--- Log opened Mon Aug 13 00:00:17 2012 | ||
wiking | blackburn: yes | 00:18 |
---|---|---|
wiking | ubuntu's package doesn't ship with gmock-config :( | 00:19 |
wiking | so we need to do soemthing else in ./configure script to detect and add compile flags for gmock on ubuntu | 00:19 |
blackburn | what is the solution? install by hand? | 00:19 |
wiking | heiko solved it by installing gmock by hand | 00:19 |
wiking | but as i said in the mailing list | 00:19 |
wiking | that's not a really good solution ;) | 00:19 |
blackburn | I think it is rather ok - we (developers) can install it and only we do need it | 00:21 |
-!- blackburn [~blackburn@109.226.80.43] has quit [Quit: Leaving.] | 02:37 | |
shogun-buildbot | build #53 of nightly_none is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/53 | 03:01 |
-!- av3ngr [av3ngr@nat/redhat/x-xdzltetycxkppxhs] has joined #shogun | 04:04 | |
-!- n4nd0 [~nando@126.Red-2-137-7.dynamicIP.rima-tde.net] has joined #shogun | 08:47 | |
-!- uricamic [~uricamic@2001:718:2:1634:7db4:863b:e25:317c] has joined #shogun | 08:55 | |
-!- n4nd0 [~nando@126.Red-2-137-7.dynamicIP.rima-tde.net] has quit [Ping timeout: 256 seconds] | 10:00 | |
wiking | bu | 10:48 |
wiking | busy bees everywher? :) | 10:56 |
-!- av3ngr [av3ngr@nat/redhat/x-xdzltetycxkppxhs] has quit [Quit: That's all folks!] | 11:46 | |
wiking | anybody around? | 12:44 |
wiking | ahhahahahaha blackburn u've fcuked me :) | 12:47 |
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has joined #shogun | 13:53 | |
-!- blackburn1 [~blackburn@109.226.80.43] has joined #shogun | 14:01 | |
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has left #shogun [] | 14:02 | |
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has joined #shogun | 14:03 | |
blackburn1 | uricamic: hey | 14:05 |
blackburn1 | uricamic: can I merge your PR now? | 14:05 |
uricamic | blackburn1: hi, I hope so | 14:05 |
blackburn1 | uricamic: I mean did you work out things Soeren pointed? | 14:05 |
uricamic | yep | 14:06 |
blackburn1 | okay I see | 14:06 |
uricamic | everything should be fixed according to comments | 14:06 |
CIA-39 | shogun: Sergey Lisitsyn master * r8c987fd / (18 files in 4 dirs): Merge pull request #699 from uricamic/BM_SOL_EXAMPLE (+8 more commits...) - http://git.io/_XWA-g | 14:06 |
blackburn1 | uricamic: btw I had a question | 14:06 |
wiking | blackburn1: mergmeme :D | 14:06 |
uricamic | blackburn1: go on :) | 14:07 |
blackburn1 | wiking: yeah will glance it over once again in 5 mins | 14:07 |
blackburn1 | uricamic: in risk function there is float64_t* R | 14:07 |
blackburn1 | is it just a pointer to one float - risk value? | 14:07 |
uricamic | wiking: there is one important change in risk function you should be aware of | 14:07 |
uricamic | blackburn1: yep | 14:07 |
blackburn1 | uricamic: actually may be we could change it not to void but float64_t return type | 14:08 |
blackburn1 | to reduce num of parameters | 14:08 |
uricamic | you mean to directly return the value of risk from that function? | 14:09 |
blackburn1 | yeah | 14:09 |
uricamic | yep, that sounds reasonable | 14:09 |
blackburn1 | uricamic: that would make director risk function easier to use | 14:09 |
wiking | uricamic: ah cool then i'll wait for the merge | 14:09 |
wiking | i mean your merge... so i change mine as well | 14:09 |
blackburn1 | now one have to do R[0] = ... | 14:10 |
uricamic | blackburn1: ok, so I will change it now | 14:10 |
blackburn1 | heh I merged too fast | 14:10 |
uricamic | ok, next PR :) | 14:10 |
blackburn1 | uricamic: btw how fast it is for MNIST? | 14:10 |
shogun-buildbot | build #319 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/319 blamelist: Michal Uricar <uricar.michal@gmail.com> | 14:11 |
shogun-buildbot | build #320 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/320 | 14:12 |
uricamic | well, to be honest, the risk function could be written in much better way (some precomputations), but now it is like 1s per iteration on whole MNIST database (i.e. 60k examples, feature vector dimensionality 784) | 14:12 |
shogun-buildbot | build #329 of deb3 - modular_interfaces is complete: Failure [failed compile libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/329 blamelist: Michal Uricar <uricar.michal@gmail.com> | 14:13 |
blackburn1 | uricamic: how many iterations does it require to converge to some consistent solution? | 14:13 |
uricamic | comparing with exactly same implementation in MATLAB it is like 8x faster | 14:13 |
uricamic | depends on regularization constant lambda, but I think several hundreds | 14:13 |
blackburn1 | uricamic: hmm it sounds like it is not really slower than linear crammer-singer | 14:14 |
blackburn1 | interesting | 14:14 |
blackburn1 | I mean liblinear and libocas implementations (they take similar time on 40K examples/ 1K features dataset) | 14:15 |
uricamic | blackburn1: but I think it can be made faster with better risk function, doing some clever precomputations | 14:16 |
blackburn1 | I see | 14:16 |
shogun-buildbot | build #222 of bsd1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/222 blamelist: Michal Uricar <uricar.michal@gmail.com> | 14:16 |
blackburn1 | wiking: so shall I wait you to rebase against uricamic changes? | 14:18 |
shogun-buildbot | build #223 of bsd1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/223 | 14:20 |
wiking | blackburn1: let me try to rebase | 14:23 |
wiking | mm rebase went fine i'll try to compule | 14:24 |
wiking | *compile if that works then u can just merge it straight away | 14:24 |
wiking | hehe yeah no i need to fix some shit | 14:26 |
blackburn1 | ~150 warnings left | 14:26 |
blackburn1 | I will kill them all soon | 14:26 |
shogun-buildbot | build #330 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/330 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Michal Uricar <uricar.michal@gmail.com> | 14:30 |
wiking | blackburn1: what should be done about the libocas update? | 14:33 |
blackburn1 | wiking: I didn't check yet | 14:34 |
wiking | blackburn1: there was this thing with the multiclass svm solver of libocas... the -1 or no -1 | 14:34 |
wiking | i guess we shouldn't do -1 as in previous cases there was no -1 ;) | 14:34 |
wiking | previous cases = previous version of libocas | 14:35 |
* wiking still compiling... stupid modular interface ;) | 14:37 | |
wiking | ohshitfuck | 14:39 |
wiking | i hate this | 14:39 |
uricamic | Is it possible to get PyCharm license for shogun? | 14:39 |
blackburn1 | I had a crack somewhere | 14:40 |
blackburn1 | :D | 14:40 |
wiking | ok i'll redo this pull request | 14:41 |
uricamic | :) but it could be possible to obtain free license for shogun developers :) | 14:41 |
wiking | otherwise it'll be just a mess | 14:41 |
uricamic | which I guess is cleaner solution | 14:41 |
blackburn1 | wiking: redo?? | 14:41 |
blackburn1 | uricamic: yeah just apply on their site as shogun developer | 14:41 |
wiking | blackburn1: do a new PR | 14:42 |
wiking | blackburn1: that'll be nicer | 14:42 |
blackburn1 | okay | 14:42 |
uricamic | blackburn1: ok, and should I apply it just for me, I mean who should be the contact person? | 14:43 |
blackburn1 | uricamic: I think it should be you | 14:44 |
uricamic | ok, thx | 14:44 |
blackburn1 | Please make sure that you meet the following criteria: | 14:44 |
blackburn1 | You are the project lead or a committer | 14:44 |
blackburn1 | You have been working on your open source project for a minimum of 3 months | 14:44 |
blackburn1 | Your community is active. This means that you have recent activity in your newsgroups or forums | 14:44 |
blackburn1 | You have an updated News section on your site | 14:44 |
blackburn1 | You release updated builds on a regular basis | 14:44 |
blackburn1 | uricamic: ^ | 14:44 |
wiking | blackburn1: this should work https://github.com/shogun-toolbox/shogun/pull/709 | 14:44 |
uricamic | blackburn1: I know, I just didn't know if it should be like the leader of the project asks for license for all developers | 14:45 |
blackburn1 | uricamic: first point is not satisfied probably but just try | 14:45 |
blackburn1 | uricamic: I don't get that too | 14:45 |
uricamic | ok, I will just try | 14:46 |
blackburn1 | if they deny lets just ask soeren | 14:46 |
blackburn1 | :D | 14:46 |
uricamic | blackburn1: sure :) | 14:47 |
blackburn1 | however I am a commiter so I could try too | 14:47 |
wiking | ok this compiled again | 14:48 |
wiking | so all's good with that PR | 14:48 |
uricamic | blackburn1: ok, I have sent it, in few days there should be some response | 14:51 |
blackburn1 | uricamic: I've used pycharm for a while but didn't really like it - do you like it? | 14:52 |
uricamic | blackburn1: I don't know, I am a python newbie :D but so far it seems nice | 14:53 |
blackburn1 | uricamic: I see | 14:53 |
uricamic | and one my college, who is a big fan of python likes it a lot | 14:53 |
blackburn1 | uricamic: what do you use to develop with C++? | 14:54 |
uricamic | now I am using QtCreator | 14:54 |
uricamic | since it seems to cooperate with CMake nicely | 14:55 |
blackburn1 | I see | 14:55 |
wiking | blackburn1: :) | 15:01 |
CIA-39 | shogun: Sergey Lisitsyn master * rd87c378 / (13 files in 8 dirs): Fixed a few documentation warnings - http://git.io/_eBOfg | 15:06 |
