Open in new window / Try shogun cloud
--- Log opened Fri Aug 03 00:00:02 2012
--- Day changed Fri Aug 03 2012
blackburnn4nd0: will you be at barcelona in february? :D00:00
n4nd0blackburn: no, I don't think so, why?00:00
n4nd0NIPS there?00:00
blackburnn4nd0: no some shity conference I hope to go through00:00
n4nd0I think I remember that there was going to be a conference taken place there00:00
blackburnbut I have no paper actually and have only one month00:00
n4nd0what's the name of it?00:00
blackburnICPRAM00:01
@sonney2kn4nd0, we need to talk a bit about release/stabilizing/documentation/nice examples and future involvement at that meeting00:01
blackburnI am pretty interested who hates us and do not want to continue and who loves and want to00:02
blackburn:)00:02
n4nd0haha00:02
@sonney2kblackburn,00:02
@sonney2khah00:02
@sonney2khate love relationship :D00:02
blackburnwe mostly support binary you know00:03
blackburnso I start to think binary too00:03
n4nd0blackburn: btw, what is it that should be prepared as a code sample for google?00:03
@sonney2kn4nd0, bern4rd did some great job with the website btw... I hope I will have some more time to finish it up :D00:03
blackburnn4nd0: ah there will be some script around00:04
@sonney2kn4nd0, don't worry blackburn has some script for that :D00:04
blackburnso you just pack everything you've done00:04
blackburnwould be better if they ask for blood sample00:04
blackburnor your foot00:04
blackburnno idea how to use code samples00:04
@sonney2kluckily only foot00:04
@sonney2knot finger00:04
@sonney2kotherwise coding gets tough *eg*00:04
n4nd0lol00:05
blackburnwhy not? we need you to send us your foot by mail to exhibit you have done something00:05
blackburnI found that fair00:05
blackburnsonney2k: n4nd0: did you know one gsoc student made that to london 2012?00:06
blackburnnot the foot thingy I mean00:06
blackburniirc he will participate in some 20k wlak00:06
blackburnwalk00:06
n4nd0blackburn: wow coder + professional sportsman/woman00:07
n4nd0nice00:07
blackburntrevor barron is the name00:07
blackburnmoreover he has some illness - I'd say that's how one live00:08
blackburnsonney2k: have you seen FREAK descriptor thing?00:09
@sonney2kblackburn, yes00:09
blackburndid you ever try?00:10
blackburnI am going to00:10
@sonney2kblackburn, I meant olympics...00:10
blackburnheh00:10
@sonney2kblackburn, FREAK - no00:10
@sonney2kn4nd0, so you do you love or hate us :D00:10
@sonney2k... or do you need a real valued output :D00:10
blackburnsonney2k: I am curious whether one can use one of string kernels to compare freaks00:11
@sonney2k-inf ... +inf :D00:11
blackburncause it is binary string00:11
@sonney2kprobably - but these may not look for the right patterns...00:11
blackburnsonney2k: well everyone uses hamming00:12
n4nd0sonney2k: I want to continue, sure00:13
n4nd0I plan to do CRFs00:13
n4nd0for the moment :)00:13
@sonney2kblackburn, at least the weighted degree kernel does too00:14
blackburnsonney2k: does what?00:14
@sonney2kblackburn, lets send lots of love or rose ponies to n4nd000:14
blackburnsonney2k: we've exchanged postcards btw00:14
n4nd0sonney2k: but I will need some backup support for theory aspects and such00:14
n4nd0I don't know if Nico or any other will be somewhat available for it?00:15
blackburnn4nd0: best way is to collaborate on research00:15
blackburnthen he will have time for you :)00:15
n4nd0I understand00:16
blackburnthat's how I bother ckwidmer_00:16
n4nd0I just need a bit of guidance, as long as a I am not completely alone, it will be ok00:17
blackburnn4nd0: I did help georg zeller with some data two weeks ago btw00:18
blackburnwe are all the big family :D00:18
n4nd0I have really enjoyed receiving help from Georg00:18
@sonney2kn4nd0, georg and nico continue research on that subject - so they will certainly help00:19
@sonney2kahh and if you get COFFIN into SO then that might even be a paper...00:19
n4nd0he have had a couple of dicussions on skype, he made lot of useful comments00:19
n4nd0sonney2k: that sounds interesting00:19
n4nd0blackburn: I count on you for that too, right?00:20
blackburnI am in in everything00:20
n4nd0:)00:20
@sonney2kn4nd0, http://memegenerator.net/instance/24349213?browsingOrder=New&browsingTimeSpan=Today00:21
n4nd0haha00:22
n4nd0sonney2k: thank you :)00:22
@sonney2kblackburn, big question is how we get shogun ready for release00:23
blackburnhardly hard00:23
@sonney2kand I mean stable :D00:23
blackburnrecalling that m$ advertisement00:24
blackburnwe are in pretty soft state00:24
@sonney2kblackburn, I am a hard guy you know ...00:25
blackburnsonney2k: good for ya :)00:26
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]00:30
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun00:30
-!- zxtx [~zv@c-24-6-91-131.hsd1.ca.comcast.net] has joined #shogun00:49
-!- zxtx [~zv@c-24-6-91-131.hsd1.ca.comcast.net] has quit [Ping timeout: 245 seconds]00:58
n4nd0good night guys01:08
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]01:09
CIA-18shogun: Sergey Lisitsyn master * r78c48e4 / (20 files in 5 dirs): Grand multitask refactoring - https://github.com/shogun-toolbox/shogun/commit/78c48e4ec97627f93ab118deee429be4cf8cc3eb01:54
