--- Log opened Fri Jun 29 00:00:17 2012 | ||
CIA-18 | shogun: Soeren Sonnenburg master * r11eba43 / src/shogun/optimization/lbfgs/lbfgs.h : drop doxygen mainpage from liblbfgs (clashes with shogun mainpage) - http://git.io/Xdrs2g | 00:17 |
---|---|---|
@sonney2k | blackburn, finally it is there: http://shogun-toolbox.org/doc/en/latest | 00:23 |
@sonney2k | but look at it http://shogun-toolbox.org/doc/en/latest/annotated.html | 00:23 |
@sonney2k | lots of new stuff is not documented... | 00:23 |
blackburn | cool | 00:24 |
blackburn | a lot of work | 00:24 |
blackburn | I like work :D | 00:24 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 240 seconds] | 00:25 | |
@sonney2k | blackburn, so then blackburn work for us :D | 00:25 |
alexlovesdata | goodnight | 00:48 |
alexlovesdata | I had a long chat with my PhDstudent Viktor | 00:48 |
blackburn | :D | 00:49 |
-!- alexlovesdata [55b21f1e@gateway/web/freenode/ip.85.178.31.30] has quit [Quit: Page closed] | 00:49 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 246 seconds] | 00:56 | |
-!- blackburn [~blackburn@31.28.43.76] has quit [Ping timeout: 248 seconds] | 02:17 | |
-!- naywhaya1e [~ryan@spoon.lugatgt.org] has joined #shogun | 03:45 | |
-!- naywhayare [~ryan@spoon.lugatgt.org] has quit [Ping timeout: 264 seconds] | 03:45 | |
-!- zxtx [~zv@205.209.7.111] has joined #shogun | 03:59 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 252 seconds] | 04:45 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 04:48 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 245 seconds] | 04:53 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 04:55 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 265 seconds] | 05:19 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 05:20 | |
-!- zxtx [~zv@205.209.7.111] has quit [Ping timeout: 246 seconds] | 05:43 | |
-!- sks [~sks@14.97.193.107] has joined #shogun | 07:01 | |
-!- gsomix [~gsomix@95.67.157.108] has joined #shogun | 07:17 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 264 seconds] | 07:56 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 07:56 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 248 seconds] | 08:05 | |
-!- sks [~sks@14.97.193.107] has quit [Ping timeout: 248 seconds] | 08:42 | |
-!- uricamic [~uricamic@2001:718:2:1634:a834:e48f:6ac0:beac] has joined #shogun | 08:50 | |
shogun-buildbot | build #1 of nightly_all is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/1 | 09:08 |
CIA-18 | shogun: Soeren Sonnenburg master * r2660a96 / (2 files): fix compile error when lapack is not available - http://git.io/i6wPkQ | 09:39 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 09:52 | |
shogun-buildbot | build #2 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/2 | 09:55 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Remote host closed the connection] | 10:16 | |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 10:16 | |
-!- zxtx [~zv@cpe-98-154-83-76.socal.res.rr.com] has joined #shogun | 10:33 | |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has joined #shogun | 10:33 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 10:38 | |
-!- zxtx [~zv@cpe-98-154-83-76.socal.res.rr.com] has quit [Ping timeout: 246 seconds] | 10:40 | |
-!- blackburn [~blackburn@31.28.43.76] has joined #shogun | 10:45 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 248 seconds] | 10:46 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 252 seconds] | 11:18 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 11:25 | |
-!- blackburn [~blackburn@31.28.43.76] has quit [Read error: Operation timed out] | 11:44 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 11:58 | |
-!- heiko [~heiko@host86-177-176-255.range86-177.btcentralplus.com] has joined #shogun | 12:00 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 250 seconds] | 12:02 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 12:07 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 246 seconds] | 12:12 | |
CIA-18 | shogun: puffin444 master * rc1e2e27 / (11 files in 5 dirs): Added Kernel Scaling in CInferenceMethod. Fixed some warnings. Fixed - http://git.io/OKg7PQ | 12:18 |