blackburn1 | wiking: I am confused whether we should -1 in ocas | 15:22 |
shogun-buildbot | build #331 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/331 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 15:34 |
wiking | blackburn1: just writing the unittest | 15:37 |
wiking | and we'll see | 15:37 |
blackburn1 | this probably maps 0 to -1 | 15:38 |
wiking | lol shogun-unit-test(85744,0x7fff7c412180) malloc: *** error for object 0x7fb913c071b8: incorrect checksum for freed object - object was probably modified after being freed. | 15:42 |
CIA-39 | shogun: Sergey Lisitsyn master * r676cefa / (2 files): Fixed java modular compilation issue with BMRM - http://git.io/IbPXjQ | 15:51 |
wiking | blackburn1: mmmm in multiclass labelling the labels should be: 0,1,2 etc? | 15:51 |
blackburn1 | yes | 15:51 |
wiking | how fast multiclassocas is actually? | 15:58 |
wiking | suppose to be fast for a 3 class problem | 15:59 |
wiking | ? | 15:59 |
wiking | 3 guassians, each having 25 examples and the vector dim is 2 | 16:00 |
wiking | mmm this doesn't work :) | 16:05 |
wiking | anybody ever tested multiclassOCAS? | 16:07 |
wiking | i see a python exmaple of it :) | 16:07 |
blackburn1 | wiking: yes it works | 16:08 |
blackburn1 | wiking: slower than liblinear but for your case should converge in a few milliseconds I believe | 16:09 |
wiking | mmm i wonder what i do wrong | 16:15 |
wiking | as it really doesn't find anything | 16:15 |
wiking | i mean everything is mapped to 0.0 | 16:15 |
wiking | most probably my bad... | 16:17 |
shogun-buildbot | build #332 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/332 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 16:19 |
wiking | blackburn1: ok i've fixed the (c) header... see the last patch in PR | 16:22 |
blackburn1 | okay | 16:23 |
CIA-39 | shogun: Viktor Gal master * rc87bf5e / (12 files in 4 dirs): [WIP] restructure latent solvers - http://git.io/N0miwQ | 16:24 |
CIA-39 | shogun: Viktor Gal master * rbbb7d9b / (23 files in 8 dirs): Various changes for latent variable learning - http://git.io/FyaCDA | 16:24 |
CIA-39 | shogun: Viktor Gal master * rd3e9250 / src/interfaces/modular/Latent.i : Fix author in copyright header - http://git.io/ueEpLA | 16:24 |
CIA-39 | shogun: Sergey Lisitsyn master * r97b8217 / (27 files in 9 dirs): Merge pull request #709 from vigsterkr/latent - http://git.io/K4r67w | 16:24 |
wiking | yey | 16:24 |
wiking | let's see if it fails :) | 16:24 |
shogun-buildbot | build #323 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/323 blamelist: Viktor Gal <viktor.gal@maeth.com> | 16:26 |
wiking | yeeey | 16:26 |
wiking | :D | 16:26 |
wiking | ah i guess i know how | 16:26 |
wiking | *why | 16:26 |
shogun-buildbot | build #324 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/324 | 16:26 |
wiking | i've missed the --amend | 16:27 |
wiking | blackburn1: i've tried now with only 2 labels (0/1) using multiclass ocas, basically the same data set that works for svmocas | 16:30 |
wiking | and it fails | 16:30 |
wiking | :S | 16:30 |
blackburn1 | wiking: fails like? | 16:31 |
wiking | all example is classified to 0 | 16:32 |
wiking | so i have 2 gaussians | 16:33 |
wiking | i mean samples of 2 different gaussians | 16:33 |
blackburn1 | how many iterations? | 16:33 |
wiking | one is labeled as 0 the other as 1 | 16:33 |
wiking | blackburn1: haven't debugged it, just a sec i'll print it | 16:34 |
CIA-39 | shogun: Sergey Lisitsyn master * rdf80e5c / (15 files in 7 dirs): Removed SLEP machine and redundant L1LogisticRegression, rebased FeatureBlockLogisticRegression and fixed missed weight initialization - http://git.io/TV8FjA | 16:36 |
wiking | CMulticlassOCAS* mocas = new CMulticlassOCAS(C, train_feats, ground_truth); | 16:36 |
wiking | mocas->io->set_loglevel(MSG_DEBUG); | 16:36 |
wiking | mmm doesn't really work :( | 16:36 |
blackburn1 | it outputs information just after training | 16:37 |
blackburn1 | with debug level | 16:37 |
wiking | yes i know | 16:37 |
wiking | but if i do this | 16:37 |
wiking | and then call mocas->train() | 16:37 |
wiking | it doesn't print out anything :S | 16:37 |
blackburn1 | then something is broken in your example :D | 16:38 |
wiking | ?! | 16:38 |
blackburn1 | I do not undrestand how can it skip debug print | 16:40 |
wiking | ?![DEBUG] Number of iterations [nIter] = 5 | 16:41 |
wiking | heheh ok i've fixed it | 16:41 |
wiking | shogun wasn't init-ed in a right way | 16:41 |
wiking | [DEBUG] Number of training errors [trn_err] = 25 | 16:41 |
wiking | :S | 16:41 |
wiking | that is funny | 16:41 |
wiking | :)) | 16:41 |
wiking | out of 50 samples :))) | 16:41 |
wiking | where 25 is 0 and 25 is 1 | 16:42 |
wiking | :) | 16:42 |
blackburn1 | what is epsilon? | 16:42 |
wiking | default | 16:42 |
wiking | lemme check what is it | 16:43 |
blackburn1 | no, not really relevant | 16:43 |
blackburn1 | I mean if it was 1e-1 I would believe heh | 16:43 |
wiking | 1e-2 | 16:43 |
blackburn1 | okay then compare | 16:43 |
blackburn1 | with liblinear | 16:43 |
blackburn1 | I am totally sure liblinear works | 16:43 |
shogun-buildbot | build #333 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/333 blamelist: Viktor Gal <viktor.gal@maeth.com> | 16:44 |
wiking | optimization finished, #iter = 164 | 16:44 |
wiking | same error with liblienar | 16:45 |
blackburn1 | then something is wrong with your features and labels | 16:45 |
wiking | blackburn1: but it works with SVMOcas | 16:45 |
wiking | the same binary case | 16:45 |
wiking | :) | 16:45 |
wiking | perfectly | 16:46 |
blackburn1 | so you set labels to | 16:46 |
blackburn1 | 0 and 1 | 16:46 |
blackburn1 | right? | 16:46 |
wiking | yes | 16:46 |
wiking | instead of -1 and 1 | 16:46 |
wiking | everything is the same | 16:46 |
wiking | just the labels are 0 and 1 instead of -1 and 1 | 16:46 |
wiking | https://github.com/shogun-toolbox/shogun/blob/master/tests/unit/classifier/svm/SVMOcas_unittest.cc | 16:48 |
wiking | the only difference is | 16:48 |
wiking | line 29 | 16:48 |
wiking | line 37 | 16:48 |
wiking | and line 40 | 16:48 |
wiking | and obviously 42 | 16:49 |
wiking | everything else is the same | 16:49 |
wiking | even the random seed is the same | 16:49 |
wiking | thus the gaussians are the same | 16:49 |
blackburn1 | that's strange | 16:50 |
wiking | yep | 16:51 |
wiking | especially that not only libocas fails | 16:51 |
wiking | but liblinear | 16:51 |
blackburn1 | I just checked on my rsr stuff - liblinear converges to right solution | 16:51 |
wiking | so i really wonder | 16:51 |
blackburn1 | did you change it to multiclasslabels? | 16:51 |
wiking | http://pastebin.com/5VQTYtZp | 16:52 |
wiking | this is the full test code | 16:52 |
wiking | and yeah | 16:52 |
wiking | i'm using CMulticlassLabels | 16:52 |
blackburn1 | interesting | 16:54 |
blackburn1 | wiking: I have one suspect - copy subset.. | 16:57 |
wiking | blackburn1: but that works in case of SVMOcas :) | 16:57 |
blackburn1 | yes because svmocas would work with subset | 16:58 |
wiking | any fast way to divide the whole set then in another way? | 16:59 |
blackburn1 | I checked code and it should be ok | 16:59 |
blackburn1 | if (i % 2 == 0) train_idx[j] = i; else test_idx[j++] = i; | 17:00 |
blackburn1 | is it correct? | 17:00 |
shogun-buildbot | build #334 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/334 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Viktor Gal <viktor.gal@maeth.com> | 17:01 |
blackburn1 | wiking: why do you set half of labels and increase j only on test indices? | 17:02 |
wiking | noup | 17:03 |
wiking | as u can see j increases every 2nd time in the iter | 17:04 |
wiking | as you want to divide the whole matrix | 17:04 |
wiking | into 2 different sets | 17:04 |
blackburn1 | I don't know - that all sounds strange | 17:14 |
blackburn1 | I wouldn't believe something is wrong both in liblinear and ocas | 17:15 |
wiking | heheh me neither | 17:24 |
wiking | i guess something is wrong on my end | 17:24 |
wiking | it's just weird that it works with svmocas | 17:25 |
shogun-buildbot | build #335 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/335 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 17:26 |
-!- uricamic [~uricamic@2001:718:2:1634:7db4:863b:e25:317c] has quit [Quit: Leaving.] | 17:44 | |
-!- heiko [~heiko@host86-176-126-57.range86-176.btcentralplus.com] has joined #shogun | 17:47 | |
wiking | heiko: is there a fast way to test whether the features in a matrix are from 2 diff gauss distrib? | 17:50 |
heiko | yes :) | 17:51 |
heiko | t-test | 17:51 |
heiko | if you are one-dimensional | 17:51 |
heiko | and if not: My statistics framework :) | 17:51 |
heiko | wiking, what kind of data do you got? | 17:51 |
wiking | heiko: the SGMatrix that is created by that function i wrote in DataGenerator | 17:52 |
heiko | Ill take a look | 17:52 |
wiking | heiko: for some reason i'm having trouble using that and trying to learn a multiclass classifier with the generated data :S | 17:52 |
heiko | oh :) | 17:52 |
wiking | so now i'm trying to find out what could be the trouble :S | 17:52 |
heiko | maybe plot it for dim=2? | 17:52 |
heiko | you could also fit a Gaussian with ML | 17:53 |
heiko | wiking | 17:54 |
heiko | what do you want to create in the method i dont get the comments | 17:54 |
