Shogun Machine Learning - IRC Logs
--- Log opened Mon Apr 22 00:00:28 2013
-!- k_matsuda [] has quit [Remote host closed the connection]02:11
-!- rgirish28 [~rgirish28@] has joined #shogun02:50
-!- iglesiasg [] has quit [Quit: leaving]02:51
-!- votjakovr [] has joined #shogun03:25
-!- ry_ [] has joined #shogun03:45
-!- vikram360 [~vikram360@] has joined #shogun04:14
shogun-buildbotbuild #369 of nightly_default is complete: Failure [failed test]  Build details are at
-!- votjakovr [] has quit [Quit: ChatZilla 0.9.90 [Firefox 20.0.1/20130409194949]]04:49
-!- rgirish28 [~rgirish28@] has quit [Quit: rgirish28]04:58
-!- rgirish28 [~rgirish28@] has joined #shogun05:00
-!- rgirish28 [~rgirish28@] has quit [Remote host closed the connection]05:31
-!- rgirish28 [~rgirish28@] has joined #shogun05:31
-!- k_matsuda [] has joined #shogun05:38
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun05:51
-!- sumit [73f91219@gateway/web/freenode/ip.] has joined #shogun05:54
-!- sumit [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]06:09
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]06:24
-!- lisitsyn [] has quit [Quit: Leaving.]06:33
-!- rgirish28 [~rgirish28@] has quit [Ping timeout: 245 seconds]06:47
-!- rgirish28 [~rgirish28@] has joined #shogun07:06
-!- rgirish28 [~rgirish28@] has quit [Client Quit]07:06
-!- rgirish28 [~rgirish28@] has joined #shogun07:10
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has joined #shogun07:14
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun07:16
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]07:19
-!- k_matsuda [] has quit [Read error: Connection reset by peer]07:22
-!- pararth [74498a2d@gateway/web/freenode/ip.] has joined #shogun07:23
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun07:53
-!- abhi_iit_10691 [6eeaec2c@gateway/web/freenode/ip.] has joined #shogun08:12
-!- iglesiasg [] has joined #shogun08:22
-!- mode/#shogun [+o iglesiasg] by ChanServ08:22
-!- gsomix [~gsomix@] has joined #shogun08:48
gsomixgood morning08:48
@iglesiasggood morning gsomix08:49
-!- abhi_iit_10691 [6eeaec2c@gateway/web/freenode/ip.] has quit [Quit: Page closed]08:55
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:02
-!- ry_ [] has quit [Remote host closed the connection]09:08
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Quit: Page closed]09:26
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun09:31
deerishi@iglesiasg: hi09:34
deerishii have implemented the scikit kpca example09:34
deerishi@iglesiasg: could u have a look at it09:38
-!- gsomix [~gsomix@] has quit [Ping timeout: 272 seconds]09:38
-!- gsomix [~gsomix@] has joined #shogun09:40
-!- rgirish28 [~rgirish28@] has quit [Quit: rgirish28]09:53
@iglesiasgdeerishi: let me see10:00
deerishican i do a pull request regarding the kernel pca test check10:03
@iglesiasgdeerishi: it could be maybe somethign to merge in graphical examples10:04
@iglesiasgdeerishi: there are some things that look a bit messy in though10:06
-!- lisitsyn [] has joined #shogun10:06
-!- mode/#shogun [+o lisitsyn] by ChanServ10:06
deerishi@iglesiasg: could you eleborate a bit more?10:07
@iglesiasgdeerishi: well, just in general the creation of the data10:09
@iglesiasgyou are just creating data or plotting two circunferences, aren't you?10:10
@iglesiasgthat's something like x^2+y^2=r^210:11
@iglesiasgI have the intuition that it should be easier to create that data10:11
deerishii am actuaaly creating the data10:15
deerishii am storing the values of the circle in a 2 dimensional array10:17
deerishithen i am concatenating the 2 arrays by hstack for kpca data10:18
deerishi@iglesiasg : where is the graphical examples folder?10:37
-!- HeikoS [] has joined #shogun10:39
deerishihello HeikoS10:40
HeikoShi deerishi10:40
deerishii did solve the scikit kpca example10:41
HeikoSdeerishi: finally :)10:41
HeikoSIll have a look at the code10:41
HeikoSdeerishi: nice, so this makes sense now!10:42
HeikoSso what I suggest now is that you merge this code into the existing kpca example10:42
HeikoSnote that input arguments10:43
HeikoSthere is a list that is used for integration testing10:43
HeikoSfor the python_modular example, please remove all plotting code and return the transformation matrix and the transformed features and the preprocessor object10:43
HeikoSbut I would also suggest to add you plotting example under the graphical examples10:43
HeikoSsince it looks very nice10:44
HeikoSplease pimp it a bit and then add as graphical example10:44
HeikoSif you have questions on the integration tests, let me know10:44
@lisitsynsomething is unhealthy with normalization10:45
HeikoSlisitsyn: ah good that you see that!10:45
HeikoSwhat is it?10:45
@lisitsynIIRC kPCA preserves the scale10:45
@lisitsynHeikoS: check axes, they are e-1010:45
-!- sijin [~smuxi@] has joined #shogun10:45
HeikoSlisitsyn: I see10:45
HeikoSI dont know10:46
HeikoSif kpca does this10:46
HeikoSbetter check10:46
@lisitsynHeikoS: kpca with linear is pca10:46
@lisitsynor mds10:46
@lisitsynthey are all the same for linear kernel10:46
HeikoSlisitsyn: yes right its linear kernel10:46
HeikoSI forgot10:46
@lisitsynso scale should be preserved10:46
HeikoSwell, then we should see same scalling10:46