-!- blackburn [~blackburn@109.226.100.87] has quit [Quit: Leaving.]02:00
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun02:08
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]02:08
-!- pluskid [6f782742@gateway/web/freenode/ip.111.120.39.66] has joined #shogun03:13
pluskidsonney2k: around?03:14
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun03:58
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]03:58
-!- cronor_ is now known as cronor03:58
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun05:49
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]05:49
-!- cronor_ is now known as cronor05:49
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun06:04
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]06:04
-!- cronor_ is now known as cronor06:04
-!- pluskid [6f782742@gateway/web/freenode/ip.111.120.39.66] has quit [Quit: Page closed]06:05
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun06:57
-!- gry [gryllida@freenode/staff/gry] has joined #shogun07:20
-!- gry [gryllida@freenode/staff/gry] has left #shogun []07:20
CIA-18shogun: Michal Uricar master * r57b2791 / (4 files): libbmrm ICP removal fixed - https://github.com/shogun-toolbox/shogun/commit/57b2791ac900e58014f99e456b7b3a687738175807:34
CIA-18shogun: Michal Uricar master * r4f7d733 / (3 files): DualLibQPBMSOSVM support for new bm solver added - https://github.com/shogun-toolbox/shogun/commit/4f7d73337f2faa6ac252a215c91a90fc5d112c8707:34
CIA-18shogun: Michal Uricar master * rabdc4fb / (2 files): libppbm - Proximal Point BMRM solver - https://github.com/shogun-toolbox/shogun/commit/abdc4fb60ceb36329d1914bfb34fa15106080bd407:34
CIA-18shogun: Soeren Sonnenburg master * r19672f3 / (6 files): Merge pull request #685 from uricamic/BM_SOL_EXAMPLE - https://github.com/shogun-toolbox/shogun/commit/19672f38bbe8935156363b1cb2d673c38ff707df07:34
-!- pluskid [~pluskid@108.171.196.83] has joined #shogun07:37
pluskidsonney2k: around?07:37
CIA-18shogun: puffin444 master * re101512 / (3 files in 2 dirs): Added FITC and visual demos. - https://github.com/shogun-toolbox/shogun/commit/e1015126bfe22765b3b4de82b136ad4ccc2eff5007:38
CIA-18shogun: Soeren Sonnenburg master * r4279e27 / (9 files in 5 dirs): Merge pull request #683 from puffin444/master - https://github.com/shogun-toolbox/shogun/commit/4279e27bb88b3074b4dcf537745285adffc2904107:38
CIA-18shogun: puffin444 master * re64e571 / (6 files in 3 dirs): FITC computations are now accurate! - https://github.com/shogun-toolbox/shogun/commit/e64e57183990457736d4d6991a45cf716f6f895c07:38
CIA-18shogun: puffin444 master * r1606ac2 / (7 files in 4 dirs): Cleaned up some of the code - https://github.com/shogun-toolbox/shogun/commit/1606ac2aa79e7b41df74852681f452fdbd5b7c4f07:38
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]07:46
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun08:06
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]08:06
-!- cronor_ is now known as cronor08:06
shogun-buildbotbuild #235 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/235  blamelist: puffin444 <walke434@msu.edu>, Soeren Sonnenburg <sonne@debian.org>08:44
-!- cronor_ [~cronor@g229081189.adsl.alicedsl.de] has joined #shogun08:45
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Read error: Connection reset by peer]08:45
-!- cronor_ is now known as cronor08:45
-!- uricamic [~uricamic@2001:718:2:1634:39be:d411:577:692b] has joined #shogun08:55
@sonney2kpluskid, now yes10:07
@sonney2kpluskid, you too?10:07
pluskidsonney2k: yeah10:07
-!- blackburn [~blackburn@109.226.100.87] has joined #shogun10:26
-!- heiko [~heiko@host86-181-156-213.range86-181.btcentralplus.com] has joined #shogun11:02
pluskidblackburn: around?11:06
blackburnpluskid: yeah11:06
pluskidblackburn: how do we paste code in shogun-tutorial with syntax highlighting?11:07
blackburnhmm11:07
pluskidpersonally, I like the minted package: http://code.google.com/p/minted/11:07
pluskidbut it requires pygments (an external python package)11:07
pluskidand you need to pass -shell-escape to latex when compiling11:08
pluskidit use external program to render the coloring11:08
blackburnheiko: what do you  think?11:08
blackburnI don't mind to use it11:08
pluskidthe result looks beautiful but slows down the compile time a little11:08
blackburnthe other alternative is listings11:08
heikopluskid is it hard to install?11:08
pluskidblackburn: I'm using it for some of my other docs, but I'm not sure other authors are wish to install pygments for this11:09
heikoThe thing is: do we really need code in the tutorial?11:09
heikowouldnt well documented examples be better?11:09
pluskidheiko: if there's no code, then what is tutorial for?11:09
heikofor explaining concepts11:09
heikoAnd refering to examples11:09
pluskidheiko: I thought shogun-tutor is for explaining concepts *within shogun* ?11:10
heikoBut actually, you are right, one could add examples in the tutorial that one explains a bit deepter11:10
pluskidheiko: "referring to examples" means to examples/undocumented/ directory?11:10
heikoyeah, I mean there should be a larger example of all things that one can do, and one could describe them in the tutorial from a greater distance, but the comments in the sourcecode are also  important for this11:12