wiking | heiko: create a matrix | 17:55 |
heiko | what kind of gaussians should the samples come from | 17:55 |
wiking | where there are m samples for each of the n gaussians | 17:55 |
heiko | you only say that there are multiple ones | 17:55 |
heiko | what are their parameters? | 17:55 |
wiking | heiko: random | 17:55 |
wiking | heiko: you mean for mu, right? | 17:56 |
heiko | yeah | 17:56 |
heiko | random? | 17:56 |
wiking | yes that is randomly generated | 17:56 |
heiko | so random mean | 17:56 |
heiko | and random variance | 17:56 |
wiking | variance is a diagonal | 17:56 |
heiko | and from each of these a fixed number of samples? | 17:56 |
wiking | eigenmatrix | 17:56 |
wiking | [1,0;0,1] | 17:56 |
heiko | do all of them have the same probability to generate a sample or are they weighted? | 17:56 |
wiking | if dim = 2 | 17:56 |
heiko | so variance is not random | 17:57 |
wiking | yep that's not random | 17:57 |
heiko | I suggest the following: first create a method (or check whether it exists): gaussian_sample(SGVector mean, SGMatrix variance) | 17:58 |
heiko | and then just call that with random means | 17:58 |
wiking | heiko: well that's more or less what's happening ;0 | 17:58 |
wiking | :) | 17:58 |
heiko | ah thats the Gaussian class right? | 17:58 |
wiking | yes | 17:58 |
wiking | Gaussian class generates the samples for the given gaussian | 17:58 |
wiking | ok maybe the problem is not linearly separable? :) | 17:59 |
heiko | btw you dont need to call this get_allocated_matrix stuff if you dont have a target matrix as parameter | 17:59 |
wiking | i mean the mean is just not good | 17:59 |
heiko | gaussians mixtures are per definition not linearly separable | 18:00 |
heiko | you might be lucky but not in general | 18:00 |
heiko | you need a large distance at least e*std-dev | 18:00 |
wiking | heiko i guess i got lucky in case of svmocas then | 18:00 |
heiko | 3*stddev | 18:00 |
heiko | I looked at the method | 18:01 |
heiko | this mean.random() | 18:01 |
heiko | breaks it | 18:01 |
heiko | you need to ensure that the means are far enough apart from each other | 18:01 |
wiking | yep | 18:01 |
wiking | htat's what i thought now | 18:01 |
wiking | any ideas | 18:01 |
wiking | how can i generate good enough means if i know the dim and the number of gaussians i have | 18:02 |
wiking | since i guess all of them should be equally far apart from each other | 18:02 |
heiko | yeah, dont use random means | 18:02 |
heiko | the samples are random anyway | 18:02 |
blackburn1 | one must be careful with 0.01 probability left with 3 sigmas | 18:03 |
blackburn1 | :D | 18:03 |
heiko | or you could order them in a grid | 18:03 |
heiko | thanks blackburn :) | 18:03 |
blackburn1 | heiko: do I remember that correctly? 0.995 is percentile or so? | 18:04 |
heiko | yeah :) | 18:04 |
heiko | wiking, just create hypercubes of gaussians | 18:04 |
heiko | Ill send you a picture | 18:04 |
wiking | heiko: great | 18:04 |
blackburn1 | 4d cube | 18:04 |
blackburn1 | :D | 18:04 |
wiking | heiko: so depending on the dim | 18:04 |
wiking | i take the 'corners' of the hypercube? | 18:05 |
wiking | i.e. if dim = 2 | 18:05 |
blackburn1 | just generate all combinations of vector of size dim consisting +-d | 18:06 |
wiking | 0,0; 0,1; 1,0; 11 | 18:06 |
wiking | *1,1 | 18:06 |
wiking | ? | 18:06 |
blackburn1 | or (-1,1) (1,1) (1,-1) (-1,-1) | 18:06 |
heiko | wiking, sent | 18:06 |
heiko | wiking, dont need the corners | 18:06 |
heiko | just do n gaussians per dim | 18:06 |
heiko | and set distance between them to a large value | 18:07 |
heiko | the picture I sent is for d=2 | 18:07 |
wiking | ah ok | 18:07 |
wiking | yeah because i was about to say what if one wants 10 gaussians but with d=2 | 18:07 |
wiking | ok so what do you reckon the distance should be? | 18:07 |
heiko | 5 should be enough | 18:08 |
blackburn1 | heiko: I found a way to evaluate OvR things | 18:08 |
heiko | but dimension wise | 18:08 |
heiko | not euclidean | 18:08 |
heiko | blackburn, tell me | 18:08 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 18:08 | |
wiking | heiko: i.e. 0,0; 0,5; 5,0; 5,5; 10,0 etc | 18:08 |
wiking | right? | 18:08 |
heiko | yeah for d=2 | 18:08 |
wiking | for d=3 5 would be good i.e. 0,0,0; 0,5,0 etc? | 18:09 |
blackburn1 | heiko: I am adding an options to store all scores in multiclass labels | 18:09 |
heiko | dim=3 would be a 3d cube " filled with" gaussians | 18:09 |
blackburn1 | the multiclassovrevaluation computes things via binaryclassevaluation | 18:09 |
heiko | great | 18:09 |
heiko | ok | 18:10 |
wiking | damn i need a grid generator code | 18:20 |
heiko | blackburn1, custom fonts with python latex are a pain is the a** | 18:23 |
heiko | computer modern is quite easy | 18:23 |
heiko | what do you think? | 18:23 |
blackburn1 | heiko: python latex? | 18:23 |
heiko | yeah, latex typo in pytohn plots | 18:23 |
heiko | for tutorial | 18:23 |
blackburn1 | matplotlib? | 18:23 |
heiko | yeah | 18:23 |
blackburn1 | I used custom fonts there | 18:24 |
blackburn1 | without much pain | 18:24 |
heiko | could you send me a script with that? | 18:24 |
blackburn1 | pain is to use cyrillic | 18:24 |
blackburn1 | :D | 18:24 |
heiko | like a small example where you use the pathpazo | 18:24 |
blackburn1 | you would never imagine how much pain cyrillic brings | 18:24 |
heiko | lol :) ok I stop complaining | 18:24 |
blackburn1 | heh | 18:24 |
blackburn1 | okay let me find something | 18:24 |
blackburn1 | heiko: all you need is to put a few things into rc | 18:25 |
heiko | ok | 18:25 |
heiko | do you have a script? | 18:25 |
blackburn1 | wait a min | 18:26 |
wiking | heiko: this shoudl work afaik | 18:26 |
wiking | - mean.random(0.0, 10.0); | 18:26 |
wiking | + mean.zero(); | 18:26 |
wiking | + for (index_t k = 0; k < dim; ++k) | 18:26 |
wiking | + mean[k] = i*grid_distance; | 18:26 |
wiking | or? | 18:26 |
wiking | where i is the i-th gaussian | 18:26 |
wiking | so basically it's 0,0... 5,5... 10,10... | 18:26 |
wiking | etc | 18:26 |
blackburn1 | heiko: https://gist.github.com/3342349 | 18:27 |
blackburn1 | lines 83,84,85 | 18:27 |
heiko | wiking, plot in 2d and 3d then youll see :) | 18:27 |
blackburn1 | for 4d case use the force! | 18:27 |
blackburn1 | :D | 18:27 |
wiking | heiko: display_matrix is not working :D | 18:27 |
blackburn1 | heiko: but show won't work | 18:28 |
wiking | or it does? :) | 18:28 |
heiko | it does | 18:28 |
wiking | oh yeah it does i'm just looking at it on a wrong way | 18:28 |
blackburn1 | heiko: show doesn't work here with latex at all | 18:28 |
blackburn1 | heiko: neither with custom nor default latex fonts | 18:28 |
heiko | blackburn, this is all annoying | 18:28 |
heiko | why isnt there an export button? | 18:28 |
blackburn1 | heiko: export? | 18:29 |
heiko | i hate this messy stuff! argh | 18:29 |
heiko | save-as you know | 18:29 |
heiko | its 2012 | 18:29 |
blackburn1 | heiko: it is in matplotlib | 18:29 |
heiko | but it doesnt integrate into latex | 18:29 |
heiko | It would be great to let latex do the typo | 18:29 |
heiko | then everything would look much nicer | 18:29 |
blackburn1 | ehmm | 18:30 |
blackburn1 | heiko: in my script latex is used with custom font and saves to .pdf | 18:30 |
blackburn1 | what else do you need? :) | 18:30 |
heiko | yeah this works, but is it comfortable? :) | 18:30 |
heiko | thats my point | 18:30 |
heiko | just complaining in general :D | 18:30 |
blackburn1 | comfortable in which way? | 18:30 |
blackburn1 | I think there is a way to make show work too | 18:31 |
blackburn1 | with latex | 18:31 |
blackburn1 | heiko: nicholas wants to normalize train and test data with preprocessor inited with train data | 18:34 |
heiko | blackburn1, what does that mean? | 18:34 |
blackburn1 | heiko: for example you compute mean of train vectors | 18:35 |
blackburn1 | and subtract it both from train and test vector | 18:35 |
blackburn1 | s | 18:35 |
blackburn1 | during the process of evaluation | 18:35 |
heiko | ok | 18:35 |
blackburn1 | got it? | 18:35 |
heiko | so he wants no normalise training data for each fold? | 18:35 |
blackburn1 | yes | 18:35 |
heiko | puh | 18:36 |
blackburn1 | heiko: it is next on my list | 18:36 |
heiko | preprocessors cannot do that? | 18:36 |
heiko | ah ok | 18:36 |
blackburn1 | heiko: they can | 18:36 |
blackburn1 | but we do not handle it | 18:36 |
blackburn1 | heiko: we should call init in cross validation I think | 18:36 |
blackburn1 | before training | 18:37 |
heiko | yeah | 18:37 |
heiko | I forgot that | 18:37 |
heiko | before | 18:37 |
heiko | its very important for some kernels | 18:37 |
blackburn1 | you could do that too if you have time | 18:37 |
heiko | maybe | 18:37 |
blackburn1 | while I am messing with ovr multiclass :) | 18:37 |
heiko | I am currently getting organised with the rest for my gsoc | 18:37 |
heiko | this sigmoid fitting still has to come | 18:37 |
heiko | and mkl weights per fold output | 18:37 |
blackburn1 | heiko: I spent some time on pre-release stuff already btw | 18:38 |
blackburn1 | there are ~100 doc warnings left and I checked tests | 18:38 |
heiko | nice, yeah I also fixed a few warning on the weekend | 18:39 |
heiko | then there is still the migration stuff | 18:39 |
heiko | and reference counts for sgvec/mat | 18:39 |