@lisitsynI am not sure if it is kpca that is wrong10:46
@lisitsynbut something is wrong for sure10:47
HeikoSlisitsyn: maybe we should have a unit test which compares KPCA with PCA10:47
HeikoSlisitsyn: haveing this example is good already10:47
HeikoSsince we can detect these things10:47
@lisitsynHeikoS: yeah10:48
sijin@lisitsyn, I just found tapkee KPCA and kernel PCA in shogun gives different result10:48
HeikoShere we go :)10:48
HeikoSwiking: hi!10:49
sijinI project features using Kernel PCA in Shogun, and compare it with the embedding result of tapkee KPCA,10:49
@lisitsynsijin ha!10:50
@lisitsynsijin: how significant?10:50
@lisitsyneither of them has a bug haha10:50
sijinwhen target dimension is 1, the difference is 1e-310:50
sijinhowever, when target_dimension is greater than 1, their ouput looks totally different10:51
@lisitsynsijin: could you plot it?10:51
sijinyou mean printing the result in text?10:52
@lisitsynsijin: no, just compare 2d results10:52
sijinnot yet. I am not familiar with shogun's plot function, any suggestion? or I send you the output of them?10:54
@lisitsynsijin: there is no plot function in shogun - what have you used to compare?10:54
sijinI just print it output.(I use small matrix for testing)10:55
@lisitsynI see10:55
sijincan I paste the output here( matrix size = 10x5)?10:58
wikingsijin: pastbin10:59
-!- van51 [] has joined #shogun11:00
sijin@wiking  thx,11:02
sijinhere is the result
-!- iglesiasg [] has quit [Quit: leaving]11:05
sijinHere is the result with target_dim = 2
sijinAs you can see, the output of shogun kernel pca is consistent. (I mean the output for target_dim ==3 is the "subset" of that of target_dim == 2)11:14
sijinsorry, the output with target_dim ==2 is the subset of result with target_dim == 311:15
-!- iglesiasg [] has joined #shogun11:15
-!- mode/#shogun [+o iglesiasg] by ChanServ11:15
sijin@lisitsyn Is there any special "normalization" in tapkee kpca ?11:18
-!- pararth [74498a2d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]11:33
sijin@listsyn I plot 2d output of kpca(shogun, blue) and kpca(tapkee, red)
-!- van51 [] has quit [Read error: Connection reset by peer]11:59
-!- van51 [] has joined #shogun12:01
wikingxfingerz this works12:10
-!- shogun-notifier- [] has joined #shogun12:10
shogun-notifier-shogun: Viktor Gal :develop * 026a413 / .travis.yml,src/Makefile:
shogun-notifier-shogun: Travis: run unit tests when building libshogun interface12:10
-!- gsomix [~gsomix@] has quit [Quit: Leaving]12:12
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]12:14
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun12:16
wikingdeerishi: see my comment for your prs12:18
deerishiseen .12:20
wikingso please redo your prs12:20
wikingas these ones are going to be ignored/closed automatically12:20
deerishiwiking: k. will do in a few minutes.12:22
wikingno worries12:22
shogun-notifier-shogun: Viktor Gal :develop * c1f397c / .travis.yml:
shogun-notifier-shogun: Fix typo in .travis.yml12:31
shogun-notifier-shogun: Viktor Gal :develop * e83dd28 / .travis.yml:
shogun-notifier-shogun: Travis: fix gmock/gtest location12:40
wikingHeikoS: have u checked/fixed bug #925 and #926?12:50
HeikoSwiking: nope12:51
HeikoSnot yet12:51
wikingHeikoS: okey, then i guess they are still leaking :P12:51
HeikoSyes ;)12:51
wikingok let's see if finally i've managed to get unit testing working on travis...12:52
wikingthen the only thing that remains is that stupid python_modular error we keep getting...12:52
wikingand it should be fixed before gsoc kicks off12:52
wikingso that PRs can be safely merged12:52
wikingif it's indicated 'good' by travis12:52
HeikoSwiking: yes12:53
HeikoSvery good that the unit-tests are working, great work! :)12:54
@sonney2kwiking, which one still fails?12:54
@sonney2kwe better blacklist the particular test if we cannot figure out what's wrong12:55
-!- travis-ci [] has joined #shogun12:56
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []12:56
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun13:00
wikingsonney2k: it's something general...13:03
wikingsonney2k: pickle and /tmp13:03
wikingsonney2k: i have no idea why that is an error atm13:04
-!- sumit [73f91219@gateway/web/freenode/ip.] has joined #shogun13:04
wikingsonney2k: /usr/lib/python2.7/ RuntimeWarning: [WARN] Could not read `labels/values' from `/tmp/fileG5ZWUc'!13:05
-!- vikram360 [~vikram360@] has quit [Ping timeout: 264 seconds]13:06
wikingsonney2k: i dont even know who's calling it :(13:06
wikingsonney2k: this is weird...13:09
wikingfirst it succeeds, but then when it's called within this test it fails13:09
wikingcd ../../../tests/integration/python_modular && \13:09
wiking( LD_LIBRARY_PATH=//usr/local/lib \13:09
wikingPYTHONPATH="//usr/local/lib/python2.7/dist-packages" python )13:09
wikingsonney2k: ideas?13:09
wikingHeikoS: not yet... hope the next build with travis succeeds ;P13:10
wikingsonney2k: i think it's something with integration script ./tester.py13:12
-!- travis-ci [] has joined #shogun13:12
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []13:12
wikingHeikoS: does the integration test run for you?13:14
* wiking is not building with python_modular interface... 13:14