CIA-18 | shogun: Heiko Strathmann master * r047feef / (11 files in 5 dirs): Merge pull request #609 from puffin444/master - http://git.io/lXNwHQ | 12:18 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 12:31 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 12:45 | |
-!- blackburn [6da9d60d@gateway/web/freenode/ip.109.169.214.13] has joined #shogun | 12:50 | |
-!- pluskid [~pluskid@111.120.67.35] has joined #shogun | 12:55 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 246 seconds] | 13:14 | |
-!- blackburn [6da9d60d@gateway/web/freenode/ip.109.169.214.13] has quit [Ping timeout: 245 seconds] | 13:22 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 13:22 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 240 seconds] | 13:43 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 13:58 | |
-!- uricamic [~uricamic@2001:718:2:1634:a834:e48f:6ac0:beac] has quit [Quit: Leaving.] | 13:59 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 244 seconds] | 14:08 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 14:21 | |
-!- gsomix [~gsomix@95.67.157.108] has quit [Ping timeout: 255 seconds] | 14:24 | |
-!- gsomix [~gsomix@109.169.240.118] has joined #shogun | 14:28 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 246 seconds] | 14:34 | |
gsomix | sonney2k, around? | 14:39 |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 14:45 | |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has quit [Quit: ( www.nnscript.com :: NoNameScript 4.22 :: www.esnation.com )] | 14:47 | |
gsomix | I just want to ask, what preprocessors I should "to direct"? Dense, Sparse, String, Dim reduction? | 14:48 |
-!- romi_ [~mizobe@187.66.121.115] has quit [Ping timeout: 248 seconds] | 14:57 | |
-!- gsomix [~gsomix@109.169.240.118] has quit [Quit: Ex-Chat] | 15:05 | |
-!- gsomix [~gsomix@109.169.240.118] has joined #shogun | 15:05 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 15:10 | |
-!- pluskid [~pluskid@111.120.67.35] has quit [Ping timeout: 252 seconds] | 15:10 | |
-!- pluskid [~pluskid@202.130.113.141] has joined #shogun | 15:10 | |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has joined #shogun | 15:23 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Read error: Connection timed out] | 15:30 | |
-!- romi_ [~mizobe@187.66.121.115] has joined #shogun | 16:00 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 16:06 | |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun | 16:29 | |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun [] | 16:29 | |
@sonney2k | gsomix, did you finish dotfeatures yet? | 16:53 |
@sonney2k | otherwise just dense preprocs | 16:53 |
-!- blackburn [~blackburn@31.28.43.76] has joined #shogun | 16:58 | |
heiko | sonney2k, how can I activate debug messages from python? | 17:00 |
blackburn | any object.io.set_loglevel(1) | 17:03 |
heiko | thanks blackburn | 17:04 |
heiko | blackburn, are you still there? | 17:13 |
heiko | I have a weird thing | 17:13 |
blackburn | sure | 17:13 |
heiko | I have a feature object | 17:14 |
heiko | say CFeatures* p | 17:14 |
heiko | and I call p->get_num_vectors() | 17:14 |
heiko | why is that possible? | 17:14 |
heiko | the method is pure | 17:14 |
heiko | but it works | 17:14 |
blackburn | but what is the actual type? | 17:14 |
heiko | DenseFeatures | 17:14 |
heiko | the method is overwritten there | 17:15 |
blackburn | so what is the problem? | 17:15 |
heiko | but why can I call it from the CFeatures interface? | 17:15 |
heiko | I thought that wasnt possible | 17:15 |
blackburn | because it is virtual | 17:15 |
heiko | but its abstract | 17:15 |
heiko | I thought the compiler forbids that, in fact I have experienced that a few times | 17:15 |
heiko | and the linker complaining about undefined references | 17:16 |
blackburn | hmm strange I've never experienced problem with pure virtual methods | 17:16 |
heiko | mmh perhaps I have kept that wrong in memory | 17:17 |
blackburn | it looks fine to me - you just declare it in base class and put impl in derived ones | 17:18 |