blackburn1 | we need to release on 1st of September | 18:39 |
blackburn1 | I like that date :D | 18:40 |
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has joined #shogun | 18:42 | |
wiking | fucking hell how do you plot this shit in matlab fast? | 18:42 |
wiking | or python | 18:42 |
wiking | i have a matrix of 2 rows and n columns | 18:42 |
blackburn1 | from pylab import * | 18:43 |
blackburn1 | X = array([yourmatrix]) | 18:43 |
blackburn1 | scatter(X[0,:],X[1,:]) | 18:43 |
blackburn1 | show() | 18:43 |
wiking | mmm | 18:48 |
wiking | wy this shit only shows 1 points :))0 | 18:48 |
blackburn1 | I expect your X is transposed then | 18:51 |
wiking | noup :) | 18:51 |
wiking | i'm switching to python | 18:51 |
blackburn1 | print X.shape | 18:51 |
wiking | i've tried scatter in matlab | 18:51 |
wiking | should do the same | 18:51 |
wiking | installing scipy.. forgot to reinstall it after os upgrade :X | 18:52 |
wiking | i should be able to copypaste the output of display_matrix into python :S | 18:55 |
wiking | heiko: SHIIITFACK | 19:01 |
wiking | :) | 19:01 |
wiking | why it no working:) | 19:01 |
heiko | hey wiking, calm down | 19:01 |
heiko | whats happening? | 19:01 |
wiking | :) | 19:02 |
heiko | one has to patient with plotting | 19:02 |
wiking | i've plotted | 19:02 |
wiking | but i think it should be ok | 19:02 |
wiking | as a lot of them are actually 0,0 | 19:02 |
wiking | i mean for the mean =[0,0] | 19:02 |
wiking | the samples are (0,0) | 19:02 |
wiking | thus when i plot i don't see it that there's equal amount there | 19:03 |
wiking | as around the other grid point | 19:03 |
wiking | but as far as i can see from the matrix | 19:03 |
wiking | it should be alright | 19:03 |
wiking | still | 19:03 |
wiking | it cannot separate it | 19:03 |
heiko | it also has to be ok from the plot | 19:05 |
heiko | matrix reading is left for keanu reeves :D | 19:05 |
wiking | ok | 19:07 |
wiking | this is weird | 19:07 |
wiking | let me pastebin it | 19:07 |
heiko | ok | 19:07 |
wiking | this is the matrix | 19:07 |
wiking | http://pastebin.com/1ykKNze5 | 19:07 |
wiking | d=2 | 19:07 |
wiking | and 2 gaussians | 19:07 |
wiking | where mean_1=(5,5) | 19:07 |
wiking | mean_2=(10,10) | 19:08 |
wiking | ah no i'm sorry | 19:08 |
wiking | mean_1= (15,15) and mean_2=(30,30) | 19:09 |
wiking | and variance is always [1,0;0,1] | 19:09 |
heiko | hey man I cant read that | 19:09 |
heiko | need images | 19:10 |
wiking | ok here's the image... (sending in email) | 19:10 |
wiking | sent | 19:12 |
heiko | I see | 19:13 |
heiko | but what happened to the hypercubes? | 19:13 |
wiking | ? | 19:13 |
wiking | it's just dim=2 | 19:13 |
wiking | so no hypercube | 19:13 |
wiking | damn | 19:14 |
wiking | i think the problem is the way i set the matrix elements | 19:14 |
heiko | but in two dimension you should get this grid in the image i sent you | 19:19 |
wiking | yep | 19:20 |
heiko | sorry phone was ringing | 19:20 |
wiking | the problem is the filling of the matrix | 19:20 |
wiking | i've printed separately the generated samples | 19:20 |
wiking | and it's good | 19:20 |
wiking | this is with 2 gaussians with 2 dim | 19:20 |
wiking | http://pastebin.com/xM2VfBKh | 19:20 |
wiking | the problem is that how i fill up the matrix | 19:21 |
heiko | you are adding the value to each dimension | 19:25 |
heiko | thats why you get this diagonal structure | 19:25 |
heiko | why not make this grid? | 19:26 |
wiking | heiko: yeah i get that diagonal struct | 19:26 |
wiking | but mainly that's not the problem | 19:26 |
wiking | i mean it should work with this diag struct as well | 19:26 |
wiking | or? | 19:26 |
wiking | but the problem is that how i fill up the matrix with these vectors | 19:27 |
wiking | i think it was a miracle that it was working with SVMOcas at the first place | 19:27 |
heiko | yes it should work | 19:27 |
heiko | really, oh no :) | 19:27 |
wiking | yeah | 19:27 |
heiko | isnt there a reference dataset that you could use? | 19:28 |
wiking | ok | 19:28 |
wiking | i have data! | 19:29 |
wiking | and it actually looks good | 19:29 |
wiking | although multiclass couldn't separate it still | 19:29 |
wiking | heiko: sent u an email | 19:31 |
wiking | with a graph where the means of the gaussians: 15,15 and 30,30 | 19:31 |
heiko | wiking these should be fine | 19:31 |
wiking | multiclass cannot separate it | 19:31 |
wiking | :D | 19:31 |
wiking | neither multiclass liblinear | 19:33 |
wiking | nor multiclass ocas | 19:33 |
heiko | are you sure you did not switch dimensions or something? | 19:33 |
wiking | well now i'm going to print out the matrices separately | 19:34 |
wiking | after i devide this big matrix of samples | 19:34 |
wiking | into train and test dataset | 19:34 |
CIA-39 | shogun: Sergey Lisitsyn master * rc49b8cf / (8 files in 5 dirs): Added multiclass OVR evaluation - http://git.io/KzIHsw | 19:35 |
wiking | the plot is very similar obviously | 19:35 |
wiking | :))) | 19:35 |
wiking | i'll print out the labeling as well next to the vector values | 19:36 |
wiking | as i really cannot see wtf is happening | 19:36 |
heiko | blackburn, matplotlib is so bad I can't believe it | 19:39 |
heiko | Now 1h trying to get along with these plots | 19:40 |
blackburn1 | heiko: what's up? | 19:40 |
blackburn1 | what is so wrong with it? | 19:40 |
heiko | everything is so complicated | 19:40 |
wiking | hah | 19:41 |
wiking | i'm right | 19:41 |
blackburn1 | heiko: yes it is very flexible so complex | 19:41 |
wiking | heiko: http://pastebin.com/MM4pzW11 | 19:41 |
wiking | this is the train set | 19:41 |
wiking | and the test set is very similar | 19:42 |
heiko | blackburn, how to change the image size | 19:42 |
heiko | I fail to do so | 19:42 |
heiko | wiking, and? | 19:42 |
wiking | heiko: and everything is classified after the training as 0 | 19:42 |
blackburn1 | heiko: image size? | 19:43 |
blackburn1 | size of what exactly? | 19:43 |
wiking | or no sorry | 19:43 |
wiking | no everything is classified as 1 | 19:43 |
heiko | blackburn, something in order to have a latex compatible size | 19:43 |
blackburn1 | heiko: why to do that? | 19:43 |
heiko | because scaling looks bad | 19:44 |
wiking | heiko: and basically what it does according to the debug that it takes the 0 examples as training errors | 19:44 |
heiko | thats why I want to produce plots that fit without scaling | 19:44 |
blackburn1 | heiko: scaling of fonts? | 19:44 |
heiko | blackburn1, yes | 19:44 |
heiko | wiking, so theres an error in multiclass? | 19:45 |
blackburn1 | heiko: okay I'd suggest to tune fontsize | 19:45 |
wiking | heiko: i have no idea... neither multiclassocas nor multiclassliblinear can separate this problem | 19:45 |
heiko | blackburn, no Just figure size, I set the font size in the figure to the same as in tutorial, which is 11 | 19:45 |
heiko | wiking, try libsvm | 19:46 |
blackburn1 | heiko: http://stackoverflow.com/questions/332289/how-do-you-change-the-size-of-figures-drawn-with-matplotlib | 19:46 |
heiko | without multiclass | 19:46 |
wiking | heiko: svmocas works smoothly | 19:46 |
heiko | blackburn you joker, I have tried the first 10 google entries | 19:46 |
blackburn1 | heiko: and? | 19:46 |
heiko | they dont work thats the problem | 19:46 |
blackburn1 | no way, why? | 19:46 |
wiking | heiko: w/o any missclassification | 19:46 |
wiking | heiko: w/o any missclassification | 19:46 |
wiking | sorry | 19:46 |
wiking | wrong window | 19:47 |
wiking | :) | 19:47 |
heiko | blackburn, and I refuse to type in more that 10 lines for that problem btw ::D | 19:47 |
blackburn1 | heiko: so rc('figure', figsize=(5, 10)) doesn't work? | 19:47 |
heiko | it looks like, but then when I save, things are discarded | 19:48 |
heiko | and when I call savefig | 19:48 |
heiko | the resulting pdf is empty | 19:48 |
heiko | and thats when I started complaining :) | 19:48 |
blackburn1 | hmm it doesn't work okay | 19:49 |
wiking | i've used not liblinear instead of libsvm | 19:49 |
wiking | WARN] reaching max number of iterations | 19:49 |
wiking | Using -s 2 may be faster(also see liblinear FAQ) | 19:49 |
blackburn1 | heiko: | 19:50 |
blackburn1 | figure(figsize=(size in inches, size in inches)) | 19:50 |
blackburn1 | just do that | 19:50 |
blackburn1 | it works here | 19:50 |
wiking | comeeooon wtf is with this :) | 19:50 |
-!- blackburn1 [~blackburn@109.226.80.43] has left #shogun [] | 19:50 | |
* wiking doesnt get it! | 19:50 | |
heiko | wiking | 19:50 |
heiko | you get this warning when you should optimise primal instead of dual | 19:50 |
heiko | try libsvm | 19:51 |
-!- blackburn1 [~blackburn@109.226.80.43] has joined #shogun | 19:51 | |
heiko | blackburn, I just figured out the problem | 19:51 |
wiking | heiko: with linear kernel? | 19:51 |
heiko | wiking yes | 19:51 |
heiko | and not multiclass | 19:51 |
blackburn1 | arghh I got why I have headache | 19:51 |
blackburn1 | I forgot to eat today | 19:51 |
blackburn1 | heiko: so now it works? | 19:53 |
heiko | blackburn, not really, still looks horrible, this causes me a headache | 19:53 |
heiko | font size is wrong and stuff | 19:53 |
heiko | since this scales things | 19:54 |
shogun-buildbot | build #336 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/336 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 19:54 |