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]13:14
-!- vikram360 [~vikram360@] has joined #shogun13:17
@sonney2kwiking, the tester runs all scripts from the very same python instance13:18
@sonney2kwiking, it is fast this way but if one of the test calls some stuff that underneath massacres memory consecutive test may fail13:19
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun13:20
wikingsonney2k: mmmm the first test fails right away with that pickle error (
deerishiHeikoS :I submitted my pull request. now i am working on Global Alignment kernels issue 86613:22
wikingbut then some succeeds13:22
HeikoSdeerishi: nice, will comment/review soon!13:23
wikinghahahaha oh13:26
wikingi see13:26
@sonney2kHeikoS, wiking, did someone change anything on labels?13:26
wikingHeikoS sonney2k
@sonney2kwiking, ?13:26
wikingsonney2k: it fails on nightly as well13:26
wikingso this is something unrelated to travis13:26
@sonney2kwiking, sure it is13:27
wikingsonney2k: HeikoS done some changes13:27
@sonney2kso just find out which version worked last13:27
@sonney2kand then we have it13:27
@sonney2kok friday everything was still fine13:28
wikingit started to crash with this commit:
wikingHeikoS: this is your work ;)13:28
wikingor let's say it started to crash after this
wikingi reckon it's somewhere here:
wikingsonney2k: lol13:32
HeikoShey guys13:33
HeikoSwas out for lunch13:33
HeikoSso whats happening?13:33
@sonney2kHeikoS, this13:34
@sonney2kand then the build bot failure13:34
wikingHeikoS: i think you've broken integration test with fixing the labels :S13:34
@sonney2ksee just the few lines above13:34
@sonney2kseems like13:34
HeikoSsonney2k: the GP demo is ultra cool13:34
HeikoSnow it needs heatmaps for the predictive distribution13:34
wikingsonney2k: tweeted ;)13:35
HeikoSsonney2k: ah13:36
HeikoSI see13:36
HeikoSI added a parameter13:36
HeikoSdid I push this to master or to develop? Currently confused about where the buildbot runs13:36
wikingHeikoS: develop13:36
HeikoSok, let me fix it13:36
wikingHeikoS: nightly buildbot runs on develop13:36
HeikoSwiking: ah ok13:36
@sonney2kwiking, hmmhh shouldn't the other buildbot also run on develop?13:37
@sonney2kI thougth we changed that13:37
wikingsonney2k: not yet :P13:37
@sonney2kwiking, nice one with the matrix :)13:37
@sonney2kwiking, no I think I did13:37
wikingsonney2k: hahaha yeah htat was fridayafternoon rushtime ;)13:38
wikingsonney2k: well they are not building afaik13:38
wikingsee for example:
@sonney2kit says branch='develop'13:38
-!- votjakovr [] has joined #shogun13:39
wikingsonney2k: yeah13:39
wikingsonney2k: what i dont understand for exmaple13:39
wikingis why the last build then is april 0713:39
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]13:39
wikingapril 513:39
-!- shogun-buildbot_ [] has joined #shogun13:39
wikingand that's still on master branch13:39
-!- shogun-buildbot_ [] has quit [Client Quit]13:39
-!- shogun-buildbot [] has joined #shogun13:39
@sonney2kshogun-buildbot, status13:40
shogun-buildbotbsd1 - libshogun: idle, last build 359h58m40s ago: build successful13:40
shogun-buildbotcyg1 - libshogun: idle, last build 163h03m46s ago: build successful13:40
shogun-buildbotdeb1 - libshogun: idle, last build 416h38m00s ago: build successful13:40
shogun-buildbotdeb2 - static_interfaces: idle, last build 299h58m08s ago: build successful13:40
shogun-buildbotdeb3 - modular_interfaces: idle, last build 138h33m54s ago: build successful13:40
shogun-buildbotnightly_all: idle, last build 10h00m23s ago: build successful13:40
shogun-buildbotnightly_default: idle, last build 9h11m58s ago: failed test13:40
shogun-buildbotnightly_none: idle, last build 10h25m02s ago: build successful13:40
shogun-buildbotrpm1 - libshogun: idle, last build 162h11m15s ago: build successful13:40
shogun-buildbotubu1 - libshogun: offline13:40
wikingok i've started ubu113:40
@sonney2kit is weird13:40
wikingbut yeah apart from nightly13:40
@sonney2kit certainly lists develop in there13:40
wikingthey are not building13:40
@sonney2kso I guess sth is wrong13:41
@sonney2kthe poller uses develop13:41
@sonney2kand the fetcher too13:41
-!- vikram360 is now known as weak_learner13:41
-!- weak_learner is now known as vikram36013:41
HeikoSsonney2k: when is the parameter version increased?13:42
HeikoSupon release?13:42
HeikoSsince I need to tell the thing when the parameter was added but PARAM_VERSION is still at one13:42
HeikoSmmh, this stuff was broken, thats why we did not increase, I remember13:43
HeikoSat some point this should be fixed, we always break compitability with previous version otherwise when we change classes13:43
@sonney2kHeikoS, then increase it now13:43
HeikoSsonney2k: problem is then serialization stops working13:43
wikingthat's weird13:43
HeikoSremember the bug?13:43
@sonney2kHeikoS, no13:44
wikingthat last none-nightly build was Apr 1613:44
@sonney2ksry but got the catch a plane13:44
wikingby deb3 - modular_interfaces13:44
HeikoSsonney2k:  we had this discussion13:44
HeikoSsonney2k: bye13:44
wikingand ever since then it's not building13:44
saketSG_PRINT("1") ; m_temp_lhs = CDenseFeatures<float64_t>(p1, m_dimensions, 1); SG_PRINT("2") ; m_temp_rhs = CDenseFeatures<float64_t>(p2, m_dimensions, 1); SG_PRINT("3") ; m_distance->init(&m_temp_lhs, &m_temp_rhs) ; SG_PRINT("4") ; return m_distance->distance(0, 0) ;13:45