heiko | I mean in C++ you always have that problem that a class cannot call its own pure virtual methods | 17:18 |
blackburn | CSGObject::get_name is pure virtual as well | 17:18 |
heiko | yes true | 17:18 |
heiko | mmg | 17:18 |
heiko | seems like, I have to read about that :D | 17:18 |
-!- pluskid [~pluskid@202.130.113.141] has quit [Quit: Leaving] | 17:36 | |
-!- romi_ [~mizobe@187.66.121.115] has quit [Remote host closed the connection] | 18:00 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 18:07 | |
@sonney2k | blackburn, can you please have a look why all sparse features examples in static interfaces crash: http://shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/9/steps/test/logs/stdio | 18:10 |
@sonney2k | heiko, blackburn that would be the 'new' buildbot http://shogun-toolbox.org/buildbot/waterfall | 18:11 |
@sonney2k | only debian 1-3 now | 18:11 |
@sonney2k | and nightly build | 18:11 |
@sonney2k | s | 18:11 |
heiko | sonney2k, nice! | 18:11 |
heiko | much more accessible :) | 18:11 |
@sonney2k | http://shogun-toolbox.org/doc/en/current/ is now the latest *released* documentation | 18:12 |
@sonney2k | and nightly we get an updated doc | 18:12 |
@sonney2k | created at | 18:12 |
@sonney2k | http://shogun-toolbox.org/doc/en/latest/ | 18:12 |
heiko | ah thats also good | 18:12 |
@sonney2k | it is run in the nightly_default process | 18:13 |
@sonney2k | heiko, so now we only need a few more machines for osx, cygwin | 18:14 |
heiko | yeh | 18:14 |
@sonney2k | and potentially other linuxes | 18:14 |
heiko | any ideas on the osx thing so far? | 18:14 |
heiko | and cygwin? | 18:14 |
@sonney2k | heiko, btw I am pretty impressed by the size of the build directory 1261504 kB for the whole build! | 18:15 |
heiko | hehe wow, | 18:15 |
heiko | but that probably are the modular interfaces right? | 18:15 |
heiko | how big is libshogun? | 18:15 |
blackburn | sonney2k: ok will take a look | 18:16 |
@sonney2k | heiko, well look at the buildbot :D | 18:16 |
@sonney2k | heiko, I already have a cygwin machine | 18:16 |
@sonney2k | tough to setup though but working on it | 18:16 |
heiko | cool | 18:16 |
@sonney2k | for osx I have absolutely no idea yet :`-( | 18:19 |
heiko | :( | 18:19 |
@sonney2k | so plan for now is to get cygwin running | 18:20 |
@sonney2k | and email notifications | 18:20 |
heiko | yeah step by step | 18:20 |
@sonney2k | I always got email notifications btw | 18:20 |
@sonney2k | so maybe I should at least put you: heiko, blackburn on the always recipient list for failures... | 18:21 |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has quit [Ping timeout: 245 seconds] | 18:21 | |
heiko | yes you can do that | 18:21 |
blackburn | I don't mind | 18:21 |
@sonney2k | for both of you your google addresses? | 18:22 |
heiko | yes | 18:22 |
blackburn | yeah | 18:22 |
@sonney2k | wiking, any news on your PR? | 18:23 |
@sonney2k | didn't you want to send it by yesterday night? | 18:23 |
@sonney2k | blackburn, all sparse stuff fails in modular too http://shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/12/steps/test/logs/stdio | 18:26 |
heiko | sonney2k, I would like to add a little scripts to the python_modular/tools folder | 18:27 |
heiko | but it is ignored by git | 18:27 |
heiko | should I add with -f? | 18:27 |
@sonney2k | heiko, is it in .gitignores? | 18:28 |
heiko | yes | 18:28 |
@sonney2k | heiko, what is its name? | 18:29 |
heiko | Id like to git add ../examples/undocumented/python_modular/tools/two_distributions_data.py | 18:29 |
heiko | but | 18:30 |
heiko | examples/undocumented/python_modular/tools is in gitignore | 18:30 |
@sonney2k | heiko, well it should not | 18:30 |
@sonney2k | so better fix git ignore | 18:30 |
heiko | kk | 18:30 |
@sonney2k | alright 30 C here | 18:30 |
@sonney2k | I better go swimming before the thunderstorm comes | 18:31 |
@sonney2k | shogun-buildbot, force build nightly_default | 18:32 |
shogun-buildbot | build #5 forced | 18:32 |
shogun-buildbot | I'll give a shout when the build finishes | 18:32 |
@sonney2k | shogun-buildbot, help | 18:32 |