heikobut as said, I agree that code snippets are a good thing in the tutorial11:12
blackburnprobably we would need11:13
blackburnsnippets11:13
blackburnfor describing some concepts11:13
pluskidheiko: btw, it requires pygments (usually can be installed with easy_install Pygments), and the package it self involves copying the code to some texmf dir I guess (not sure since I use my system package manager to install it)11:13
blackburnlike multiclass machines and etc11:13
heikook, well then its fine for me11:13
blackburnheiko: pygments is in ubuntu repo so no need to use easy_install11:13
heikowe should have a live version of the pdf online anyway11:13
heiko(for people that dont like to install this stuff)11:14
pluskidheiko: good idea11:14
pluskidheiko: but where to put it and how to keep it sync-ed ?11:14
heikopluskid, just make the buildbot compile it everytime its updated and then have a link to the latest pdf11:15
heikojust like the shogun doxygen11:16
pluskidheiko: OK, I see11:16
heikopluskid, could you add some example on how to use this highlighting?11:20
pluskidheiko: yes, I'm writing the multiclass part, I'll make a PR later today with some examples11:20
heikocool11:21
@sonney2kpluskid, heiko - I would suggest to put examples only in the examples folder and then have some magic to insert these examples directly in the doc11:46
@sonney2kif it is only short snippets of code11:46
@sonney2kthen things are different of course11:46
@sonney2k(not full examples)11:46
@sonney2kthunderstorm again!11:47
pluskidsonney2k: I'm trying to use a full example and explain part by part in the tutorial, any suggestion for this?11:47
@sonney2kpluskid, I tried the same in shogun doc on the website...11:48
@sonney2kit partially includes examples from shogun/examples11:48
@sonney2kand partially not11:48
@sonney2kno idea how to do it better11:48
pluskidhmm11:49
@sonney2kproblem with external tutorial is that it is likely to become outdated if not regularly updated11:49
pluskidexternal?11:49
@sonney2kuricamic, please fix the warning in http://shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/235/steps/compile/logs/stdio11:50
pluskidinternal tutorial will also get out-dated11:50
@sonney2kpluskid, not in code11:50
pluskidoh, get it11:50
@sonney2kpluskid, that is why we use doxygen and shogun/examples11:50
@sonney2kthis requires just the author of the code to update stuff and the buildbot will tell us about any breakage / errors11:51
@sonney2kthat is a problem in the tutorial11:51
pluskidsonney2k: I thought (and I guess this is also Cheng's idea) shogun/examples are mainly simple test cases to let buildbot to check the code11:51
@sonney2kmachine learning 'theory' is luckily not affected :D11:51
@sonney2kpluskid, nope11:51
@sonney2kthese are examples showing how to minimally use a certain machine11:51
@sonney2kbut they return some objects at the end such that we can use them for regression tests too11:52
@sonney2k(checking if results stay the same over versions)11:52
pluskidsonney2k: Cheng suggest that tutorial code work on some real world examples (like USPS), I don't think it will be afardable for build-bot to constantly run those examples, isn't it?11:52
@sonney2kwe cannot run it on every commit - maybe weekly or sth11:53
@sonney2kpluskid, we even have shogun/applications11:53
@sonney2kand already that gets outdated (because not run in tests...)11:53
pluskidoh11:53
pluskidI see11:54
pluskidhistory is a good lesson11:54
@sonney2kI guess if we had 100 developers then things might be better11:54
@sonney2k...or more chaotic :D11:54
pluskidthen you'll have no time with your kids any more :D11:54
pluskidOK, then hope the build-bot for tutorial code could be set up soon11:55
pluskidbrb11:55
-!- blackburn [~blackburn@109.226.100.87] has quit [Quit: Leaving.]11:56
-!- blackburn [~blackburn@109.226.100.87] has joined #shogun11:56
@sonney2kblackburn, any ideas why the buildbot fails - what does one need to include for eigen3?12:03
-!- blackburn [~blackburn@109.226.100.87] has quit [Ping timeout: 240 seconds]12:05
-!- gsomix [~gsomix@95.153.188.241] has joined #shogun12:08
gsomixgood day12:08
@sonney2kgsomix, hey!12:08
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun12:09
-!- cronor [~cronor@g229081189.adsl.alicedsl.de] has quit [Quit: cronor]12:11
gsomixsonney2k, my rest at sea is over.12:13
@sonney2kgsomix, thunderstorm in progress again here...12:13
@sonney2kgsomix, hope you enjoyed it...12:13
gsomixwoohoo :)12:13
-!- blackburn [~blackburn@109.226.80.43] has quit [Client Quit]12:13
@sonney2kgsomix, and I am looking forward to your PRs with the zero-memory-copy for python modular and label python buffer stuff12:14
gsomixsonney2k, I will be in the train 4-5 August. And then I will be absolutely free.12:15
gsomixufm, I'll send this PR at 6 August.12:15
CIA-18shogun: Soeren Sonnenburg master * r02d643e / src/shogun/regression/gp/FITCInferenceMethod.h : include eigen3 libs to fix compile failure - https://github.com/shogun-toolbox/shogun/commit/02d643e65ab24b36b73fa8cf8cfaa11f954be9e112:15
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun12:16
@sonney2kgsomix, long train rides...12:16
gsomixyep, over 42 hours12:17