wikingokeeey let's see if unittesting will run13:45
saketI am getting Segmentation fault in m_distance->init(...) .any idea why?13:46
saketp1 and p2 are float64_t *13:46
shogun-notifier-shogun: Heiko Strathmann :develop * 74c1f6a / src/shogun/labels/DenseLabels.cpp:
shogun-notifier-shogun: removed parameter from class to make tests work again13:48
shogun-notifier-shogun: Heiko Strathmann :develop * c1d2bdf / src/shogun/labels/DenseLabels.cpp:
shogun-notifier-shogun: Merge pull request #1002 from karlnapf/develop13:48
shogun-notifier-shogun: fix integration tests13:48
HeikoSwiking: this fixed the integration tests locally here13:49
HeikoSsorry for the mess13:49
wikingHeikoS: ok lets see how that effects travis :P13:49
HeikoSwiking: indeed :)13:49
wikingHeikoS: ok unit testing works13:51
HeikoSwiking: aweseom!13:51
-!- rgirish28 [~rgirish28@] has joined #shogun14:01
votjakovrHi everyone :)14:02
-!- travis-ci [] has joined #shogun14:04
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []14:04
wikingmmm i think i'll have to hack something with ruby...14:06
-!- wiking [] has quit [Changing host]14:06
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun14:06
votjakovrLooking through the code of GP framework, I don't understand the idea, why base class CLikelihoodModel has a protected field m_df (degrees of freedom)... I think it's better to declare it, where it really needs (for example in CStudentsTLikelihood). What do you think about that?14:10
@iglesiasgvotjakovr: better in what sense?14:11
@iglesiasgI don't know much about the GP framework we have, so just asking out of curiosity14:12
wikingjacob wrote that code... but i haven't seen him around ever since gsoc'12 finished14:14
votjakovriglesiasg: I think that base class should contain properties, which are common for all derived.14:14
votjakovrBut degrees of freedom aren't used in GaussianLikelihood, for example14:17
@iglesiasgvotjakovr: yeah, that makes sense. But if all the derived classes have some degrees of freedom, doesn't it make sense to have it there? Even if the number of degrees of freedom is different among them14:17
@iglesiasgvotjakovr: aham I see14:17
-!- gsomix [~gsomix@] has joined #shogun14:20
-!- travis-ci [] has joined #shogun14:23
travis-ci[travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []14:23
-!- rgirish28 [~rgirish28@] has quit [Quit: rgirish28]14:33
vikram360Hi, I'm just curious to know what the git workflow of some of the veteran shogun developers is.14:33
votjakovriglesiasg: If all derived classes from has degrees of freedom field, I think it's better to have it in base class. But it's not true in our case.14:33
@iglesiasgvotjakovr: ok14:39
votjakovriglesiasg: So, I'd like to remove it from base class and declare in derived. What do you think?14:39
@iglesiasgvotjakovr: how many derived classes are there that use that member?14:40
votjakovriglesiasg: just one14:40
@iglesiasgvotjakovr: then yeah, it could make sense to have it as you are saying14:41
votjakovriglesiasg: thanks, then i'll do it in next PR :)14:42
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:45
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.] has joined #shogun14:59
saketlisitsyn: Hi, yesterday we talked out multiple v_array classes in the package ... we resolved it using a deeper namespace... but then each time i do "sudo make install" I have to add changes to auto-generated file /src/interfaces/modshoung_wrap.cxx about the namespaces15:00
saketlisitsyn: do you know how to avoid this ?15:00
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun15:00
@lisitsynsaket: no, not really15:00
@iglesiasgsaket: have you tried doing make clean and building all over again?15:01
@iglesiasgsometimes that helps15:01
saketyes i tried15:01
saketdidn;'t work15:01
@lisitsynsaket: where the second one v_array is used?15:02
saketin /classifer/vw/somefiles here15:02
saketI think there were a few more .. I am unable to recall now .. was fixing them as and when they showed in the errors15:03
@iglesiasgsee you later guys15:03
-!- iglesiasg [] has quit [Quit: leaving]15:04
-!- rgirish28 [~rgirish28@] has joined #shogun15:04
vgorbatilisitsyn: hello, did you have time to review the PR?15:04
@lisitsynvgorbati: no sorry15:04
-!- rgirish28 [~rgirish28@] has quit [Read error: Connection reset by peer]15:04
saketlisitsyn: /io/AsciiFile.h too15:04
@lisitsynsaket: would that make sense to remove it from jlcovertree?15:05
-!- rgirish28 [~rgirish28@] has joined #shogun15:05
saketI think the other(jlcov...)  is also used in other files15:06
vgorbatilisitsyn: for what?) btw, what about the issue "Develop an adapter for t-SNE" - I am planning to start working on it today? I mean, is it still open?15:07
@lisitsynvgorbati: what for what?15:07
vgorbatilisitsyn: sorry for what?)15:08
vgorbatilisitsyn: it isn't actually a question)15:09
@lisitsynvgorbati: that I hadn't find time for that yet15:09
vgorbatilisitsyn: I meant to say that its okay)15:09
saketlisitsyn: Apart, @iglesias said you will be able to tell better for the PR( ) in the third last comment15:09