shogun-buildbot | Get help on what? (try 'help <foo>', or 'commands' for a command list) | 18:32 |
@sonney2k | shogun-buildbot, commands | 18:32 |
shogun-buildbot | buildbot commands: commands, dance, destroy, force, hello, help, last, list, mute, notify, source, status, stop, unmute, version, watch | 18:32 |
@sonney2k | shogun-buildbot, list | 18:32 |
shogun-buildbot | try 'list builders' | 18:32 |
@sonney2k | shogun-buildbot, list builders | 18:32 |
shogun-buildbot | Configured builders: deb1 - libshogun deb2 - static_interfaces deb3 - modular_interfaces nightly_all nightly_default nightly_none | 18:32 |
@sonney2k | shogun-buildbot, help watch | 18:32 |
shogun-buildbot | Usage: watch <which> - announce the completion of an active build | 18:32 |
CIA-18 | shogun: Heiko Strathmann master * rccf6742 / (14 files in 6 dirs): Merge pull request #610 from karlnapf/master (+7 more commits...) - http://git.io/FAeKkQ | 18:36 |
-!- blackburn1 [~blackburn@31.28.43.76] has joined #shogun | 18:42 | |
puffin444 | Heiko, I have a question about the linear kernel | 18:42 |
heiko | pufin444, yes, hi | 18:42 |
puffin444 | It does not appear that there are any registered parameters for the LinearKernel. Is my observation correct? | 18:44 |
heiko | yes | 18:45 |
heiko | the two should be registered | 18:45 |
heiko | this is often forgotten :) | 18:45 |
puffin444 | Normal and normal_length? | 18:46 |
heiko | yes | 18:46 |
blackburn1 | too bad C++ have no annotations | 18:47 |
puffin444 | Do you think it makes sense to make them model selection parameters? I am not sure how a vector could be modified in the framework. | 18:48 |
puffin444 | Or at least the normal vector | 18:48 |
heiko | no I think they shouldnt be available | 18:49 |
heiko | As far as I understand it they are only used in the optimisation case and set from outside anyway | 18:50 |
blackburn1 | agree | 18:50 |
puffin444 | Okay. So I don't think I can add any kernel derivatives for the linear kernel then. | 18:50 |
heiko | but thats natural isnt it? | 18:50 |
heiko | for a linear kernel, what can you do? :) | 18:51 |
heiko | they is nothing to vary | 18:51 |
puffin444 | I think so. GPML has a linear kernel with ARD that has kernel derivatives available | 18:51 |
heiko | linear kernel with ARD? | 18:51 |
heiko | how does that work? | 18:51 |
heiko | for ARD you would need some kind of parameter for every dimension | 18:52 |
heiko | do you have a link? | 18:52 |
heiko | so I can read that? | 18:52 |
heiko | what are their kernel parameters? | 18:52 |
puffin444 | Yes its under covLINard.m at http://gaussianprocess.org/gpml/code/matlab/release/gpml-matlab-v3.1-2010-09-27.zip | 18:52 |
puffin444 | a parameter for each dimension as you said. | 18:53 |
blackburn1 | two kernel statisticians walk into a bar | 18:53 |
heiko | puffin44, I see, they bsically add a factor to each dimension | 18:54 |
heiko | you could add that to our linear kernel | 18:54 |
heiko | the kernel is then k(x,y)= x1*y1*alpha1+x2*y2*alpha2 .... | 18:55 |
heiko | blackburn and then? :) | 18:55 |
heiko | puffin444, I like the idea | 18:55 |
puffin444 | How would this fit into the model selection framework? Wouldn't the dimensionality have to be set at construction? | 18:55 |
puffin444 | If the dimension of the inputs change then somehow the modelselection parameters would have to be deleted and created over again. | 18:56 |
blackburn1 | heiko: and then your conversation :) | 18:56 |
heiko | but why should the dimension change when you use a linear kernel? | 18:57 |
heiko | or do you mean when you register one parameter per dimension? | 18:58 |
puffin444 | I'm not sure. Suppose you allocated a linear kernel for some data set, use it, call cleanup(), and then init() if for some other dataset with different dimensions. | 18:58 |
puffin444 | one parameter per dimension if I want to integrate it into hyperparameter learning. | 18:58 |
heiko | ah, ok, I would say, just register a vector with the weights per dimension | 18:58 |
heiko | then the parameter (also for modelselection) is *one* vector | 18:58 |
heiko | would that be a problem in the lerning methods you wrote? | 18:59 |