gsomixin cool train w/o air conditioning12:17
blackburnsonney2k: you said your deutchebahns suck?12:18
@sonney2kblackburn, we sometimes have air con12:18
@sonney2kbut for sure no windows to open12:18
@sonney2kso if it is 30 C outside - pray!12:19
blackburnsonney2k: do you have to stay for 7 hours in a city for no reason?12:19
blackburn:)12:19
blackburnchelyabinsk - st.petersburg train does it12:19
@sonney2kblackburn, in germany that is not necessary - we just walk home ;-)12:19
pluskidI once was on a train for 33 hours (without a ticket for seat) with people all around12:20
blackburnwhat a bad boy12:20
pluskidI got on that train through the window actually12:20
blackburnbad^212:20
shogun-buildbotbuild #236 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/236  blamelist: Soeren Sonnenburg <sonne@debian.org>12:20
blackburnpluskid: doesn't it lead to execution in china? :D12:21
@sonney2kpluskid, you would have to ride on the roof of the ICE12:21
@sonney2k(in germany)12:21
@sonney2kbut if it runs 250km/h you certainly don't need to worry about air con :D12:21
pluskidblackburn: no, you can get on the train and then buy a ticket (a ticket without a seat of course), especially at the crowd time12:21
blackburnI see12:22
pluskidthere're soooooooooooo many people12:22
pluskidit's a nightmare to buy ticket for summer/winter holiday12:22
@sonney2kI can imagine12:22
@sonney2kpeople already complain here12:23
pluskidI once queued outside for 12 hours overnight for a ticket.12:23
blackburnI bet small city in china is a city with 'only' 2 millions of people12:23
pluskidblackburn: haha, I have no sense of numbers12:23
@sonney2kblackburn, almost like the capital of germany :D12:23
blackburnsonney2k: I thought it is sth like 5M..12:24
@sonney2kblackburn, same order of magnitude :D12:24
blackburnhmm 3.512:24
blackburnpretty small12:24
pluskidall of my friends who has been abroad would say: " there's almost no one in the street there" :D12:24
@sonney2kblackburn, buildbot fails due to not having proper ifdef have_eigen3 stuff12:24
@sonney2kthe class is not known12:24
blackburnsonney2k: CHALLENGE ACCEPTED12:25
blackburn:D12:25
@sonney2kpluskid, when visiting china only hope would be that chinese live in the 1.60-1.70m height so I can breathe fresh air above them :D12:26
blackburnhmm interesting12:26
pluskidsonney2k: sure :D12:26
blackburnI can't :(12:26
gsomixblackburn, oh no, I always remember pic with Sasha Gray and title "CHALLENGE ACCEPTED"12:26
blackburngsomix: like pics with sasha grey?12:27
blackburn:D12:27
blackburnsonney2k: I am having a strange issue with eigen312:27
@sonney2kCFITCInferenceMethod is unknown if eigen3 is not available12:28
blackburnit produces a lot of warnings when included via -I flag12:28
@sonney2kand our class list detects it12:28
blackburnwhy does it detect it then?12:28
@sonney2kso either we make some empty class available12:28
gsomixblackburn, ooops, Grey, of course12:28
gsomix^^'12:28
pluskidbtw: is the meeting in half an hour or actually tomorrow?12:29
@sonney2kgsomix, taht thing http://beerandstupidity.blogspot.de/2011/07/challenge-accepted.html12:30
@sonney2kpluskid, in 30 mins yes12:30
uricamicsonney2k: ok, I will fix it with next PR12:30
gsomixsonney2k, yes! :D12:31
CIA-18shogun: Soeren Sonnenburg master * r5e4b294 / src/shogun/base/class_list.cpp.py : make class list generator script be aware of HAVE_EIGEN3 - https://github.com/shogun-toolbox/shogun/commit/5e4b294c8dba439741de7c98693018e9d3aa8e8512:32
@sonney2kblackburn, next attempt12:32
@sonney2kpluskid, btw maintaining all the interfaces is not that much work with swig (and no library / std stuff being used in *headers*)12:33
@sonney2kactually close to none12:33
pluskidthat's cool12:33
blackburnsonney2k: I have a strange issue with eigen12:35
blackburnsonney2k: do you have it installed?12:35
@sonney2kpluskid, that is why I always insist that no one uses magic in headers ...12:35
@sonney2kblackburn, locally yes12:35
shogun-buildbotbuild #237 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/23712:35
pluskidsonney2k: OK, I see12:35
blackburnsonney2k: and you see no warnings, right?12:35
@sonney2kblackburn, tons of warnings12:36
blackburnsonney2k: from eigen?12:36
@sonney2khey a successful build12:36
@sonney2kyes12:36
@sonney2keigen has lots of shadowed variables12:36
blackburnsonney2k: they go away when I include eigen3/Eigen/Dense12:36
blackburninstead of Eigen/Dense12:36
blackburnI do not understand why12:36
@sonney2kblackburn, I once got bitten by a variable naming clash (shadowed variable) ... was hunting that bug for quite a while. since then this warning is turned on12:39
@sonney2kblackburn, do we get other warnings from eigen12:39
@sonney2kwould the eigen people 'fix' that?12:39
@sonney2kor should we disable this warning again...12:39
blackburnsonney2k: but why it goes away with different way of include?12:40
shogun-buildbotbuild #143 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/143  blamelist: Soeren Sonnenburg <sonne@debian.org>12:41
@sonney2kblackburn, no clue12:46
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has joined #shogun12:48
-!- shogun-toolbox [~shogun@7nn.de] has quit [Ping timeout: 244 seconds]12:52