@lisitsynsorry guys quite overwhelmed right now - will have some time tonight15:10
saketlisitsyn: I sent a new one on develop branch as per chat with @sonney2k ..can you please look into it too ( )15:10
saketlisitsyn: okay... will catch you after some time :-)15:11
-!- madhura [~madhura@] has joined #shogun15:36
-!- madhura [~madhura@] has quit [Ping timeout: 272 seconds]15:52
-!- rgirish28 [~rgirish28@] has quit [Quit: rgirish28]15:53
-!- madhura [~madhura@] has joined #shogun15:53
wikingHeikoS: ok we are back at square one: setting 1/215:55
HeikoSwiking: this was the state we were in before right? :)15:55
HeikoSserialisation needs a fix in order for us to be able to add new parameters to classes15:55
HeikoSthere are things broken15:56
HeikoSbut at least now we can detect things15:56
HeikoSwhich  makes it much easier to fix15:56
wikingHeikoS: ok but how to detect this error, since as far as i understand it fails only on travis :S15:57
HeikoSwiking: oh really?15:57
wikingor anybody else gets this error?15:57
HeikoSI talked about the other problems15:57
HeikoSno idea15:57
wikingHeikoS: does it fail for you? in the integration test?15:57
HeikoSlemecheck15:57 setting 1/215:58
wikingin integration test15:58
HeikoSI dont have json installed so yes fails15:58
wikingit's a json test?15:59
-!- madhura [~madhura@] has left #shogun ["Leaving"]16:00
wikingis it tests/serialization_complex_example0.txt or tests/serialization_complex_example1.txt16:02
wikingor it's none of that?16:04
wikingi really dont get what's happening in that testsuite :P16:04
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:07
-!- rgirish28 [~rgirish28@] has joined #shogun16:14
-!- vikram360 [~vikram360@] has quit [Quit: Leaving]16:17
-!- gsomix [~gsomix@] has quit [Read error: Connection reset by peer]16:31
-!- rgirish28 [~rgirish28@] has quit [Quit: rgirish28]16:38
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:41
-!- shogun-notifier- [] has quit [Quit: transmission timeout]16:48
-!- sumit [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:54
-!- dvalcarce [c1903df0@gateway/web/freenode/ip.] has joined #shogun17:20
-!- lisitsyn [] has quit [Quit: Leaving.]17:23
-!- towelenee [~towelenee@] has joined #shogun17:27
toweleneehello, I would like to know do you need any contribution in decision trees?17:28
-!- vikram360 [~vikram360@] has joined #shogun17:29
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun17:55
-!- jaypanda [0e8b5206@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:01
-!- nbbk [653febbb@gateway/web/freenode/ip.] has joined #shogun18:11
-!- bharath_kmr [~bharath@] has joined #shogun18:24
-!- nbbk [653febbb@gateway/web/freenode/ip.] has quit [Quit: Page closed]18:25
-!- iglesiasg [] has joined #shogun18:33
-!- mode/#shogun [+o iglesiasg] by ChanServ18:33
-!- bharath_kmr [~bharath@] has quit [Ping timeout: 245 seconds]18:34
-!- bharath_kmr [~bharath@] has joined #shogun18:34
@sonney2khurray I am back online :D19:03
@iglesiasgthe captain is back \o/19:03
@sonney2kiglesiasg, I am more close to you than ever now19:04
@sonney2kdirect view to the eiffel tower19:04
@iglesiasgsonney2k: where are you?19:04
@iglesiasgyou know I am not in Spain right?19:04
@iglesiasgnor in France hehe19:04
@sonney2kiglesiasg, ah forgot19:05
@sonney2kbut if you were paris is closer to spain than berlin :D19:05
@sonney2kso I cann assume I am more close to dvalcarce  now19:05
@iglesiasgyou are close to dvalcarce19:05
@iglesiasgyeah exactly, if he is in coru?a I mean19:06
@sonney2kwiking, HeikoS - it shouldn't be a json thing19:07
@sonney2kIIRC I disabled json altogether19:07
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:27
-!- vikram360 [~vikram360@] has quit [Ping timeout: 256 seconds]19:38
HeikoSsonney2k: its just on my computer, I dont have json installed so the integration tests fails complaining about not finding json19:39
-!- lisitsyn [~blackburn@] has joined #shogun19:40
-!- mode/#shogun [+o lisitsyn] by ChanServ19:40
HeikoSsonney2k: btw my name is now registered with the nickserver19:45
-!- mode/#shogun [+o HeikoS] by ChanServ19:46
@HeikoSnot logged in :19:47
-!- HeikoS [] has left #shogun []19:47
-!- HeikoS [] has joined #shogun19:47
-!- mode/#shogun [+o HeikoS] by ChanServ19:47
-!- vgorbati_ [d4029f22@gateway/web/freenode/ip.] has joined #shogun19:50
-!- van51 [] has left #shogun []19:56
-!- HeikoS [] has quit [Quit: Leaving.]19:58
-!- bharath_kmr [~bharath@] has left #shogun []20:04
-!- towelenee [~towelenee@] has quit []20:07
-!- mikhailBelous [~towelenee@] has joined #shogun20:08
-!- gsomix [~gsomix@] has joined #shogun20:10
gsomixsonney2k, but you are farther from Russia. Go back sooner. :)20:16
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun20:16
saketrandom question: is there a way other than using SG_PRINT to debug the code ?20:16
gsomixsaket, gdb, valgrind >:320:16
saketgsomix: for using gdb what parameters i need to pass to make ?20:17
gsomixsaket, I dunno. I just did "gdb example_that_covers_my_code".20:20
saketgsomix: which interface do you use ?20:21
gsomixsaket, native c++. :)20:21
saketgsomix: okhay, thanks :) anyidea if we can debug through python interface ?20:23
gsomixsaket, pdb, maybe.