puffin444 | This might complicate the kernel derivatives, as right now they are accessed by name. | 18:59 |
heiko | you would have to treat all elements of the vector as single parameter | 18:59 |
heiko | cant you just access the whole vector and then take only the desired index? | 19:00 |
heiko | mmh the derivative of a linear kernel which is linear in its parameters | 19:00 |
heiko | wait, what would be the derivative of this kernel? | 19:01 |
puffin444 | The derivative would be respect to each dimension. | 19:02 |
puffin444 | K = -2*x(:,i)*x(:,i)'; | 19:02 |
heiko | yeah | 19:02 |
heiko | I mean, the parameter vanishes | 19:03 |
puffin444 | The issue is that in order to abstractly access kernel parameters, I current take the total number of registered model selection parameters and use their name to get their derivative. | 19:03 |
shogun-buildbot | Hey! build nightly_default #5 is complete: Failure [failed test] | 19:04 |
shogun-buildbot | Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/5 | 19:04 |
puffin444 | Thanks buildbot | 19:04 |
blackburn1 | shogun-buildbot: Hey! | 19:04 |
shogun-buildbot | What you say! | 19:04 |
heiko | damn that was my fault :) | 19:05 |
puffin444 | I am not sure how the indices of a vector parameter would fit it. | 19:05 |
puffin444 | The buildbot seems cheerful considering the build failed. | 19:05 |
heiko | puffin444, mmh, hard to tell for me currently | 19:07 |
heiko | I mean if you can access the vector itself | 19:07 |
heiko | and you can get the derivative with respect to all these parameters, then you would be fine right? | 19:07 |
heiko | the way they do it in the GP box is the same, they just compute the derivative wrt all dimensions | 19:08 |
blackburn | ping | 19:09 |
blackburn1 | pong | 19:09 |
blackburn1 | lol | 19:09 |
heiko | blackburn wtf ? :) | 19:09 |
heiko | have you split? | 19:09 |
blackburn1 | heiko: transferring data from old notebook to new | 19:09 |
blackburn1 | so work on both now | 19:09 |
heiko | make backups | 19:09 |
heiko | I recently deleted all my data :) | 19:10 |
heiko | when doing exactly that | 19:10 |
blackburn1 | that should be funny | 19:10 |
puffin444 | One possible, but perhaps awkward way to do it is have the derivative function in the kernel return a list of matrices instead of one. | 19:10 |
puffin444 | Wait forget it that would mess up the GradientSelection stuff. | 19:11 |
blackburn1 | oh I like having 8 gb of ram | 19:11 |
blackburn | bastard | 19:11 |
heiko | puffin444, no too much memory | 19:12 |
puffin444 | The problem is currently the framework treats each registered parameter as one dimension to be optimized. Access is by name to bypass possibly different indicies used for the same parameters in different data structures. | 19:12 |
heiko | yes I see | 19:12 |
CIA-18 | shogun: Heiko Strathmann master * r11c6ed7 / examples/undocumented/libshogun/statistics_quadratic_time_mmd.cpp : corrected compile error - http://git.io/j4Z0Qg | 19:13 |
CIA-18 | shogun: Heiko Strathmann master * rc54c124 / src/shogun/statistics/TestStatistic.h : added compute threshold dummy - http://git.io/r8xbww | 19:13 |
CIA-18 | shogun: Heiko Strathmann master * r029a375 / examples/undocumented/python_modular/statistics_linear_time_mmd.py : minor fixes - http://git.io/q5_ryQ | 19:13 |
CIA-18 | shogun: Heiko Strathmann master * r5e9cd50 / (3 files in 3 dirs): Merge pull request #611 from karlnapf/master - http://git.io/XfMBHg | 19:13 |
heiko | but puffin444, when the kernel matrix is derived wrt to say first dimension parameter | 19:14 |
puffin444 | Yes | 19:15 |
heiko | what is the difference when you derive wrt to second? | 19:15 |
heiko | no, this doesnt lead somewhere :( | 19:16 |
heiko | how hard would it be to extend the framework to also work on vectors? | 19:16 |
heiko | (and matrices) of parameters | 19:16 |
puffin444 | I really don't know. | 19:17 |
heiko | this problem arises with any ard right? | 19:17 |
-!- blackburn [~blackburn@31.28.43.76] has left #shogun [] | 19:17 | |
-!- blackburn1 is now known as blackburn | 19:18 | |