--- Log closed Fri Aug 03 12:52:32 2012
--- Log opened Fri Aug 03 12:52:44 2012
-!- shogun-t1olbox [~shogun@7nn.de] has joined #shogun12:52
-!- Irssi: #shogun: Total of 15 nicks [1 ops, 0 halfops, 0 voices, 14 normal]12:52
-!- Irssi: Join to #shogun was synced in 9 secs12:52
-!- shogun-toolbox [~shogun@7nn.de] has quit [Ping timeout: 244 seconds]12:53
-!- blackburn [~blackburn@109.226.80.43] has quit [Ping timeout: 240 seconds]12:53
-!- ckwidmer_ [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds]12:53
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Ping timeout: 244 seconds]12:53
wikingmeeting in 10?12:53
chengI'm ready.12:54
-!- shogun-buildbot_ [~shogun-bu@7nn.de] has joined #shogun12:55
-!- alexlovesdata [6d2d0040@gateway/web/freenode/ip.109.45.0.64] has joined #shogun12:55
-!- n4nd0 [~nando@192.36.80.8] has joined #shogun12:56
n4nd0hey!12:56
n4nd0nice internet in the airport :)12:57
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun12:57
n4nd0and free!12:57
pluskidn4nd0: cool12:57
pluskidn4nd0: which country are you in now?12:57
n4nd0pluskid: Sweden12:57
gsomixmimimi12:58
@sonney2kn4nd0, wow you made it?12:59
@sonney2kwho is missing?12:59
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun12:59
n4nd0blackburn?12:59
@sonney2kpuffin444, welcome12:59
-!- ckwidmer [45c9b18c@gateway/web/freenode/ip.69.201.177.140] has joined #shogun12:59
@sonney2kchris12:59
-!- cronor [~cronor@141.23.76.80] has joined #shogun12:59
puffin444hello13:00
@sonney2kheiko, will arthur join?13:00
ckwidmerhi all13:00
@sonney2kn4nd0, how much time do you have at the airport?13:00
heikosonney2k, unfortunately I have no response so far, sent a text and an email, I feel bad that this always happens since we talk every few days, but I cannot do anything about it13:01
n4nd0sonney2k: I think the boarding will start in about 1h20min13:01
n4nd0so as long as the connection holds, I think I can make it for the whole meeting13:01
@sonney2kwiking, what about alex?13:02
@sonney2kalexlovesdata, there13:02
@sonney2kok13:02
@sonney2kthen lets just start13:02
@sonney2kWelcome everyone to our 4th meeting. In the last weeks we have seen various PRs from everybody and things have been progressing nicely.13:02
@sonney2kThat said, the program is coming to an end rather soon.13:02
@sonney2kWe are very proud what you all have achieved and naturally want to keep you as contributors (maybe even becoming core developers like blackburn/heiko) after GSoC.13:02
@sonney2k(it would be blackburns part here to say yes yes yes we want you)13:02
n4nd0:)13:03
@sonney2kfor some reason he got disconnected few minutes ago13:03
@sonney2kso heiko you have to take over this part :D13:03
heikolol :)13:03
heikoyes yes yes we want you,13:03
heikojoin the team, its great fun!13:03
@sonney2kwe all know real live will take over after gsoc13:04
@sonney2kso one cannot commit a lot of time but just occassionally13:04
heikoits pretty cool seeing that your stuff is being used13:06
heikoand that usually takes some time, so stay, stay! :D13:06
@sonney2kheiko did that last year and eventually became a core developer13:06
@sonney2k(we cannot mention this often enough)13:06
@sonney2kFor sure we want to have all of your code in the upcoming shogun 2.0 release.13:06
@sonney2kHowever, to be able to do a release we need to shift focus from features to stability, documentation and examples.13:06
@sonney2kSo please *everyone* help us to get the buildbot to build on all archs with close to 0 warnings and create nice examples (graphical too) such that we can illustrate the work you did.13:06
heikoand btw it opens a lot of doors in uni when you work on an open-source project13:06
@sonney2kideally you use shogun to do baseline experiments and then from time to time contribute some extensions13:07
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun13:07
heikosonney2k, it would be great if we could sort out the testing-branch issues in order to maybe have some tests for the release13:07
@sonney2kthat works well in academia and even industry13:07
shogun-buildbot_build #247 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/247  blamelist: Soeren Sonnenburg <sonne@debian.org>13:08
wikingheiko: second that... i think we should discuss it today on the meeting to have a consensus about that13:08
n4nd0haha that was very appropriate from shogun-buildbot_13:08
@sonney2kheiko, wiking ok noted down13:08
pluskidbuildbot wants to join the meeting13:08
@sonney2kpluskid, yeah it complains about my merge of puffin444's code :D13:08
blackburnehm my connection sucks13:08
@sonney2kalright13:09
@sonney2kMentors - again who wants to go to mountain view on Oct 20/21 to join the mentors summit? That event is a good place to get to know core developers from other organisations (like swig, orange, opencv, ...).13:09
blackburnckwidmer wants I am sure :)13:09
ckwidmersonney2k, same as last time: interested, but let others go if anyone else wants to13:10
shogun-buildbot_build #205 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/205  blamelist: Soeren Sonnenburg <sonne@debian.org>13:10
blackburnckwidmer: it should be easier for you this time hehe13:10
ckwidmerblackburn yes, true :)13:11
heikosonney2k, no students allowed there right?13:11