saketgsomix: thanks will ave a look20:24
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun20:30
-!- votjakovr [] has quit [Quit: ChatZilla 0.9.90 [Firefox 20.0.1/20130409194949]]20:35
vgorbati_lisitsyn: hello, are you here?20:36
@lisitsynvgorbati_: yes20:36
vgorbati_lisitsyn: where can I take a look at tapkee implementation of t-SNE?20:36
@lisitsynvgorbati_: it is not merged yet I think20:37
@lisitsynvgorbati_: okay about your PR20:37
@lisitsynvgorbati_: does it work?20:37
vgorbati_lisitsyn: the test passes, if you are talking about it)20:37
@lisitsynwhat is the intersection rate?20:38
@lisitsynokay let me merge latest tapkee now20:38
vgorbati_lisitsyn: wait a second20:38
vgorbati_lisitsyn: when initial n_dimension=4, target=3, neigbor_count=40 (total_samples=100), the minimal intersection rate is 25/4020:42
vgorbati_lisitsyn: average is about 30/4020:42
@lisitsynalright that looks more or less ok for me20:42
vgorbati_have you looked at PR already?20:43
-!- dvalcarce [c1903df0@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:46
@lisitsynvgorbati_: I have a major issue there - you didn't set up your git properly20:46
vgorbati_what did I missed?20:47
@lisitsynvgorbati_: ah no sorry20:47
@lisitsynthat looked like an email20:48
-!- pararth [0e8b6149@gateway/web/freenode/ip.] has joined #shogun20:49
@lisitsynvgorbati_: didn't you reinvent ?20:49
vgorbati_lisitsyn: I thought of it, but then decided to optimize a little) I mean, set_intersection would have required a temporary set..20:50
@lisitsynI see20:50
vgorbati_it would me more readable though)20:51
vgorbati_should I change to set_intersection?20:51
@lisitsynno, that's ok for me20:51
@lisitsynI am a bit concerned with the randomization of that test20:52
@lisitsynit is out of kinda 3 sigmas of course but it may fail20:52
@lisitsynI mean there is somewhat may be 1e-5 probability of failure20:53
vgorbati_ah, you mean that the min_similarity level is close to min matches20:53
@lisitsynand in general20:53
@lisitsynthe graph *may* get a few connected components20:53
vgorbati_so that neighborhood graph won't be connected?20:54
@lisitsynit may happen with non-zero Pr20:55
-!- pararth [0e8b6149@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:55
vgorbati_can you please explain me (if it is not too long to explain), what does 'neighborhood graph is not connected' means when we are talking about vectors?20:56
@lisitsynjust imagine a graph that has edge from i to j if and only if jth vector is in k nearest neighborhood of ith vector20:56
vgorbati_got it, so if we have to clusters with lets say 2*k points in each, that are far from each other - it would be two connected components?20:58
-!- rgirish28 [~rgirish28@] has joined #shogun20:58
vgorbati_two cluster*20:58
@lisitsynand we are in trouble like in taylor swift song20:58
-!- rgirish28 [~rgirish28@] has quit [Client Quit]20:59
vgorbati_:) okay, than should I change the test to predefined values?20:59
@lisitsynvgorbati_: no I think we are ok currently21:00
vgorbati_lisitsyn: so what do you suggest? Leaving the test as it is?21:01
-!- rgirish28 [~rgirish28@] has joined #shogun21:01
@lisitsynvgorbati_: yeah we can improve later I think21:01
vgorbati_so, can the issue be considered as closed?21:02
-!- shogun-notifier- [] has joined #shogun21:03
shogun-notifier-shogun: :develop * 7dbc220 / tests/unit/converter/
shogun-notifier-shogun: Added unit test for locally linear embedding method, that checks, whether it preserves neighbors (according to
shogun-notifier-shogun: :develop * c20178b / tests/unit/Makefile:
shogun-notifier-shogun: Modified unit tests makefile, so that it adds GTEST_DIR and GMOCK_DIR to include paths, when compiling unit tests - since they do include gtest and gmock files.21:03
shogun-notifier-shogun: :develop * 2d7d574 / / (7 files):
shogun-notifier-shogun: Merge branch 'develop' of into vgorbati-develop21:03
shogun-notifier-shogun: :develop * eb32fa6 / / (8 files):
shogun-notifier-shogun: Merge branch 'develop' of into vgorbati-develop21:03
shogun-notifier-shogun: :develop * 4786c8c / tests/unit/converter/
@lisitsynvgorbati_: closed, merged21:03
shogun-notifier-shogun: Modified unit test for Locally Linear Embedding, so that it allows 'misses' of neighbors before and after the dimensionality reduction.21:03
shogun-notifier-shogun: Sergey Lisitsyn :develop * 6debd28 / tests/unit/Makefile,tests/unit/converter/
shogun-notifier-shogun: Merge pull request #990 from vladislav-horbatiuk/vgorbati-develop21:03
vgorbati_wow, so many lines)21:03
vgorbati_lisitsyn: about t-SNE: I looked at (embedtDistributedStochasticNeighborEmbedding method) - is that the method that will be used?21:05
@lisitsynvgorbati_: exactly21:05
@lisitsynvgorbati_: I am now merging the latest code and then you can go21:05
vgorbati_lisitsyn: great21:05
vgorbati_lisitsyn: also wanted to ask: do you have some method of finding the distance between strings in shogun? I did looked a bit at distance folder, but still not sure21:07
@lisitsynvgorbati_: the usual way is to use kernel distance21:07
@lisitsynvgorbati_: simples is exp{-k(x,x')/width}21:08
@lisitsynvgorbati_: one may use the hamming distance of course21:08
@lisitsynbah! proposals are now accepted21:09
vgorbati_what about sequence alignment?21:09
vgorbati_when we have the penalty for mismatched letter and the penalty for inserting a gap21:10
vgorbati_and find the optimal way to insert gaps in both strings21:11
vgorbati_then the resulting penalty would be a distance21:11
@lisitsynvgorbati_: you are talking about some alignment kernel I guess21:11
@lisitsynI am not an expert in that unfortunately as I am more into the computer vision21:12
vgorbati_I just recently have heard about the algorithm, so I decided to pretend smart:)21:13
-!- mikhailBelous [~towelenee@] has quit []21:13
vgorbati_however, the algorithm is quadratic in a maximum length of both strings, so I am not sure whether its suitable21:14
