--- Log opened Sat Aug 18 00:00:17 2012 | ||
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has joined #shogun | 02:10 | |
CIA-21 | shogun: Sergey Lisitsyn master * r38642f1 / (3 files in 3 dirs): Fixed a few issues with feature blocked logit regression - http://git.io/As1TYg | 02:19 |
---|---|---|
-!- blackburn [~blackburn@62.106.106.114] has quit [Quit: Leaving.] | 02:33 | |
shogun-buildbot_ | build #385 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/385 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 02:39 |
shogun-buildbot_ | build #58 of nightly_none is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/58 | 03:02 |
shogun-buildbot_ | build #68 of nightly_default is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/68 | 03:23 |
shogun-buildbot_ | build #54 of nightly_all is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/54 | 03:45 |
-!- emrecelikten [~emre@trir-5d800125.pool.mediaWays.net] has quit [Ping timeout: 244 seconds] | 04:31 | |
-!- emrecelikten [~emre@trir-5d800450.pool.mediaWays.net] has joined #shogun | 04:45 | |
shogun-buildbot_ | build #69 of nightly_default is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/69 | 07:02 |
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has quit [Ping timeout: 246 seconds] | 08:11 | |
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has joined #shogun | 11:24 | |
-!- gsomix [~gsomix@178.45.70.165] has joined #shogun | 11:52 | |
gsomix | hi | 11:52 |
-!- heiko [~heiko@host86-185-9-87.range86-185.btcentralplus.com] has joined #shogun | 12:30 | |
-!- heiko [~heiko@host86-185-9-87.range86-185.btcentralplus.com] has quit [Remote host closed the connection] | 12:58 | |
CIA-21 | shogun: Heiko Strathmann master * rcf6beaf / src/shogun/evaluation/CrossValidationMKLStorage.cpp : fixed memory bug when cross-validation was performed twice with different number of runs - http://git.io/fX837w | 13:15 |
CIA-21 | shogun: Heiko Strathmann master * r8e8fff9 / examples/undocumented/libshogun/evaluation_cross_validation_mkl_weight_storage.cpp : added example for case that was failing before - http://git.io/fYRM5g | 13:15 |
CIA-21 | shogun: Heiko Strathmann master * r7cfee21 / (2 files in 2 dirs): Merge pull request #738 from karlnapf/master - http://git.io/u4zt9Q | 13:15 |
-!- heiko [~heiko@host86-185-9-87.range86-185.btcentralplus.com] has joined #shogun | 13:15 | |
-!- sr___ [u5548@gateway/web/irccloud.com/x-abthilurbjnvpfrb] has quit [Remote host closed the connection] | 13:33 | |
shogun-buildbot_ | build #386 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/386 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 13:34 |
-!- heiko [~heiko@host86-185-9-87.range86-185.btcentralplus.com] has left #shogun [] | 13:39 | |
-!- blackburn [~blackburn@62.106.106.114] has joined #shogun | 13:52 | |
shogun-buildbot_ | build #387 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/387 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 13:53 |
-!- sr___ [u5548@gateway/web/irccloud.com/x-fgcntzhdfnowhati] has joined #shogun | 14:25 | |
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has quit [Ping timeout: 246 seconds] | 15:15 | |
-!- yoo [575b08cb@gateway/web/freenode/ip.87.91.8.203] has joined #shogun | 15:44 | |
yoo | hi all | 15:44 |
yoo | I got lots of warning at compilation | 15:44 |
yoo | related to the use of eigen3 | 15:45 |
yoo | which version of eigen3 do you use ? | 15:47 |
-!- yoo [575b08cb@gateway/web/freenode/ip.87.91.8.203] has quit [Quit: Page closed] | 15:51 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun | 16:22 | |
n4nd0 | yoo: you can turn off the warnings produced by eigen | 16:23 |
n4nd0 | yoo: just open your .config and substitute the -I of eigen for -isystem | 16:23 |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] | 18:10 | |
CIA-21 | shogun: Chiyuan Zhang master * reafbb83 / examples/undocumented/python_modular/graphical/regression_lars.py : use shogun built-in evaluation. - http://git.io/xBC_Zw | 18:28 |
CIA-21 | shogun: Sergey Lisitsyn master * r13750d9 / examples/undocumented/python_modular/graphical/regression_lars.py : Merge pull request #739 from pluskid/multiclass - http://git.io/kE7_sQ | 18:28 |
shogun-buildbot_ | build #388 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/388 blamelist: Chiyuan Zhang <pluskid@gmail.com> | 18:43 |