-!- blackburn [~blackburn@31.28.43.76] has left #shogun [] | 19:18 | |
puffin444 | Yes, and now that I look at Combined Kernel we have the same problem I think with computing derivatives. | 19:19 |
puffin444 | With the vector of subkernel weights | 19:20 |
heiko | yes I see | 19:20 |
heiko | mmh, we have to think about it some more | 19:20 |
heiko | I would currently say to extend the proceure | 19:21 |
heiko | you currently get parameters by name, but since you can access the TParameter construct you also got type information | 19:21 |
heiko | and using that you can make a distinction between CT_SCALAR, CT_VECTOR etc | 19:22 |
puffin444 | Okay. | 19:22 |
-!- blackburn [~blackburn@31.28.43.76] has joined #shogun | 19:22 | |
heiko | and then iterator over all indices | 19:22 |
heiko | then whats missing is how to get the derivative from the kernel | 19:22 |
heiko | since we only know the index we are currently at | 19:22 |
heiko | this could be solved by simply passing the index as parameter | 19:23 |
heiko | but then theres the problem of having a unified interface for derivatives in general | 19:23 |
puffin444 | Yeah, I could overload the function by adding one that takes both a name and index. | 19:23 |
heiko | yes that would work for vectors | 19:23 |
heiko | could also do it for matrices | 19:24 |
puffin444 | I call one of them based on the TParameter Type. | 19:24 |
heiko | and since these are the only container types shogun got | 19:24 |
heiko | there are only three possibilites | 19:24 |
blackburn | heiko: statistics_quadratic_time_mmd | 19:24 |
heiko | blackburn, already fixed | 19:24 |
blackburn | ah forgot to pull sorry | 19:24 |
heiko | puffin444, I think that would be nice | 19:25 |
blackburn | heiko: statistics_linear_time_mmd | 19:25 |
heiko | blackburn, damn :) | 19:25 |
puffin444 | Another issue is within the Sum and Combined Kernels. Suppose a Combined Kernel has a Gaussian Subkernel that has a width parameter. | 19:25 |
puffin444 | Forget it. If I pass the name to the subkernels it shouldn't be an issue. | 19:26 |
CIA-18 | shogun: Heiko Strathmann master * r11ba3d7 / examples/undocumented/python_modular/statistics_quadratic_time_mmd.py : corrected compile error - http://git.io/en8few | 19:28 |
CIA-18 | shogun: Heiko Strathmann master * r263031f / examples/undocumented/libshogun/statistics_linear_time_mmd.cpp : another example fix - http://git.io/f3f_Ow | 19:28 |
CIA-18 | shogun: Heiko Strathmann master * r166b510 / examples/undocumented/python_modular/statistics_quadratic_time_mmd.py : Wrong file. - http://git.io/8G2SSw | 19:28 |
CIA-18 | shogun: Heiko Strathmann master * rdb38a5d / examples/undocumented/libshogun/statistics_linear_time_mmd.cpp : Merge pull request #612 from karlnapf/master - http://git.io/RKkQcQ | 19:28 |
heiko | puffin444, ok | 19:28 |
heiko | puffin444, well this addition by vectors and matrices would make the framework even nicer :) | 19:29 |
heiko | and the problem with the number of dimensions (dont know how many parameters to register) is resolved on the fly | 19:29 |
puffin444 | This will take more time than expected though. | 19:30 |
heiko | shogun-buildbot faster! | 19:30 |
heiko | puffin444, yes, these framework things always take ages. I still remember the horror of these parameter trees last year :) | 19:30 |
-!- blackburn [~blackburn@31.28.43.76] has quit [Quit: Leaving.] | 19:41 | |
-!- blackburn [~blackburn@31.28.43.76] has joined #shogun | 19:42 | |
heiko | blackburn, is there a corresponding thing to #ifdef HAVE_LAPACK in python_modular | 19:46 |
blackburn | hmm no probably no | 19:47 |
heiko | how to proceed with that? tests will fail if i use the methods | 19:47 |
blackburn | I am curious how do we handle LDA then | 19:47 |
blackburn | heiko: any idea? :D | 19:53 |
heiko | blackburn, not really :D | 19:53 |
CIA-18 | shogun: Heiko Strathmann master * re9b94ce / (5 files in 3 dirs): Merge pull request #613 from karlnapf/master (+5 more commits...) - http://git.io/V7s_cA | 19:59 |
heiko | I just added the example | 19:59 |
heiko | blackburn I have an idea | 19:59 |
heiko | we could add a method to shogun class | 19:59 |