@lisitsynvgorbati_: check shogun/kernel/string if you are interested21:15
@lisitsynvgorbati_: it plays its role in bioinformatics21:15
vgorbati_yeah, its used for DNA sequences alignment21:16
vgorbati_wow, that's quite a lot of kernels)21:16
@lisitsynvgorbati_: just as an example - genes are interrupted by introns in RNA of such creatures like humans21:17
@lisitsynerr DNA21:17
@lisitsynwhatever :D21:17
@lisitsyn[gene region] [kawabanga! intron] [gene region continued]21:17
@lisitsynI am afraid to let sonney2k die from laugh with my lame things but it is kind of that IIRC21:18
naywhayaresonney2k: quick GSoC question that I can't seem to find an answer to: is it possible that a person can sign up to be a mentor, but end up without a student because nobody was interested in his particular project?21:18
naywhayareit would seem unreasonable if that weren't the case, but I can't seem to find it in writing anywhere...21:18
@lisitsynnaywhayare: I think you'd have to ask carol directly21:19
vgorbati_lisitsyn: :) sonney2k is biologist?21:19
naywhayarelisitsyn: you think so?  I imagine this would be a fairly commonplace situation21:19
@lisitsynnaywhayare: ahh sorry21:19
@lisitsynI misread21:19
@lisitsynnaywhayare: I thought you asked about mentor -> student transformation21:19
@lisitsynvgorbati_: former bioinformatician yes21:20
naywhayarenope, not in this case :)21:20
@lisitsynnaywhayare: yes that's totally ok21:20
@lisitsynhe won't be a mentor though21:20
@lisitsynwon't receive t-shirt etc :)21:20
naywhayareI figured.  I have some people who would like to be mentors, but only for the right student and the right project21:20
@lisitsynnaywhayare: we have 14 mentors - you know we won't receive 14 slots :)21:21
naywhayarewow, 14? nice.21:21
@lisitsynnaywhayare: btw what's up with your project? how is it going?21:22
-!- travis-ci [] has joined #shogun21:26
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []21:26
naywhayarelisitsyn: it's going pretty well.  it takes a lot of time to answer all the emails I get21:27
@lisitsynnaywhayare: hah I feel your pai21:27
naywhayaremlpack is not a particularly mature project, so, in some ways it hasn't really developed a community, so the influx of GSoC applicants is quite a shock21:28
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Quit: Page closed]21:28
naywhayareI don't work on weekends, so Mondays have been pretty brutal; I show up and I have like 20 emails to answer...21:29
@lisitsynnaywhayare: yeah I had to answer somewhat 10 on sunday21:30
vgorbati_I found it - :)21:31
naywhayareI'd probably have an easier time if I answered them on weekends too, but I find it's really good for my stress levels to do completely different things on weekends21:32
vgorbati_lisitsyn: still there?21:54
@lisitsynnaywhayare: yes21:54
@lisitsynvgorbati_: yes21:54
@sonney2knaywhayare, sure that can happen21:56
vgorbati_lisitsyn: I am trying to write a documentation for getter and setter for perplexity parameter in t-SNE embedding class. Am I right that perplexity determines the width of gaussian 'bell' around each data point?21:56
@sonney2knaywhayare, you then just don't officially mentor21:56
@sonney2kvgorbati, sure - want to know where your genes are?21:57
naywhayaresonney2k: thanks.  that's what I figured -- I was just making sure that my mentors aren't commiting themselves to something they might now want to do21:57
vgorbati_sonney2k: you mean where are they located?)21:57
@lisitsynvgorbati_: kind of21:58
@lisitsynlet me cite21:58
@lisitsyn"The perplexity can be interpreted as a smooth measure of the effective number of neighbors"21:58
-!- utsavjain [~tiripS@] has joined #shogun21:58
vgorbati_lisitsyn: :) "Note that the perplexity increases monotonically with the variance"21:58
wikingwtf is that 1/2 test for serialization21:58
@sonney2knaywhayare, it will work like this: once you have your student applications in21:59
@sonney2kmentors can say 'I want to mentor this project'21:59
@sonney2kso you can have multiple mentors / project21:59
@sonney2kor none21:59
vgorbati_lisitsyn: ok, but how to write this in function specification? Or i can just leave 'setter for perplexity', 'getter for perplexity'?21:59
@lisitsynvgorbati_: yes that's ok22:00
@sonney2kwiking, yes that is crazy22:00
@sonney2kit should work yes / no22:00
@sonney2kwiking, only thing I see we can do is print out the different objects before failing22:00
@sonney2kvgorbati, yes22:00
wikingsonney2k: but what does actually happen in that script?22:00
@sonney2kvgorbati, where on DNA22:01
@sonney2kwiking, what do you mean?22:01
@sonney2kwiking, the big picture?22:01
wikingsonney2k: in that exact script22:01
wikinglisitsyn: wtf man22:01
@sonney2kwiking, just runs the different settings22:01
wikinglisitsyn: BAD22:01
wikinglisitsyn: should not merge!22:01
@sonney2kit is calling that function in each python script22:01
@lisitsynwiking: makefile?22:01
@sonney2kwith all the parameter settings in that script22:02
@sonney2kfor each parameter setting it is just python.pickling the results22:02
wikinglisitsyn: see my comment22:02
@sonney2kthen is doing the same again - and loads the serialized reference data from disk comparing things22:02
wikinglisitsyn: so somebody fix this22:03
@lisitsynwiking: I see - I thought it was missed22:03
wikinglisitsyn: it was not22:03
wikingwho's vladislav-horbatiuk22:03
vgorbati_wiking: are you sure? When I tried it without -I GTEST_DIR ..., it gave me an error: 'gtest/gtest.h' - not found22:03
wikingvgorbati_: that was *your* fault22:03
vgorbati_wiking: its me22:03
wikingvgorbati_: should have checked those variables before doing this patch22:04
wikingvgorbati_: r u sure ./configure found your gmock and gtest src?22:04
wikingfrom the configure:22:05
wikingso i'm pretty sure that it was there22:05
wikingand since for everybody else who uses gmock/gtest this was working22:05