heiko | its so stupid, since it is an easy problem that every scientist has to approach | 19:54 |
blackburn1 | heiko: font size of what? | 19:54 |
blackburn1 | axes? | 19:54 |
heiko | I dont get why there are no good solutions | 19:54 |
heiko | no all | 19:54 |
blackburn1 | wait but is size of plot correct? | 19:55 |
wiking | heiko: works just fine | 19:56 |
wiking | heiko: as with liblinear and svmocas | 19:56 |
wiking | when it's binary | 19:56 |
CIA-39 | shogun: Sergey Lisitsyn master * ra6cdaf5 / src/shogun/evaluation/CrossValidation.cpp : Added preprocessor init calls in crossvalidation - http://git.io/gzmNOw | 19:57 |
wiking | binary classification just works fine | 19:58 |
wiking | heiko: i've tried changing dimension | 20:01 |
wiking | but that did not confuse at all the binary classifiers | 20:01 |
heiko | then there is a bug somewhere | 20:01 |
wiking | heiko: in case of multiclass if i put higher dimension | 20:03 |
wiking | then the error gets smaller | 20:03 |
wiking | but still nothing near to the binary case | 20:03 |
wiking | i just need a confirmation | 20:09 |
wiking | ok i'm gonna feed this train/test pair | 20:11 |
wiking | directly to liblinear | 20:11 |
wiking | just to see wtf | 20:11 |
heiko | wiking I dont know | 20:12 |
heiko | it should give the same answer when you hand in the same (easy) data | 20:13 |
wiking | heiko: heheh let's see i'm very very curious now | 20:13 |
wiking | Accuracy = 36% (27/75) | 20:16 |
wiking | :) | 20:16 |
wiking | if i give the train set as a test set | 20:16 |
shogun-buildbot | build #337 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/337 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 20:19 |
wiking | hehe | 20:20 |
wiking | ok this si just not separable | 20:20 |
wiking | :))) | 20:20 |
wiking | heiko: any ideas how we could solve this that there's a datagenerator function that creates a dataset that is separable in a way that i can do a unittest for it? | 20:21 |
heiko | do the unit test on a dataset that is definately lin. separable | 20:21 |
heiko | you could use the mean dataset | 20:22 |
heiko | but actually | 20:22 |
heiko | I would use a fixed one | 20:22 |
heiko | so that you can assert outputs | 20:22 |
wiking | heiko: heheh so add a header file that has the dataset? | 20:22 |
heiko | no | 20:23 |
wiking | and one could include it when he wants to test a classifier? | 20:23 |
heiko | just produce it in your code | 20:23 |
wiking | heiko: ok how ?:) | 20:23 |
wiking | as the gaussian doesn't work | 20:23 |
wiking | :) | 20:23 |
heiko | wiking, just via signs or so | 20:23 |
wiking | :D | 20:28 |
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has joined #shogun | 20:31 | |
n4nd0 | blackburn1: hey! | 20:31 |
blackburn1 | n4nd0: hey there | 20:31 |
n4nd0 | blackburn1: how are you? | 20:31 |
wiking | !@#$ :) | 20:31 |
blackburn1 | n4nd0: fine, watching wiking playing around with gaussians :D | 20:32 |
n4nd0 | haha | 20:32 |
n4nd0 | wiking: how is it going with them? | 20:32 |
n4nd0 | hehe | 20:32 |
wiking | n4nd0: shitfuck | 20:32 |
wiking | wasted half a day | 20:32 |
wiking | and it's not working at all | 20:32 |
heiko | blackburn, I got it finally with the plots | 20:32 |
n4nd0 | hehe | 20:32 |
n4nd0 | wiking: I wish you luck! | 20:33 |
heiko | n4nd0, you were also interested in that right? | 20:33 |
n4nd0 | blackburn1: I am going to rebase the branch now | 20:33 |
n4nd0 | heiko: yes | 20:33 |
n4nd0 | blackburn1: does that mean that we'll be ready to merge afterwards?? | 20:33 |
n4nd0 | I am really looking forward to it! | 20:33 |
blackburn1 | heiko: yeah I think so | 20:33 |
n4nd0 | blackburn1: was that last message regarding my branch? | 20:35 |
blackburn1 | about rebase? yeah | 20:36 |
blackburn1 | wow wiking you did weekly report in time :) | 20:36 |
wiking | :D | 20:37 |
wiking | it was already in draft | 20:37 |
wiking | but the the hell with the multiclassocas unittesting | 20:38 |
n4nd0 | wiking: was the change CStructuredData -> CData finally merged? | 20:40 |
n4nd0 | I took a look to your latent branch but didn't see it | 20:40 |
wiking | n4nd0: i haven't changed your code | 20:41 |
wiking | n4nd0: just did the latent part | 20:41 |
wiking | n4nd0: and introduced lib/CData... | 20:41 |
@sonney2k | heiko, did you discuss the 'e.V.' issue? | 20:41 |
blackburn1 | master sonney2k in da hous | 20:41 |
blackburn1 | e | 20:41 |
@sonney2k | heiko, about now we have to start to organize formalities w/ google | 20:42 |
heiko | sonney2k, not really, but I got a friend who studies law visiting, he hinted me some reference to check stuff out, Ill do it later this week | 20:42 |
@sonney2k | blackburn1, bow before the leader ;-) | 20:42 |
heiko | busy with plotting currently | 20:42 |
heiko | hope to sort that out soon | 20:42 |
@sonney2k | heiko, ok - it is starting to be a bit pressing though | 20:43 |
heiko | yeah, sorry that this took so long | 20:43 |
@sonney2k | wiking, n4nd0 what is CData? | 20:43 |
blackburn1 | we need an offshore to keep money we get from stealing people and oil | 20:43 |
@sonney2k | wiking, n4nd0 it sounds like a very generic name doing what CFeatures was invented for - no? | 20:44 |
alexlovesdata | why not an island close to the pacific trash vortex?? | 20:44 |
* sonney2k also prefers an island | 20:44 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 20:45 | |
blackburn1 | well cyprus is an island I heard it is ok to keep stolen money | 20:45 |
n4nd0 | sonney2k: it is what I called CStructuredData | 20:45 |
n4nd0 | wiking prefers to use CData for that | 20:45 |
n4nd0 | to be honest, I am ok with any of them... | 20:45 |
@sonney2k | n4nd0, structured data is CFeatures + CLabels right? | 20:46 |
n4nd0 | sonney2k: no | 20:46 |
wiking | heiko: !!!! | 20:46 |
wiking | heiko: i think i did it | 20:46 |
n4nd0 | CStructuredData is the building block of CStructuredLabels | 20:47 |
n4nd0 | CStructuredLabels is a DynamicObjectArray of CStructuredData | 20:47 |
heiko | blawiking, what happened? :) | 20:47 |
CIA-39 | shogun: Heiko Strathmann master * rbe9da8e / (5 files): Merge pull request #710 from karlnapf/master - http://git.io/ID4pQw | 20:47 |
CIA-39 | shogun: Heiko Strathmann master * rdaabd3f / (2 files): updated graphical examples - http://git.io/6-7Eig | 20:47 |
CIA-39 | shogun: Heiko Strathmann master * r15cb24b / (2 files): renamed file - http://git.io/RCmIbA | 20:47 |
CIA-39 | shogun: Heiko Strathmann master * ra99f6fe / examples/undocumented/python_modular/graphical/latex_plot_inits.py : added import to make matlibplot plots look nice - http://git.io/As19Hg | 20:47 |
CIA-39 | shogun: Heiko Strathmann master * r9b66a85 / (2 files): minor updates - http://git.io/NTy-tA | 20:47 |
blackburn1 | blawiking? my hamming based reception thought you are calling me and then kaboom | 20:48 |
n4nd0 | sonney2k: CStructuredData can be e.g. a sequence (for HM-SVM), a real number (for the multiclass example) a tree, etc | 20:48 |
heiko | n4nd0, blackburn, sonney2k, check out nice looking plots in tutorial! | 20:48 |
n4nd0 | heiko: let's see :) | 20:48 |
@sonney2k | n4nd0, ok then the name is a bit misleading maybe ... in fact it is a part of the label | 20:52 |
n4nd0 | sonney2k: yeah, it could be | 20:54 |
n4nd0 | sonney2k, wiking: what name would you like to have instead? | 20:54 |
@sonney2k | heiko, nice figures :D | 20:54 |
heiko | sonney2k, thanks :) they depict the tests very nice, take them for the new webpage | 20:55 |
blackburn1 | heiko: what the heck is plotted? | 20:56 |
blackburn1 | :D | 20:56 |
n4nd0 | heiko: I am checking shogun-tutorial/fig/statistical-learning, is that the correct place? | 20:56 |
heiko | yeah, but better look at the compiled pdf to the how the fit in | 20:57 |
heiko | blackburn, read the tutorial :D | 20:57 |
blackburn1 | heiko: I sneeze when I read statistics | 20:57 |
blackburn1 | all the time | 20:58 |
n4nd0 | heiko: nice! | 20:58 |
@sonney2k | btw heiko this reminds me I haven't received any figures (apart the ones from you) | 20:59 |
n4nd0 | sonney2k: I won't be able to make it for the meeting this time :( | 20:59 |
wiking | heiko: here? | 20:59 |
@sonney2k | n4nd0, well you have been around last time - btw nico should be back so update him :D | 21:00 |
heiko | wiking, yes | 21:00 |
wiking | heiko: https://github.com/shogun-toolbox/shogun/pull/711 | 21:00 |
n4nd0 | sonney2k: yeah he sent us a mail, I am writing back to him now | 21:00 |
wiking | heiko: it works :))) | 21:00 |
wiking | heiko: both multiclass and binary | 21:00 |
heiko | wiking, well done :)! | 21:01 |
CIA-39 | shogun: Heiko Strathmann master * rf1dc712 / (4 files in 4 dirs): Merge pull request #711 from vigsterkr/utest - http://git.io/s7B6jg | 21:01 |
CIA-39 | shogun: Viktor Gal master * r6cbc4ee / (4 files in 4 dirs): Fix gaussian_generator in DataGenertor - http://git.io/OYYwzg | 21:01 |
heiko | wiking, gmock is still not workin ghere | 21:01 |
heiko | thats my plan for this weelk | 21:01 |
heiko | week | 21:01 |
n4nd0 | sonney2k: btw, I am spending quite a bit of time lately reading about speech recognition. I'd like to apply our HM-SVM to some problems in this area. What do you think about it? | 21:02 |
heiko | sonney2, should we write an email about the figures? | 21:02 |