heiko | boolean have_lapack() | 19:59 |
heiko | which then returns true/false depending on #ifdef stuff | 19:59 |
blackburn | well you may handle this via import | 19:59 |
heiko | that would work | 19:59 |
heiko | then in python you can ask that | 20:00 |
blackburn | if import fails you have no your stuff available | 20:00 |
heiko | but still the example fails | 20:00 |
blackburn | no, why | 20:00 |
heiko | oh can one check for failing imports in pythong? | 20:00 |
heiko | python? | 20:00 |
blackburn | just catch exception and do nothing | 20:00 |
blackburn | yeah try except | 20:00 |
heiko | mmmh | 20:01 |
heiko | but in my case only one method is defined or not | 20:01 |
heiko | import would work | 20:01 |
blackburn | heiko: you can check if method is here | 20:02 |
heiko | "mymethod" in dir(dyn) | 20:02 |
heiko | nice I will add that | 20:03 |
CIA-18 | shogun: Heiko Strathmann master * r42dd334 / examples/undocumented/python_modular/statistics_quadratic_time_mmd.py : added check for method availability to avoid errors when lapack isnt installed - http://git.io/WAIA1g | 20:06 |
CIA-18 | shogun: Heiko Strathmann master * r9962e1d / examples/undocumented/python_modular/statistics_quadratic_time_mmd.py : Merge pull request #614 from karlnapf/master - http://git.io/LAVJ7Q | 20:06 |
heiko | blackburn, nice works | 20:07 |
heiko | I will leave now, see you later | 20:07 |
blackburn | heiko: see you | 20:07 |
puffin444 | hey Heiko | 20:12 |
-!- blackburn [~blackburn@31.28.43.76] has quit [Read error: Connection reset by peer] | 20:31 | |
-!- blackburn [~blackburn@31.28.62.192] has joined #shogun | 20:35 | |
-!- blackburn1 [~blackburn@81.28.187.0] has joined #shogun | 21:09 | |
-!- blackburn [~blackburn@31.28.62.192] has quit [Ping timeout: 265 seconds] | 21:09 | |
-!- heiko [~heiko@host86-177-176-255.range86-177.btcentralplus.com] has quit [Ping timeout: 264 seconds] | 21:19 | |
-!- gsomix [~gsomix@109.169.240.118] has quit [Quit: Ex-Chat] | 21:47 | |
-!- gsomix [~gsomix@109.169.240.118] has joined #shogun | 21:47 | |
-!- blackburn1 [~blackburn@81.28.187.0] has quit [Quit: Leaving.] | 22:08 | |
-!- blackburn [~blackburn@81.28.187.0] has joined #shogun | 22:34 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed] | 22:34 | |
@sonney2k | wiking, could you please give me an update? | 22:38 |
blackburn | sonney2k: hey have 5 mins? | 22:46 |
blackburn | gsomix asked me about add_to_dense_sgvec | 22:47 |
blackburn | I am a little confused with it as well | 22:47 |
@sonney2k | gsomix, why don't you ask me? | 22:52 |
@sonney2k | blackburn, sure | 22:52 |
blackburn | sonney2k: okay we both are little confused with SGVector parameter | 22:53 |
gsomix | sonney2k, blackburn is closer to me... :) | 22:53 |
blackburn | we copy memory here you know | 22:53 |
blackburn | how can it work at all? | 22:53 |
@sonney2k | blackburn, I don't understand the question | 22:53 |
blackburn | sonney2k: add_to_sgvec has SGVector parameter | 22:53 |
blackburn | argh | 22:54 |
blackburn | :D | 22:54 |
blackburn | lol nevermind | 22:54 |
@sonney2k | problem is you cannot do any operations with pointers | 22:54 |
@sonney2k | so you need some sgvector interface | 22:54 |
@sonney2k | gsomix, I see | 22:54 |
blackburn | I mixed up what is being added and where | 22:54 |
blackburn | gsomix: add_to_sgvec adds given vector (you map it from numpy) | 22:54 |
@sonney2k | so the idea is that this sg vector based function has to be overloaded | 22:54 |
blackburn | to ith one | 22:54 |
blackburn | sonney2k: I got it - that's ok | 22:55 |
@sonney2k | and the sg_vector based function will be called from the standard one | 22:55 |
@sonney2k | k | 22:55 |
blackburn | for some reason I had idea that something modifies given vector | 22:55 |
blackburn | so there would be a problem | 22:55 |
@sonney2k | gsomix, btw the other thing on the plan once all this is finished are the typemaps for python for easier model selection interface | 22:55 |
blackburn | sonney2k: I actually think we should stop at some point | 22:56 |
blackburn | not with >5 of director classes | 22:56 |