shogun-buildbot_ | build #389 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/389 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 19:02 |
-!- blackburn [~blackburn@62.106.106.114] has quit [Quit: Leaving.] | 19:21 | |
-!- blackburn [~blackburn@62.106.106.114] has joined #shogun | 19:23 | |
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has joined #shogun | 20:21 | |
-!- zxtx [~zv@c-76-121-51-43.hsd1.wa.comcast.net] has quit [Ping timeout: 240 seconds] | 20:42 | |
CIA-21 | shogun: Sergey Lisitsyn master * r37d4ed6 / src/shogun/lib/slep/slep_solver.cpp : Removed redundant positive and negative weighting in slep solver - http://git.io/ZYGyKw | 21:19 |
CIA-21 | shogun: Sergey Lisitsyn master * r63f32c9 / (2 files): Added tree regularized multiclass logistic regression - http://git.io/ssRU-Q | 21:19 |
shogun-buildbot_ | build #380 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/380 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:21 |
shogun-buildbot_ | build #390 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/390 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:33 |
-!- gsomix [~gsomix@178.45.70.165] has quit [Ping timeout: 246 seconds] | 21:43 | |
-!- gsomix [~gsomix@178.45.39.145] has joined #shogun | 21:57 | |
@sonney2k | hey gsomix | 22:29 |
gsomix | sonney2k, hey | 22:29 |
@sonney2k | what are you working on? | 22:29 |
gsomix | protocols for custom kernel | 22:30 |
gsomix | sonney2k, btw, I need to fix typemap | 22:31 |
gsomix | 'in' typemap for SGVector | 22:31 |
@sonney2k | gsomix, wrt what? | 22:31 |
@sonney2k | what's wrong? | 22:31 |
blackburn | does anybody remember by heart if dynamic object array should be ->reserve() before setting? | 22:32 |
@sonney2k | ??? | 22:32 |
@sonney2k | reserve? | 22:32 |
blackburn | yeah or sth like tha | 22:32 |
blackburn | ahhhh we have a check in set_machine | 22:32 |
@sonney2k | it grows autmagically | 22:32 |
blackburn | yeah but it doesn't work like set(0) set(1) | 22:33 |
blackburn | but push back should be used | 22:33 |
gsomix | sonney2k, for directors. it's needed, because sometimes args in methods is actually SGVector's not numpy.arrays. | 22:33 |
@sonney2k | gsomix, I still don't understand - you are talking about directorin typemaps? | 22:33 |
gsomix | nope, about 'in' | 22:34 |
gsomix | hm, what about directorin? | 22:34 |
@sonney2k | gsomix, what do you want to 'fix' in in typemaps? | 22:34 |
gsomix | I want to add case where the arg is SGVector. | 22:35 |
gsomix | at now only numpy.arrays may be processed | 22:36 |
@sonney2k | ahh - what do you want to do when arg is SGVector? shouldn't that work already? | 22:36 |
gsomix | if arg is SGVector => do nothing. :) | 22:37 |
gsomix | because this is actually SGVector | 22:38 |
@sonney2k | gsomix, but swig should have wrapped SGVector already - so one should be able to pass SGVectors to functions | 22:38 |
@sonney2k | gsomix, I think what is missing is some SGVector constructor that takes an SGVector as input - such that one can create an SGVector via a numpy array | 22:43 |
gsomix | sonney2k, awwww, it seems that I'm stupid and there is no problem. | 22:45 |
@sonney2k | gsomix, will you bring the director examples back? | 22:46 |
gsomix | yep | 22:47 |
@sonney2k | gsomix, so except for typemaps for model selection - what is missing now? | 22:50 |
blackburn | sonney2k: can't we avoid typemaps for modelselection but extend class for that? | 22:51 |
@sonney2k | blackburn, I don't understand? | 22:51 |
@sonney2k | what do you want to extend? | 22:51 |
gsomix | sonney2k, protocols for custom kernel and examples. | 22:51 |
blackburn | sonney2k: add python native constructor with all {'C' : ...} stuff | 22:51 |
@sonney2k | blackburn, no - that is what typemaps are for | 22:52 |
@sonney2k | gsomix, ok then do these first maybe we can still manage to do these modsel typemaps before 2.0 - but lets see | 22:52 |
gsomix | ok | 22:53 |
blackburn | sonney2k: I think handling that in C++ is a wrong way | 22:59 |
blackburn | what I suggest is a faster in means of development way to do the same | 22:59 |
@sonney2k | blackburn, please explain more... | 23:06 |
blackburn | sonney2k: that {} map thing you were thinking of works only for python, right? | 23:06 |
blackburn | {'C' : {1,2,3} } or os? | 23:07 |
blackburn | so? | 23:07 |
@sonney2k | for any language that supports stuff like this... | 23:07 |
@sonney2k | but lets say only python - so now what? | 23:08 |