wiking | n4nd0 sonney2k anything is good for me :) but since it's we use the same concept in latentdata and structureddata... i thoguth that it would be nice to have only one class like that | 21:02 |
@sonney2k | heiko, yes | 21:02 |
@sonney2k | heiko, please do | 21:02 |
heiko | sonney2k, which time is the meeting again? | 21:02 |
heiko | sonney2k, Ill do | 21:02 |
wiking | heiko blackburn1 now i can use this unit test to see what happens with the update of libocas to 0.96 \o/ | 21:02 |
heiko | sonney2k, sorry for the question just saw your mail | 21:02 |
blackburn1 | nice | 21:03 |
shogun-buildbot | build #338 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/338 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:04 |
wiking | FAILURE I AM? :) | 21:04 |
heiko | ehm | 21:05 |
heiko | these errors are strange | 21:05 |
heiko | wiking no not you | 21:05 |
heiko | and not me | 21:05 |
wiking | hah fuck just a sec i need to send in a quickfix | 21:06 |
wiking | good that bots are not running automatically make unit-tests :D | 21:06 |
heiko | ah so thats whats happening? | 21:06 |
wiking | nono | 21:06 |
heiko | I gotta go, wiking Ill merge tomorrow | 21:08 |
-!- heiko [~heiko@host86-176-126-57.range86-176.btcentralplus.com] has quit [Quit: Leaving.] | 21:08 | |
CIA-39 | shogun: Sergey Lisitsyn master * r5df0dcd / src/interfaces/modular/modshogun_ignores.i : Ignored bmrm stuff in modular - http://git.io/wI1WIQ | 21:13 |
n4nd0 | sonney2k, blackburn1: I have just rebased a few moments ago | 21:16 |
blackburn1 | sonney2k: are you ok to merge? | 21:16 |
@sonney2k | blackburn1, sec | 21:17 |
@sonney2k | n4nd0, is you PR updated - so show I re-review it? | 21:17 |
n4nd0 | sonney2k: I did the changes you told me in your last revision | 21:19 |
n4nd0 | and now I have just rebased solving a couple of minor conflicts | 21:19 |
@sonney2k | n4nd0, ok then let me have a quick look | 21:20 |
@sonney2k | n4nd0, what I still don't understand is why you didn't replace the constructor | 21:21 |
@sonney2k | SGMatrix< ST >* feats, int32_t num_vecs, int32_t num_feats | 21:21 |
@sonney2k | with the SGMatrixStringList<ST> one | 21:21 |
@sonney2k | then we could create typemaps for that type | 21:21 |
@sonney2k | n4nd0, also I think the SGMatrix< ST > feats, int32_t feat_length, int32_t num_vecs constructor | 21:22 |
shogun-buildbot | build #339 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/339 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:22 |
@sonney2k | should rather be one that takes an SGNDArray as input | 21:22 |
n4nd0 | sonney2k: ok with the first thing about introducing SGMatrixList | 21:25 |
n4nd0 | sonney2k: but the second change, it is another different constructor | 21:25 |
@sonney2k | n4nd0, please explain - what it is supposed to do then | 21:26 |
n4nd0 | sonney2k: have you seen the explanation in the header file? | 21:26 |
wiking | blackburn1: can u merge this: https://github.com/shogun-toolbox/shogun/pull/712 | 21:27 |
CIA-39 | shogun: Sergey Lisitsyn master * r60278ef / (2 files): Merge pull request #712 from vigsterkr/utest - http://git.io/X4t3iA | 21:27 |
CIA-39 | shogun: Viktor Gal master * r2e361a5 / (2 files): Fix unit testing of LatentSVM - http://git.io/qeJ2nA | 21:27 |
wiking | thnx | 21:28 |
@sonney2k | n4nd0, it is the one where I get *parse error* | 21:28 |
n4nd0 | sonney2k: hehe where? | 21:28 |
n4nd0 | so you construct a MatrixFeatures (a list of matrices) | 21:28 |
n4nd0 | from a big matrix | 21:28 |
n4nd0 | you give it a matrix [ ........ ] | 21:29 |
@sonney2k | n4nd0, yeah but why not use SGNDArray then? you can have a tensor that is NxMxL big | 21:29 |
@sonney2k | where L is the number of matrices | 21:29 |
@sonney2k | and NxM the matrix size | 21:29 |
n4nd0 | and partition it in a list of [...][...][...] | 21:29 |
n4nd0 | sonney2k: ok, but why not to have also one constructor like this? | 21:30 |
n4nd0 | I mean, it can be useful to do it from a matrix | 21:30 |
n4nd0 | I just did it to use it from CTwoStateModel::simulate_data | 21:30 |
n4nd0 | I ported a function from the hmsvm toolbox that simulates two data and stores it in a matrix | 21:31 |
@sonney2k | n4nd0, I understand that constructor now though I don't like it - it is probably the same as having an SGVector as input for a constructor of SGMatrix | 21:33 |
n4nd0 | sonney2k: yes | 21:33 |
@sonney2k | anyways this one certainly is not a showstopper | 21:33 |
n4nd0 | I can modify the simulate data function so this constructor is not required though | 21:34 |
n4nd0 | in any case I don't like the idea of getting rid of this constructor ... | 21:35 |
n4nd0 | I can use data generated from the hmsvm toolbox in matlab from python scripts thanks to it | 21:36 |
n4nd0 | see examples python_modular/structure_hmsvm_mosek.py | 21:36 |
@sonney2k | n4nd0, well you could easily do one reshape to get this to work then | 21:37 |
n4nd0 | sonney2k: ok :) | 21:37 |
n4nd0 | sonney2k: should I change the other constructor now and ammend? | 21:38 |
@sonney2k | n4nd0, yes do that now and then let blackburn1 or so merge :D | 21:38 |
n4nd0 | I will stay here to feel guilty if a disaster happens :-O | 21:39 |
shogun-buildbot | build #340 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/340 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Viktor Gal <viktor.gal@maeth.com> | 21:40 |
wiking | hahaha!!! | 21:40 |
n4nd0 | I'll not be the only one in the blamelist at least :D | 21:41 |
wiking | RUN ] MulticlassOCASTest.train | 21:41 |
wiking | make: *** [build] Segmentation fault: 11 | 21:41 |
wiking | \o/ | 21:41 |
blackburn1 | wiking: -1? | 21:41 |
wiking | not exactly the way i've imagined buuuut it indicates that it doesn't work! | 21:41 |
wiking | blackburn1: hehe yes | 21:44 |
wiking | ok i'm sending in the patch | 21:44 |
wiking | for that pr | 21:45 |
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has left #shogun [] | 21:51 | |
wiking | blackburn1: done | 21:53 |
wiking | blackburn1: https://github.com/shogun-toolbox/shogun/pull/697 | 21:53 |
wiking | the unit tests for svmocas and multiclassocas runs as expected | 21:53 |
CIA-39 | shogun: Viktor Gal master * r11c1730 / src/shogun/lib/external/libocas.cpp : Add minor fixes for libocas 0.96 update - http://git.io/sQMkAQ | 21:54 |
CIA-39 | shogun: Viktor Gal master * r3548686 / src/shogun/lib/external/libocas.cpp : Fix labeling in msvm_ocas_solver - http://git.io/FnrIjA | 21:54 |
CIA-39 | shogun: Sergey Lisitsyn master * r892a159 / (2 files): Merge pull request #697 from vigsterkr/master - http://git.io/K2-kBg | 21:54 |
CIA-39 | shogun: Viktor Gal master * r0ffe237 / (2 files): Update libocas to version 0.96 - http://git.io/rzXUgw | 21:54 |
wiking | so now we'll only need to create a wrapper for svm_ocas_solver_nnw ;) | 21:55 |
blackburn1 | wiking: why to use it? | 21:56 |
wiking | hahah good question... but if anybody who would need it... | 21:57 |
wiking | anyhow i guess after this doing train unittests for other classifiers should be straightforward ;P | 21:57 |
blackburn1 | vojta said it was developed for some computer vision guys | 21:57 |
blackburn1 | I am curious what is the case to require positive weights | 21:58 |
blackburn1 | with only positive alphas we get better understanding of class representatives | 21:58 |
blackburn1 | but what do we get with positive weights? | 21:59 |
shogun-buildbot | build #341 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/341 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Viktor Gal <viktor.gal@maeth.com> | 21:59 |
wiking | hah i have to do there something | 22:00 |
wiking | as it's my code that breaks java | 22:00 |
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun | 22:01 | |
blackburn1 | wiking: why do you inherit from dense labels? | 22:01 |
blackburn1 | KMcQuisten: hey there, what is the issue today? ;) | 22:01 |
wiking | that i have to redesign now | 22:02 |
KMcQuisten | Not so much an issue. | 22:02 |
KMcQuisten | I just want to take a chainsaw to part of your object model. Specifically, the machine evaluation code as well as the result code. | 22:03 |
@sonney2k | KMcQuisten, I would very much prefer if you do these things publicly on the mailinglist | 22:03 |
KMcQuisten | That's fine, sonney. I can certainly do that. I was just hoping to have a conversation first. | 22:04 |
KMcQuisten | I don't have any bugs to report. everything seems to be working as it should | 22:04 |
wiking | hahah cache and pipelining kicked in in my cpu | 22:04 |
blackburn1 | sonney2k: we failed to discuss some issues at mailing list :) | 22:04 |
wiking | compiling is getting faster and faster :) | 22:04 |
blackburn1 | KMcQuisten: what do you mean? | 22:04 |
wiking | blackburn1: i guess he means that all's good ;) | 22:05 |
blackburn1 | no about chainsaw and code | 22:05 |
blackburn1 | sounds scary | 22:05 |
KMcQuisten | Hehehe. Not so much scary | 22:05 |
blackburn1 | anyway please elaborate | 22:05 |
KMcQuisten | Here's my take: When constructing cross-validation, you must pass the evaluationCriterion object to the crossValidation object, which calculates the statistics, averages them, and then passes them to the EvaluationResult object. | 22:07 |
blackburn1 | rigggght | 22:08 |
blackburn1 | so? | 22:08 |