@sonney2kheiko, exactly13:11
n4nd0:(13:11
@sonney2konly mentors13:11
@sonney2kbecome a mentor next year and you two can go :D13:11
@sonney2kckwidmer, I guess it will be us two then13:12
ckwidmerheiko, mentors get paid in cool rather than money ;)13:12
ckwidmersonney2k, I'm not complaining :)13:12
heikockwidmer, yeah :)13:12
@sonney2kNow as usual, please every mentor/student team - give us a status update, telling us what you did since the last meeting, whether you are behind or on schedule and what has already been merged into shogun and what will be the final steps.13:12
blackburnckwidmer: sonney2k said he is a hard boy so be careful13:12
ckwidmerheh13:13
@sonney2kblackburn, it is not the first time I share a room with ckwidmer - so he knows pretty well ;-)13:13
blackburnI don't want to know what happened last time!13:13
blackburn:D13:13
ckwidmeryes, sonney2k snores13:13
heikolol :)13:14
@sonney2kckwidmer, that is how I code at night!13:14
@sonney2kso who wants to start?13:14
ckwidmerheh13:14
@sonney2kckwidmer, blackburn you?13:14
blackburnwhy not13:14
ckwidmersonney2k, sure13:14
blackburnokay what we've done13:14
ckwidmerblackburn, fire away13:14
blackburnI did not prepare the speech lol13:14
blackburnthere are a few multitask algorithms available now13:15
@sonney2kmaybe n4nd0 next and then vojtech/uricamic13:15
blackburnsome time was spent on making it work with xval13:15
ckwidmerblackburn, just make a rap video from your commits and upload them to youtube13:15
blackburnwe actually use these algos to predict on ovarian cancer data13:15
blackburnnamely, algorithms are L1/Lq task tree/group logistic regression/least squares regression13:16
blackburnL1/L2 following argyriou probably popular thing13:16
blackburntrace norm regularized logistic regression and clustered multitask logistic regression13:16
blackburnSLEP-based algorithms do not support xval yet but it is on the way13:16
blackburnmoreover I am working on kernel mean matching now13:17
blackburnthat's something from domain adaptation I want to implement13:17
@sonney2kblackburn, so what do you want to have in shogun 2.0?13:17
ckwidmerso seems like we have most of the popular MTL methods (non deep learning) stuff in shogun now, which is really nice13:17
@sonney2k(and what you would like to do after?)13:17
blackburnsonney2k: after gsoc I will get back to edrt and made yet another attempt13:18
blackburnas for MTL13:18
blackburnwell I hope ckwidmer will catch me on the train so we will continue to play around with data13:18
blackburnso if something cool appears I can extend MTL part13:19
blackburnactually MTL is not so useless sonney2k13:19
blackburnit appears to be best classifier on the data we work on now13:19
ckwidmerblackburn, never give up hope ;)13:19
@sonney2kblackburn, ckwidmer alright thanks - so what is missing then is documentaion / examples mostly right?13:20
blackburnsonney2k: I actually want to polish these things because they are not user-friendly yet13:20
blackburnyes, doc, examples and safety13:20
ckwidmerthe good thing is, quite a bunch of ppl here at MSK get interested in this stuff, when realizing that their data set could benefit from these methods13:20
blackburnif you pass something wrong it could kaboom13:20
@sonney2krecall that we at least intended to release Sept 1.13:21
ckwidmerwill help us to polish13:21
@sonney2kckwidmer, users are the best testers o_O13:21
ckwidmer:)13:21
ckwidmerthey will complain13:21
@sonney2kalright thanks13:21
@sonney2kI think we have to continue13:21
@sonney2kn4nd0, could you please?13:21
n4nd0sure13:22
n4nd0I implemented the HM-SVM based on the hmsvm toolbox. I have compared the model learnt13:22
n4nd0by my implementation with the one in the toolbox and the results are correct. However, it13:22
n4nd0just works with discrete data and I am currently working on a extension to handle13:22
n4nd0real data for the observations using PLiFs.13:22
n4nd0After this I will focus on improve the documentation and the examples.13:22
@sonney2kn4nd0, so plifs will still be done in gsoc and doc / example right?13:22
@sonney2kgsoc timeframe13:22
n4nd0yes13:22
n4nd0right now the plifs are taking a bit more time because I am trying to use what it is already in SHOGUN13:23
@sonney2kand in case you stick around (which is what we all hope) coffin, etc13:23
n4nd0exactly13:23
-!- alexlovesdata [6d2d0040@gateway/web/freenode/ip.109.45.0.64] has quit [Ping timeout: 245 seconds]13:23
n4nd0Unfortunately, it has not been possible to implement conditional random fields. This part was included on the initial schedule.13:23
@sonney2kn4nd0, ok thanks - questions to that anyone13:23
blackburnactually KMM would work with SO too13:24
blackburnso we can have domain adaptation structured output learning13:24
@sonney2kblackburn, hehe post shogun 2.013:24
blackburna few more words and it would sound like something cool13:24
wikingn4nd0: i'd like to have some changes in SO for supporting COFFIN fw. i dont know whether we should discuss it now or later...?13:24
n4nd0wiking: this will be something to do after gsoc13:25
@sonney2kwiking, n4nd0 after the meeting I guess13:25
wikingsonney2k n4nd0 ok13:25
n4nd0wiking: but we can discuss about it at any moment :)13:25