blackburn | this can be handled in python effectively | 23:08 |
blackburn | so it should be python native extension, like extended constructor | 23:08 |
blackburn | I see no reason to get hands dirty with that kind of parsing | 23:09 |
blackburn | this structure can be effectively parsed in python - just creating model selection parameters and setting childs | 23:10 |
blackburn | any C++ version will do the same with same performance but much harder to develop, debug and maintain | 23:10 |
@sonney2k | blackburn, except one thing: you cannot directly pass this dictionary to the modsel routines | 23:14 |
@sonney2k | you would have to call some function to get the C++ object | 23:14 |
blackburn | IIRC everything this dictionary does can be done in python | 23:14 |
blackburn | am I wrong? | 23:14 |
@sonney2k | blackburn, the model selection itself runs in C++ in shogun - so you need to pass this 'modsel' object to libshogun | 23:15 |
@sonney2k | so what you can do is create this object via python code | 23:15 |
@sonney2k | but you cannot have the mapping dictionary -> 'modsel' object | 23:16 |
@sonney2k | that is only possible via typemaps | 23:16 |
blackburn | sonney2k: how should that dictionary look like? | 23:16 |
@sonney2k | gsomix, could you please forward blackburn my email with the syntax | 23:17 |
blackburn | what I suggest is | 23:17 |
blackburn | ModelSelectionParameters({'C': [0.1,0.2,0.3],'width': [1.0,5.0]}) | 23:18 |
blackburn | sth like that | 23:18 |
blackburn | this can be done in python natively | 23:19 |
@sonney2k | sure that would work | 23:19 |
@sonney2k | but you could avoid the modelsectionparameters | 23:19 |
@sonney2k | if this is done as typemap | 23:19 |
blackburn | how? | 23:19 |
@sonney2k | blackburn, the same way we can pass numpy arrays... | 23:20 |
gsomix | sonney2k, I cannot find this letter. O_o bad | 23:20 |
blackburn | pass to modelselection? | 23:20 |
@sonney2k | parse the python object - going through the dictionary/lists is not that much effort | 23:21 |
blackburn | well it is up to you :) | 23:21 |
@sonney2k | forwarded | 23:22 |
gsomix | sonney2k, thanks! | 23:23 |
blackburn | sonney2k: can't that be done via modelselectionparameters(the-dictionary)? | 23:25 |
@sonney2k | blackburn, sure but it can be done w/o the modelselectionparameters(the-dictionary) - just the-dictionary | 23:26 |
blackburn | I like the thing I suggested because it is like 1 md while typemapping can take a while | 23:27 |
@sonney2k | blackburn, as I said parsing dictionary / list is not so much the issue .. much more creating the object | 23:28 |
blackburn | ok | 23:28 |
blackburn | sonney2k: we have an issue with eigen3 thing - one puffin444's class header uses it | 23:31 |
@sonney2k | blackburn, hmmh can he avoid that? | 23:31 |
blackburn | I asked him he will do that | 23:31 |
blackburn | thanks to eigen authors we can easily map our matrices to eigen matrices | 23:32 |
@sonney2k | yeah that sounds good | 23:33 |
blackburn | sonney2k: do you know how can we use -isystem instead of -I? | 23:34 |
@sonney2k | blackburn, what does this mean? | 23:34 |
@sonney2k | -isystem??? | 23:34 |
blackburn | I mean we use pkg-config | 23:34 |
blackburn | -isystem is just -I but w/o warnings | 23:34 |
@sonney2k | blackburn, you mean the include path? | 23:34 |
blackburn | yes | 23:34 |
blackburn | we use pkg-config and it produces -I | 23:35 |
@sonney2k | blackburn, then simple sed magic - but do we want that? | 23:35 |
blackburn | yeah warnings are confusing | 23:35 |
blackburn | no idea if they will disappear at some point | 23:35 |
@sonney2k | blackburn, so the assumption would be - eigen3 warnings are useless and what they did is ok right? | 23:36 |
blackburn | yes sure, we trust it | 23:36 |
@sonney2k | gsomix, once you are done with the stuff you have on you todo - could you please try out parsing the dicts - I mean just to get an idea how difficult this is (w/o doing anything in shogun) | 23:37 |
gsomix | ok | 23:37 |
blackburn | sonney2k: can we unify preprocessors to support sparse somehow? | 23:50 |
@sonney2k | blackburn, I don't think so ... even substracting mean will turn a sparse solution into a non-sparse one | 23:58 |
@sonney2k | maybe for sparsity preserving operations? | 23:58 |
blackburn | sonney2k: hmm | 23:59 |
--- Log closed Sun Aug 19 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!