KMcQuisten | I think thats backwards. The crossValidation object should pass the observed/predicted pairs to the evaluationResult object, and the evaluationCriterion object should be passed to the evaluationResult using a method. This way one run of crossvalidation could be used, and then multiple evaluationCriterion could be thrown at it to look at CV results in multiple ways without having to rerun a full CV for each evaluationcriterion on | 22:09 |
KMcQuisten | As it stands, It's impossible for me to actually look at the obs/pred pairs generated by the builting cv, so I can't do more in depth analysis of my CV results if I use the builtin methods | 22:10 |
KMcQuisten | I'd like the opportunity to refactor the code to make it more flexible | 22:11 |
KMcQuisten | If there's a formal proposal process for such things I'd like to follow it, of course. | 22:11 |
blackburn1 | if you want to improve anything just do and sent a PR | 22:12 |
blackburn1 | (pull request) | 22:12 |
KMcQuisten | Yep Yep. | 22:12 |
KMcQuisten | I know that. I have my own branch forked off. | 22:12 |
KMcQuisten | I'd rather tell folks what i'm planning to do first. Common courtesy and all, since it would involve some big changes. | 22:13 |
KMcQuisten | sonney, would you like me to write something up for the mailing list before I get started? | 22:13 |
blackburn1 | sonney2k: ^ | 22:14 |
blackburn1 | KMcQuisten: so you want to use multiple | 22:14 |
blackburn1 | evaluations | 22:14 |
@sonney2k | http://web.mit.edu/newsoffice/2012/autonomous-robotic-plane-flies-indoors-0810.html | 22:15 |
@sonney2k | blackburn1, any idea where gsomix is | 22:15 |
blackburn1 | sonney2k: will appear soon | 22:16 |
blackburn1 | sonney2k: I am surprised to see kind of machine leraning here | 22:16 |
-!- gsomix [~gsomix@80.234.28.229] has joined #shogun | 22:17 | |
gsomix | good evening | 22:17 |
KMcQuisten | Yes. Without having to completely rerun CV for each criterion. Assuming the splitting criterion doesn't change, each time CV is run on a dataset with a fixed set of parameters, it would produce the same observed/predicted label pairs, so why generate them over and over for different evaluations? | 22:17 |
blackburn1 | KMcQuisten: but you optimize only one criterion right? | 22:18 |
KMcQuisten | This would also allow output for evaluations to return different types, and return them as values to wrapped languages, instead of having to rely only on stdout | 22:18 |
@sonney2k | KMcQuisten, I understand this desire - but what do you do if you have test data sets? then you would need to store all outputs for all evaluations in memory right? | 22:18 |
@sonney2k | gsomix, hey - so what's going on? did you manage to do the example fixes? | 22:19 |
blackburn1 | KMcQuisten: but why no to refactor modelselectionoutput a little to store things? | 22:19 |
@sonney2k | gsomix, (or did I miss anything?) | 22:19 |
blackburn1 | not to print but to store I mean | 22:19 |
blackburn1 | it already receives all you need | 22:19 |
blackburn1 | and it supports custom additional evaluations too | 22:19 |
KMcQuisten | See, this is why I wanted to talk about this before I did anything :) . I'm still familiarizing myself with the modelselection code, so I am unsure. Does it allow for the comparison of arbitrary sets of features and kernels, or must those be fixed and only the parameters varied? | 22:21 |
gsomix | sonney2k, hey. what do you mean? I plan to debug my DirectorDotFeatures example and solve problem with SGVector in directors. But little later. At now I'm working on zero-copying mechanism. At next week plan to start work on new typemaps for modelselection (it seems, that it's little easy that sg in director or zero-copy). | 22:22 |
blackburn1 | KMcQuisten: why to do that? | 22:22 |
gsomix | ah, and I forgot about examples for Labels. | 22:23 |
gsomix | too bad | 22:23 |
blackburn1 | I am a bit lost, comparison of what? | 22:23 |
@sonney2k | gsomix, that is quite a bit of work - not sure how you want to do that within 2 weeks... | 22:23 |
blackburn1 | sonney2k: why to +1? | 22:23 |
shogun-buildbot | build #342 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/342 blamelist: Viktor Gal <viktor.gal@maeth.com> | 22:24 |
@sonney2k | blackburn1, if I understand correctly ocas uses labels 1... <nr_classes> | 22:24 |
@sonney2k | in shogun we use 0...<nr_classes-1> | 22:24 |
blackburn1 | sonney2k: there is no +1 in multiclass ocas we had before | 22:24 |
@sonney2k | so we should +1 shogun's labels for ocas to work | 22:24 |
gsomix | sonney2k, 2 weeks? when is the release of new version? | 22:25 |
@sonney2k | hopefully 1st of september | 22:25 |
gsomix | oops | 22:25 |
KMcQuisten | Comparison of the CV accuracies for SVMs applied to my data under various choices of feature expression, kernel selection, and kernel parameter. | 22:25 |
KMcQuisten | blackburn1: ^ | 22:25 |
gsomix | ok | 22:25 |
@sonney2k | but we need at least 1 week to just do testing / bugfixing / documentation an no more feature additions | 22:25 |
blackburn1 | KMcQuisten: okay lets specify things here | 22:26 |
gsomix | zero-copying mechanism - 4-5 hours | 22:26 |
@sonney2k | I would actually say 2 weeks ... | 22:26 |
gsomix | solve problem with SGVector - 1-2 day | 22:26 |
@sonney2k | gsomix, well do what you can... | 22:26 |
KMcQuisten | blackburn1: Yes, let's. | 22:26 |
@sonney2k | gsomix, what is the new SGVector problem? | 22:26 |
@sonney2k | gsomix, I thought you resolved it? | 22:26 |
gsomix | sonney2k, nope. | 22:27 |
blackburn1 | KMcQuisten: cross validation splits to train/test subsets | 22:27 |
KMcQuisten | Yes | 22:27 |
blackburn1 | you need these indices, right? | 22:27 |
blackburn1 | when you construct parameter tree you may specify everything you want to vary | 22:28 |
KMcQuisten | blackburn1: No, I don't . I don't want to change anything about how the CV code does splitting, training, or testing of the untrained portion of the data. All that's fine, and I don't want to reinvent the wheel. | 22:28 |
gsomix | sonney2k, LibLinear's train in my example says that 'not a numpy or... bla-bla'. I need to debug it. | 22:28 |
wiking | sonney2k blackburn1 it's now good at is... ;) | 22:28 |
gsomix | not many time, I think | 22:29 |
blackburn1 | KMcQuisten: okay | 22:29 |
blackburn1 | KMcQuisten: so you need a few more evaluation to see | 22:29 |
blackburn1 | evaluations* | 22:29 |
blackburn1 | are you ok with optimizing by only one evaluation parameter? | 22:29 |
-!- sr___ [u5548@gateway/web/irccloud.com/x-abthilurbjnvpfrb] has joined #shogun | 22:29 | |
n4nd0 | 3 | 22:30 |
n4nd0 | wrong tab... | 22:30 |
@sonney2k | blackburn1, wiking I think the best thing to do would be to compare mc ocas output (of stand alone ocas) against shogun once | 22:32 |
@sonney2k | and then feel better | 22:32 |
wiking | sonney2k: :DDDD | 22:32 |
wiking | sonney2k: i've compared libocas v0.95 and how that was ported in shogun | 22:32 |
KMcQuisten | Yes. I don't want to pass multiple evaluation criterion to the crossvalidation object. My point is that structurally I don't think the evaluationcriterion belongs in the CV code. The CV code should just spit out obs/pred pairs to the result object, and the evaluationCriterions can be passed to the result object. | 22:33 |
wiking | sonney2k: and there the -1 was removed basically | 22:33 |
@sonney2k | gsomix, well then do 0-copy now ... | 22:33 |
gsomix | aha | 22:33 |
KMcQuisten | The question is, as sonney said, how that will work for very very large datasets | 22:33 |
@sonney2k | KMcQuisten, make it optional... | 22:33 |
@sonney2k | then it is ok | 22:34 |
blackburn1 | KMcQuisten: but why not to use modelselectionoutput? | 22:34 |
@sonney2k | wiking, what where the class labels you used? | 22:34 |
@sonney2k | 0...2? | 22:34 |
blackburn1 | I don't get that point yet | 22:34 |
KMcQuisten | I have to look more into the modelselection code before I can give you a good answer to that. | 22:34 |
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has quit [Ping timeout: 244 seconds] | 22:34 | |
blackburn1 | KMcQuisten: you may add a few more evaluation instances to modelselectionoutput | 22:34 |
blackburn1 | and it will be printed | 22:35 |
KMcQuisten | If that's a way I can get what I want without messing with code, then it's fine, | 22:35 |
wiking | sonney2k: yeps | 22:35 |
blackburn1 | if you want to make it not printed but stored - we can do that | 22:35 |
KMcQuisten | Aha. That's the other problem. I'd rather just get that value and keep in internally rather than having to scrape from stdout | 22:35 |
@sonney2k | blackburn1, would that enable us to also store trained models? | 22:36 |
KMcQuisten | But I can conceive of eval criteria that aren't simple floats, but rather arrays of floats, i.e. a list of binned accuracies across the entire prediction range of a regression problem | 22:36 |
blackburn1 | sonney2k: it prints trained model right now | 22:37 |
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has joined #shogun | 22:37 | |
@sonney2k | wiking, I would suggest to email vojtech (after reaing ocas doc) | 22:37 |
blackburn1 | if you pass model selection output to the instance | 22:37 |
@sonney2k | blackburn1, *store* | 22:37 |
blackburn1 | sonney2k: yes, if we change print to store - it would enable to do that | 22:37 |
blackburn1 | but I planned to do that optionally of course | 22:37 |