wikingn4nd0: hehe yeah let's do it after as that is what i need as well in latent... anyways after the meeting then if u have time13:25
@sonney2kalright next uricamic/vojtech?13:26
vojtechok13:26
uricamicok13:26
vojtechhe work Michal did since the last meeting:13:26
vojtechi. He implemented better cutting plane management for both BMRM and BM.  ii. He implemented our variant of Bundle Method for risk minimization.  iii. Finally, he has been implementing multiple-cutting plane model improvement (similar to n-slack of Joachims).   He will finish in 3-4 days. And that's it. The final steps: to collect some nice benchmarks for StructSVM, testing, documentation, examples.13:26
vojtechsorry for the bad formating13:26
@sonney2kvojtech, ok nice - I think we really need some nice/cool example for that otherwise people won't know that this is actually cool stuff13:27
vojtechI agree13:27
@sonney2kn4nd0, blackburn holds for you too13:27
n4nd0uricamic, vojtech : I got a question here13:27
vojtechgo ahead13:28
uricamicso far I have just the multiclass example, I have tried and compared to matlab implementation on MNIST data13:28
@sonney2kvojtech, uricamic any chance of continuing afterwards doing libqp business (exctracting things like pr loqo etc)?13:28
n4nd0could I use the models I have worked in my project, the HMSVMModel or the MulticlassModel, with one of your algorithms directly?13:28
wikingn4nd0: you need a risk function...13:28
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has joined #shogun13:28
n4nd0wiking: aham, I see13:29
blackburnah btw vojtech I patched libqp_gsmo a little13:29
blackburnmay be you would want to push it into your version13:29
vojtechsonney2k: I'm continously improving LIBQP and LIBOCAS if there is a substantial motivation, e.g. recently I added positivity constraint to LIBOCAS which was requested by some computer vision people13:29
uricamic n4nd0 yes, but you have to create the RiskFunction for that, anyway in that function it should be possible to use your stuff13:29
wikingvojtech: oh that's great news! i'm looking forward to use that extension in latent methods ;)13:29
n4nd0uricamic: nice, thank you!13:29
uricamicbut you have to wait for the next PR, where I will add support for user data13:29
vojtechblackburn, thanks I'll look at it13:30
@sonney2kckwidmer, could you please continue to lead the meeting13:30
n4nd0uricamic: ok13:30
@sonney2k(my kids just woke up...)13:30
ckwidmersure13:30
vojtechwiking: you can already downloaded13:30
ckwidmerI guess I have to wake up, too then (around 7AM) ;)13:30
wikinguricamic: remember the chat we had 2 weeks ago, i'll give u some changes i've done to your code to support latent variable in bmrm_data_t13:30
vojtechwiking: http://cmp.felk.cvut.cz/~xfrancv/ocas/html/index.html13:30
wikingvojtech: ok i'll then try to merge the code into shogun!13:30
blackburnvojtech: basically one needs to initialize solution that doesn't violate \sum constraint13:30
chengckwidmer: I had to cut my Friday beers short. ;-)13:30
wikingblackburn: we can do that straight away right?13:31
blackburnwiking: what?13:31
ckwidmercheng haha. Ill probably fall asleep in mine13:31
uricamicwiking: yep, I remember and I am almost finished with new design to support whatever user data you want13:31
wikingblackburn: just take the new release of libocas and update the code in shogun13:31
wikinguricamic: great let's see how our two ways collide ;)13:31
blackburnwiking: sure why not just push everything13:32
wikingblackburn: okie13:32
ckwidmerok13:32
blackburnwho's next?13:32
ckwidmerwho wants to go next?13:32
wikingi can go13:32
wikingi mean alexlovesdata and me13:32
ckwidmerok13:32
wikingif that's ok13:32
ckwidmerwikiing, sure go ahead13:32
wikingok so basically 2 things happened ever since the last meeting13:32
uricamicwiking: but anyways I would like to see your version supporting latend variables, maybe then I will modify the current version to meet better you requirements13:33
wikinga) most importantly latent implementations has been PR-ed and merged into shogun master branch. that includes LatentSVM and a preliminary LatentSOSVM. i've added a very simple example but we've agreed with alexlovesdata that we should do a more complex and nicer example13:34
ckwidmer(the others can maybe start preparing some short summary)13:34
wikingthus i'm currently working on finishing a full blown example for latentSVM and LatentSOSVM based on image-net data13:34
wikingbasically it'll be a computer vision based problem and i think that'll be a very nice example of how to use these two methods and what are they capable of13:35
wikingwhat is still missing is to finish this example and add modular interface for both of the classifiers13:35
wikingand that i add support for bmrm based LatentSOSVM13:35
blackburnwiking: I've been studying freak a little13:36
wikingb) with heiko i've created a unit testing branch on my repo to add unit testing for the c++ code in shogun because we both thought that it would be great to have it for shogun13:36
alexlovesdataand I owe you still the Hog code viking ;)13:36
blackburnit seems that it could be a good example of latent model13:36
blackburnshould pretty fast actually I mean13:36