blackburn | once example is here it is easy to continue | 22:56 |
blackburn | so for example I'd rather let say n4nd0 do it by given example - he just knows his domain | 22:57 |
blackburn | it takes some time for gsomix to get into stuff he modifies | 22:57 |
blackburn | sonney2k: wait I was wrong | 22:58 |
blackburn | I am still confused | 22:58 |
blackburn | sonney2k: you input vec2 (SGVector) | 22:58 |
blackburn | and want to modify it | 22:58 |
blackburn | how can it go back if we typemap w/ copying | 22:58 |
gsomix | sonney2k, ok. | 23:02 |
@sonney2k | blackburn, it will be copied, then modified from python and then the modified thing is copied again and goes back to the add_to_dense_vec | 23:02 |
blackburn | sonney2k: uhh how does it work? | 23:02 |
@sonney2k | blackburn, as you would expect | 23:03 |
blackburn | does swig maps it back and forth? | 23:03 |
@sonney2k | via the in / out typemaps | 23:03 |
blackburn | I thought out typemaps are only for return types | 23:03 |
@sonney2k | one can even define new director in / out typemaps btw | 23:03 |
@sonney2k | they are | 23:03 |
@sonney2k | but you forget that this thing is going to python | 23:03 |
@sonney2k | and somehow has to go back to C | 23:03 |
@sonney2k | ++ | 23:03 |
@sonney2k | so C -> python -> C | 23:04 |
blackburn | okay then I expect director dot features a little slow | 23:04 |
@sonney2k | and arrows use the director stuff | 23:04 |
@sonney2k | same with kernel computation btw | 23:04 |
gsomix | Hhmm, and what about vec2 (SGVector)? Actually it's not numpy array, just pointer. I think there is a little problem, because vec2 is <Swig Object of type 'shogun::SGVector< double > *' at 0x1d069d0>. | 23:05 |
@sonney2k | we used some get_vector there - well only once | 23:05 |
@sonney2k | gsomix, which function are you talking about? | 23:05 |
gsomix | add_to_dense_sgvec | 23:06 |
@sonney2k | well it is a SGVector | 23:06 |
@sonney2k | what you can do is use its operations | 23:06 |
@sonney2k | like += | 23:06 |
@sonney2k | whatever | 23:06 |
@sonney2k | like in the example | 23:07 |
gsomix | But I thought he should to be transformed into an array. | 23:07 |
gsomix | *numpy array | 23:07 |
@sonney2k | hmmhh you are right | 23:08 |
gsomix | And this is numpy array. After +=, I mean. | 23:08 |
@sonney2k | maybe it is not after all | 23:08 |
@sonney2k | I wonder if this only happens when we use directorin / out typemaps | 23:08 |
gsomix | sonney2k, should I try use directorin/out in typemaps? | 23:11 |
@sonney2k | gsomix, we don't need them for Dotfeatures it seems | 23:11 |
@sonney2k | it is sufficient to have the += operator overloaded | 23:11 |
@sonney2k | hmmhh strong thunderstorm outside... | 23:11 |
gsomix | I'm afraid of thunderstorms... last was with a hurricane | 23:13 |
@sonney2k | nah not this one | 23:13 |
@sonney2k | gsomix, you might need to overload *= etc operators to get other dot feature operations to work | 23:14 |
gsomix | sonney2k, All right. I need to think a little and sleep. | 23:21 |
gsomix | sonney2k, see you tomorrow, ok? =___= | 23:21 |
gsomix | good night | 23:21 |
-!- gsomix [~gsomix@109.169.240.118] has quit [Ping timeout: 264 seconds] | 23:40 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 23:44 | |
n4nd0 | sonney2k: hi there! do you have a moment for a question? | 23:45 |
-!- alexlovesdata [55b21452@gateway/web/freenode/ip.85.178.20.82] has joined #shogun | 23:49 | |
n4nd0 | hey alexlovesdata, how are you doing? | 23:50 |
alexlovesdata | hey nando | 23:50 |
alexlovesdata | preparing to try out somethin for medical data ... but in one hour | 23:50 |
alexlovesdata | currently yet busy ;) | 23:50 |
n4nd0 | cool | 23:51 |
n4nd0 | alexlovesdata: can I ask you something about hmsvm? something about theory, nothing fancy I think | 23:52 |
alexlovesdata | ye, but maybe later | 23:53 |
n4nd0 | all right | 23:53 |
alexlovesdata | I am still helping a friend with an insurance | 23:53 |
-!- blackburn [~blackburn@81.28.187.0] has quit [Quit: Leaving.] | 23:58 | |
--- Log closed Sat Jun 30 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!