shogun-notifier-shogun: Sergey Lisitsyn :develop * f2dae9d / tests/unit/Makefile:
shogun-notifier-shogun: Removed unnecessary compilation include paths22:05
wikingi'm pretty sure that there is something wrong at your end22:05
vgorbati_wiking: I did not put my gtest/gmock at default paths22:05
@lisitsynwiking: alright solved22:06
wikingvgorbati_: then you should have told that to the configure script, right?22:06
wikinglisitsyn: thnx22:06
wikingvgorbati_: ./configure --gmock=... --gtest=...22:06
wikingand check what configure's output is actually for gmock and gtest22:06
wikingi.e. if it finds it or not22:06
vgorbati_wiking: now I get it, sorry for that much text22:06
@lisitsynoutrageous wiking with an axe is killing people22:07
gsomixsonney2k, there? how long are you going to stay in France?22:07
wikingsonney2k: ok so am i right if i assume that it fails with this particular serialized model: data/testsuite/tests/serialization_complex_example0.txt?22:08
@sonney2kgsomix, until wednesday22:08
wikinglisitsyn: indeed22:08
@sonney2kwiking, yes22:08
gsomixsonney2k, ok, thanks.22:08
wikingsonney2k: mmmhm... i'll try to somehow reproduce this error :S22:08
wikingsonney2k: which part of .fr?22:09
@lisitsynwiking: he is looking at the eiffel tower22:10
@sonney2kand if one runs -d it would give you a python shell when things deviate22:10
@sonney2kso you have two variables a  and b then22:10
@sonney2kwhich are lists of what got returned22:10
@sonney2kwiking, the problem is that I cannot reproduce the issue here22:10
@sonney2kwiking, paris22:10
@sonney2kyeah I am at this event here22:10
vgorbati_lisitsyn: theta and perplexity are the only parameters for t-SNE, aren't they?22:11
@lisitsynvgorbati_: yes22:11
@lisitsyntheta is even not the t-sne but barnes-hut-sne22:11
@iglesiasgsonney2k: nice event!22:11
naywhayaresonney2k: okay, thanks for the info22:11
wikingsonney2k: ah cool! although paris i hate :P22:12
@lisitsynhaha I put russian elections 2011 to the t-SNE22:12
wikinglisitsyn: :DDDDDD22:12
wikingsonney2k: have u managed to figure out wtf is going on with the buildbots? or no time22:12
@sonney2kno time and internet connection of this hotel is too unstable22:13
wikinginternet in a hotel in .fr is always shiat22:14
vgorbati_lisitsyn: I thought theta is basically parameter of all gradient-descend-like methods22:14
@sonney2knot sure if there is some boring session tomorrow - then I would fix it22:14
wikinglisitsyn: that's a brain? :DDD22:14
@lisitsynwiking: usual picture for t-sne22:14
@lisitsynvgorbati_: yes, theta=0 is t-sne22:18
@lisitsyntheta \ne 0 is barnes-hut22:18
vgorbati_so for t-SNE there's only one parameter - perplexity?22:20
vgorbati_ah, we are actually using 'barnes-hut' in the end, so theta is used22:21
@lisitsynvgorbati_: yes for pure t-sne only perplexity matters22:22
@lisitsynit is kind of exact22:22
vgorbati_lisitsyn: ok, going to sleep now, hope will finish it by tomorrow22:24
@lisitsynvgorbati_: alright22:24
-!- vgorbati_ [d4029f22@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:24
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:28
-!- gsomix [~gsomix@] has quit [Ping timeout: 245 seconds]22:35
-!- travis-ci [] has joined #shogun22:40
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []22:40
-!- rgirish28 [~rgirish28@] has quit [Ping timeout: 252 seconds]22:41
-!- iglesiasg [] has quit [Quit: leaving]22:43
-!- van51 [] has joined #shogun22:45
-!- van51 [] has left #shogun []22:46
-!- iglesiasg [] has joined #shogun22:48
-!- mode/#shogun [+o iglesiasg] by ChanServ22:48
-!- van51 [] has joined #shogun22:48
van51hello all23:13
@iglesiasghey van5123:13
van51i am trying to tackle an entry task23:13
van51can you tell if it is possible to do a partial build?23:13
van51and if yes, how?23:14
@iglesiasgvan51: what do you mean with a partial build_23:15
@iglesiasgnot to build all shogun modules_23:15
@iglesiasg_ -> ?23:15
van51to re-build just the components that I have changed23:15
@iglesiasgno, unfortunately we have not such an option23:16
@iglesiasgwell, if you are talking about re-building23:16
@iglesiasgas long as you have not modified headers, make will do that automatically23:16
van51oh ok23:16
van51I had modified a header last time and that's why it did it all over again I guess23:16
@iglesiasgif you configure with --disable-optimization23:16
@iglesiasgit compiles faster too23:17
@iglesiasgand also using a compiler cache23:17
van51nice, good to know!23:17
@iglesiasgvan51: yeah, that's the case when you modify a header23:17
@iglesiasgvan51: I really want to see cmake integration in shogun so we could manage this partial builds nicely23:17
van51iglesiasg: I wish I knew what you were talking about :p23:18
van51it's my first time "developing" on such a large scale framework23:18
@iglesiasghehe :)23:18
@iglesiasghave you heard of cmake?23:18
van51I believe I have used it23:19
van51but I was just copying commands23:19
@iglesiasgso the partial buil issue you were talking about23:19
@iglesiasgcan be handled well with cmake23:19
@iglesiasgsee for example PCL (another open source project)23:19
@iglesiasgwhat you do there instead of doing a configure23:19
@iglesiasgyou do cmake23:19
@iglesiasgand that creates the Makefile23:20
@iglesiasgalso you can configure cmake with its cousin ccmake23:20
@iglesiasgccmake gets you to a nice ncurses interface23:20
@iglesiasgand you may be able to choose options, such as the modules you want to build23:20
van51I see23:21
van51and would it be difficult to make this transition?23:21
@iglesiasglet's say not easy :)23:21
@iglesiasgI guess one would require to learn some cmake first23:22
van51It could be an upcoming project23:22
@iglesiasgand study shogun structure to some depth23:22
van51thank you for the information23:24
@iglesiasgnp :)23:24
-!- van51 [] has left #shogun []23:51
--- Log closed Tue Apr 23 00:00:28 2013