:27
@sonney2kanyway for isolated features all these branches are very helpful23:28
@sonney2kfor such intrusive change as the SGVector* one it does not help - this change stays a nightmare23:28
@sonney2kblackburn, though we made good progress... python typemaps are there more or less and heiko did the serialization support already23:30
blackburnsonney2k: 3 days to start of gsoc :)23:30
@sonney2kit all needs testing now, support for r_modular/octave_modular and more documentation / changes in the classes23:30
@sonney2kblackburn, yeah that will also mean weekly student reports...23:31
blackburnI would report on my progress on this sunday23:31
@sonney2kbut heh I hope some projects really get rolling23:31
@sonney2kwell 1 week after gsoc starts is sufficient :)23:32
blackburnI hope to finish ~1/2 of my project before my exams23:32
blackburnand another 1/2 right before mid-term :)23:32
@sonney2kI wish you best of luck!23:32
blackburnafter mid-term I want to make a paper about dimreduction in shogun and some cleanup/support and so on23:33
@sonney2kblackburn, I really hope that around mid-term we have the SGVector* transition finished and also the renaming of CClassifier -> CMethod23:34
@sonney2kor CMachine23:35
blackburnI think CMethod could be a nightmare23:35
@sonney2kprobably better - we do MACHINE learning23:35
blackburn// method of method that methods the method23:35
@sonney2kI wish I had more time to do these transitions now - all I have is 1hr/day max currently23:36
@sonney2kspeaking of which - I need to sleep23:36
blackburnit just like name some class CClass23:36
@sonney2kcu you around and serialhex good luck with ruby23:37
blackburnsee you23:37
@sonney2kblackburn, CGoodNight!23:37
serialhexnp i mean :P23:37
blackburnsonney2k: new CHaveANiceDream()23:37
blackburnor more java-style23:37
blackburnInitialContext ic = new InitialContext();23:37
serialhex we're prepared 4 zombie apocalypse,  r u??23:38
@sonney2kb= new CBrain(); delete b;23:38
serialhexlol sonney2k23:38
blackburnObject ref = ic.lookup("HaveANiceDreamFactory")23:38
blackburnHaveANiceDreamFactoryRemoteHome factory = (HaveANiceDreamFactoryRemoteHome) PortableRemoteObject.narrow(ref, HaveANiceDreamFactoryRemoteHome.class);23:38
blackburnHaveANiceDreamFactoryRemote = factory.create();23:39
blackburnHaveANiceDream dream = factoryRemote.create();23:40
blackburnfuck I have to sleep now23:40
serialhexsoeren.sleep :dream => :good_ones23:40
blackburnserialhex: looks commonlispy23:40
serialhexit's ruby23:40
serialhexwith a hash arg :D23:40
blackburnah, in ruby they use colons too23:40
serialhexthe colons are for symbols, which are lightweight & immutable strings23:41
blackburnit is the lisp way23:41
serialhexthey're spiffy for use in hashes & other things23:41
blackburnand of course it is clojure way too :)23:42
serialhexyeah, cause coujure >= lisp :P23:42
blackburnI played with clojure a little but didn't learnt anything cool for some reason23:43
blackburnanyway now I use only three languages: python, java and C++ :)23:44
serialhexhmm, i need to get thru this book that will teach me the basics for 7 langs23:44
blackburnI think it is useless :)23:45
serialhexyeah me == ruby :P23:45
serialhexwith << C/C++23:45
blackburnanyone will forget these things about not-needed-languages23:45
blackburnI have used a common lisp for a while but now cannot remember anything - I just don't need it23:46
blackburnsame thing with physics :D23:46
serialhexyeah, if u dont use it then u wont remember it23:46
blackburnwell it is a bit late23:47
blackburngoing sleep just like Soeren did23:47
blackburnserialhex: have a nice eh.. evening/23:47
serialhexyeah, thanks, nite blackburn!23:47
blackburnsee you23:47
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]23:48
--- Log closed Fri May 20 00:00:13 2011