@sonney2k | wiking, I am pretty sure he uses labels 1...nr_class | 22:37 |
@sonney2k | makes sense | 22:37 |
blackburn1 | KMcQuisten: | 22:39 |
blackburn1 | so now you may put | 22:39 |
blackburn1 | ModelSelectionOutput instance | 22:39 |
blackburn1 | to ModelSelection instance | 22:39 |
blackburn1 | and it will print stuff you need | 22:39 |
blackburn1 | you may add anything to ModelSelectionOutput | 22:39 |
blackburn1 | like more evaluations | 22:39 |
blackburn1 | if you add some new complex evaluation I'd suggest to add its handling to the ModelSelectionOutput | 22:40 |
blackburn1 | I will change it to store things today/tomorrow | 22:40 |
KMcQuisten | That sounds fine to me. I will take a look and see what I can do. | 22:40 |
shogun-buildbot | build #343 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/343 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Viktor Gal <viktor.gal@maeth.com> | 22:41 |
blackburn1 | KMcQuisten: I just have to think how can we add this kind of storage here | 22:42 |
KMcQuisten | Well, hold up on that. It might not need to be stored if it can just be calculated and returned | 22:43 |
@sonney2k | KMcQuisten, well it can take quite a while... | 22:44 |
* sonney2k Zzzz... | 22:44 | |
blackburn1 | KMcQuisten: I hardly have an API in mind for that | 22:45 |
KMcQuisten | blackburn1: Ok, cool. I'll familiarize myself a bit more with the modelselection code and hopefully I will be able to see how it can help me. | 22:46 |
blackburn1 | KMcQuisten: remind me please - are you using python? | 22:47 |
KMcQuisten | Yes | 22:47 |
blackburn1 | the problem is that storage should be quite generic | 22:47 |
KMcQuisten | Yes. | 22:48 |
KMcQuisten | with the SO machines and all being put together it has to work for them too | 22:48 |
KMcQuisten | not everyon'es outputs are nice floats. :) | 22:48 |
blackburn1 | well evaluation is always float - a score | 22:49 |
KMcQuisten | Yes, I know, but they predictions that the machines make are not always nice floats | 22:49 |
n4nd0 | sonney2k, blackburn1: I updated the branch with the new constructor | 22:49 |
KMcQuisten | and I can think of ways that the score is actualyl an array of scores, as opposedto a single float, so even that assumption isn't quite right wither | 22:50 |
KMcQuisten | either | 22:50 |
blackburn1 | KMcQuisten: would be nice to have map: ParameterCombination -> some bunch of results | 22:51 |
blackburn1 | but it is very inconvenient for user | 22:51 |
KMcQuisten | Perhaps. I'm not entirely sure we're talking about the same thing here. | 22:51 |
KMcQuisten | I want to dig into the modelselection code before we discuss this further. | 22:52 |
blackburn1 | KMcQuisten: okay | 22:52 |
blackburn1 | you are not the only user that wants that stuff btw | 22:52 |
blackburn1 | second one already :) | 22:52 |
n4nd0 | blackburn1: should we merge or do you prefer to wait for Soeren until tom? | 22:54 |
KMcQuisten | RIghto | 22:54 |
blackburn1 | n4nd0: did you have to include .cpp? | 22:55 |
blackburn1 | that sounds to be a blocker for you lets just merge | 22:55 |
n4nd0 | yes, it looks like this now | 22:55 |
n4nd0 | from SGMatrixList.h it is included SGMatrix.h | 22:56 |
blackburn1 | however I do not like that that much | 22:56 |
blackburn1 | ah | 22:56 |
blackburn1 | is .cpp included? | 22:56 |
blackburn1 | n4nd0: do you mind to make it static method of sgmatrixlist? | 22:56 |
n4nd0 | in SGMatrix.h it is SGMatrixList forward declared | 22:57 |
n4nd0 | and from SGMatrix.cpp, SGMatrixList.cpp is included | 22:58 |
n4nd0 | blackburn1: yeah, but I can tell you why it is done like that | 22:58 |
n4nd0 | blackburn1: ok | 22:58 |
blackburn1 | n4nd0: including .cpp could be bad | 22:58 |
n4nd0 | blackburn1: it is just because of the style in which templated classes are defined in SHOGUN | 22:59 |
n4nd0 | blackburn1: because it is in the .cpp where the types are specified | 22:59 |
blackburn1 | yeah | 22:59 |
blackburn1 | i see | 22:59 |
blackburn1 | n4nd0: anyway .cpp should not be included - it can become a mystery problem later | 23:00 |
n4nd0 | I agree | 23:00 |
n4nd0 | I am changing it to static right now :) | 23:01 |
blackburn1 | KMcQuisten: there are now 3 guys wanting to check xval results in a fancy way | 23:02 |
KMcQuisten | Hehe. Fancy | 23:02 |
KMcQuisten | I feel like i need a parasol now LOL | 23:03 |
* blackburn1 never knew what is parasol | 23:05 | |
KMcQuisten | fancy umbrella for protecting fair-skinned ladies form the sun | 23:06 |
KMcQuisten | from | 23:06 |
blackburn1 | yeah I googled | 23:06 |
KMcQuisten | Ah yes | 23:06 |
KMcQuisten | SHould have known ;) | 23:06 |
blackburn1 | the only word for that I knew is umbrella | 23:06 |
blackburn1 | :) | 23:06 |
blackburn1 | but there are quite a few like rainshade and others | 23:06 |
blackburn1 | parapluie | 23:07 |
blackburn1 | wtf | 23:07 |
blackburn1 | :D | 23:07 |
KMcQuisten | Hahaha! I'd not hear that one | 23:07 |
KMcQuisten | Ok, i'm going to study modelselection, and then I'll know better what can and cannot be easily fiddled with :) | 23:07 |
KMcQuisten | thanks for your help again | 23:07 |
blackburn1 | welcome | 23:07 |
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Quit: Page closed] | 23:08 | |
n4nd0 | blackburn1: you know, parasol is a funny word | 23:10 |
blackburn1 | yeah sounds funny at least | 23:10 |
n4nd0 | we use it in Spanish too | 23:11 |
n4nd0 | it means like this | 23:11 |
n4nd0 | para = stop | 23:11 |
n4nd0 | sol = sun | 23:11 |
n4nd0 | :D | 23:11 |
blackburn1 | heh | 23:11 |
blackburn1 | I remember song or so | 23:11 |
blackburn1 | dia de sol | 23:11 |
n4nd0 | which one? | 23:11 |
blackburn1 | sunny day? | 23:11 |
n4nd0 | yeah | 23:11 |
blackburn1 | huh I know spanish hurray | 23:11 |
blackburn1 | :D | 23:11 |
blackburn1 | I am watching participants list in your PR | 23:12 |
blackburn1 | two are normal guys | 23:13 |
blackburn1 | and only I am a cat | 23:13 |
blackburn1 | that cat is so damn awesome you shall not resist | 23:13 |
n4nd0 | haha | 23:13 |
CIA-39 | shogun: Sergey Lisitsyn master * r31528b4 / (2 files in 2 dirs): A few more warnings fixes - http://git.io/xN9S4A | 23:14 |
* wiking cannot keep up with rebasing :) | 23:19 | |
wiking | lol | 23:29 |
wiking | blackburn1: here? have u managed to get unit testing working? :) | 23:30 |
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has quit [Ping timeout: 248 seconds] | 23:30 | |
blackburn1 | wiking: no I didn't try :) | 23:30 |
wiking | i've just made a simple pre-commit hook for git | 23:30 |
wiking | and it works great with the unit testing | 23:30 |
wiking | if it fails, it does not let you commit :) | 23:30 |
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has joined #shogun | 23:31 | |
wiking | blackburn1: should i add that into README.developer? | 23:33 |
blackburn1 | I am not a big fan of that solution to be honest | 23:33 |
wiking | blackburn1: ok i'll add it to the README.developer to kindly ask developers to add that to their pre-commit hook | 23:34 |
wiking | but of course we cannot force it | 23:34 |
blackburn1 | yeah sure | 23:34 |
wiking | so it'll be just a suggestion in a way | 23:34 |
shogun-buildbot | build #344 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/344 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:39 |
blackburn1 | wiking: can you change base of latentlabels? | 23:41 |
wiking | blackburn1: yep i'll change now | 23:42 |
wiking | just a sec | 23:42 |
n4nd0 | blackburn1, wiking: hey guys | 23:47 |
blackburn1 | yes? | 23:47 |
n4nd0 | something I don't really understand entered in a commit | 23:47 |
-!- blackburn1 is now known as blackburn | 23:47 | |
blackburn | uh finally I got that back | 23:47 |
n4nd0 | it is related to the data submodule | 23:48 |
n4nd0 | take a look here please https://github.com/shogun-toolbox/shogun/pull/684/files | 23:48 |
n4nd0 | I ignore whether this is or not relevant | 23:48 |
n4nd0 | did you get my messages about the problem ^? | 23:48 |
n4nd0 | my connection sucks... | 23:49 |
blackburn | n4nd0: no please remove that | 23:50 |
blackburn | your data can be outdated and you trying to restore it to your state | 23:50 |
blackburn | if it appears in your changes just update data folder , i.e. enter to shogun/data and git pull origin master | 23:51 |
wiking | or git submodule update | 23:51 |
blackburn | yeah | 23:52 |
wiking | blackburn: https://github.com/shogun-toolbox/shogun/pull/713 | 23:54 |
CIA-39 | shogun: Viktor Gal master * rdabe375 / src/README.developer : Add unit testing description to README.developer - http://git.io/z7hRqA | 23:54 |
wiking | blackburn: deep enough description? | 23:54 |
CIA-39 | shogun: Sergey Lisitsyn master * r2250325 / src/README.developer : Merge pull request #713 from vigsterkr/latent - http://git.io/gICzfA | 23:54 |
blackburn | may be I should read it before | 23:54 |
blackburn | :D | 23:54 |
wiking | hahahahah | 23:54 |
wiking | yeah | 23:54 |
blackburn | yeah deep enough | 23:54 |
wiking | ok let's see that label thingy | 23:55 |
wiking | as a quickfix | 23:55 |
n4nd0 | blackburn: I do not manage to take out that change from the past commit though | 23:58 |
blackburn | n4nd0: which change? | 23:59 |
--- Log closed Tue Aug 14 00:00:14 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!