wikingcurrently the PR for that attempt is at https://github.com/shogun-toolbox/shogun/pull/66713:36
ckwidmerwiking, ok thank you13:36
wikingthank you very much for all the comments13:36
wikingBUT13:36
@sonney2kvojtech, I would very much prefer to have all the qp solver stuff in your libqp and only use that from shogun via some standard interface - then we can (from shogun) just switch solvers on the fly for any of the chunking svms etc13:36
wikingckwidmer: just as a last thing!13:37
ckwidmerwiking, sure13:37
wikingso i think13:37
heikoblackburn, sonney2k, could you have a look on the current approach so that we can decide how to proceed with the testing today?13:37
wikingthat we should really discuss how to proceed with the unit testing13:37
wikingbecause it's there i've added examples for both simple unit testing and mocking13:37
blackburnlet me check13:37
vojtechsonney2k: ok, I have increased the priority counter for LIBQP by one13:37
wikingnow we should really try to agree on how to add this into shogun13:37
wikingbefore the 2.0 release13:38
wikingnaturally i want to add both latentSVM and latentSOSVM into 2.0 release of shogun13:38
wikingand i'm comfident that that will happen13:38
wikingi'm more 'concerned' about the unit testing13:38
blackburnone problem I see13:38
wikingsince that should be done by all of us together13:38
wiking;)13:38
wikingand yeas13:38
@sonney2kwiking, sure we do that as submodule :)13:38
wikingi've forgot13:38
blackburnwe wouldn't be able to complete all the tests13:39
blackburnbefore release13:39
heikoblackburn, but this is incremental13:39
wikingsonney2k: i've given my reasons why i think submoduling will be really bad for the whole concept of unit testing13:39
blackburnyeah13:39
@sonney2kblackburn, that is not a problem13:39
@sonney2kwiking, why?13:39
blackburnsonney2k: wiking: yes, submoduling can be a problem13:39
heikoso, it can be included in the release13:39
blackburnbecause they are not synchronized13:39
@sonney2kthey are13:39
wikingsonney2k: "my concern is here more about the whole method of how to add new code + corresponding unit tests.13:39
wikingthat's going to be at least 2 commits + 2 pushes and one extra pulls to get in sync with the submodule..."13:39
@sonney2kthats true13:40
@sonney2kthough I don't see a problem with that13:40
wikingand yeah we've try to use shogun with blackburn on a large scale image categorisation challenge13:40
wikingit's part of this years Pascal VOC13:40
@sonney2kI think we need to postpone the discussion until after this meeting though13:40
wikingso if u feel like joining the team of shogun in this13:40
wikingplease come and contribute because we need you!13:41
blackburnwiking: yeah we need to start some progresson that13:41
wikingblackburn: indeed :)13:41
heikowiking, what are deadlines and prices? :)13:41
alexlovesdataOMG, viking you should have told me about pascal VOC13:41
blackburnalexlovesdata: why?13:41
wikingalexlovesdata: now i'm saying it ;)13:41
alexlovesdataI won last years imageclef photoannotation13:41
alexlovesdataI have good features13:41
wikingalexlovesdata: :D then come and join us :)13:41
alexlovesdataI only lack disk space13:41
blackburnhah13:41
wikingalexlovesdata: i have that :)13:41
wikingalexlovesdata: 7 TB13:41
blackburnalexlovesdata: wiking has a lot of13:41
wikingif that's enough13:41
wikingsonney2k: okey13:42
alexlovesdatagreat .... we should talk after the meeting about pascal voc ;)13:42
ckwidmerhehe, ok13:42
wikingso please everybody stick around if u can for a max 10 mins13:42
ckwidmerseems like sonney2k is back13:42
wikingto join the conversation about unit testing13:42
blackburnpluskid: cheng may be you next?13:42
ckwidmerI would suggest finishing up the reports first13:42
pluskidok, shall I go?13:43
@sonney2kalright cheng / pluskid want to contiue?13:43
@sonney2kheh13:43
ckwidmerbut it's fine to discuss unit tests after that13:43
chengI volunteer pluskid to type... ;-)13:43
pluskidI implemented the RelaxedTree algorithm. It tooks a bit longer than expected due to my under-estimate of the complexity of its colorization algorithm. But I finally get it hunted. Before the end of GSoC, I'm planning to add content to the shogun-tutorial of multiclass learning and implement an algorithm from "Gun".13:43
@sonney2k(and after that puffin444 and gsomix)13:43
pluskidand that's it :)13:43
@sonney2kpluskid, considering the number of algorithms you contributed you will get lots of users13:44
chengJust a comment here that Gunnar's algorithm is basically cutting planes. Hopefully some of the other members can help.13:44
@sonney2kpluskid, dont' forget about examples (graphical if possible)13:44
pluskidsonney2k: OK, maybe we could include some nice figures in shogun-tutor13:45
@sonney2kotherwise people have no idea how to use all that stuff13:45
blackburncheng: pluskid: I can actually prepare a script for road sign recognition stuff13:45
@sonney2kand more importantly what you can do with all that13:45
blackburnthere you can measure performance13:45
chengblackburn: That's a cool app.13:45
pluskidblackburn: cool, thanks!13:45
@sonney2kpluskid, if you beat blackburn on that task he will cry for at least 1 hour :D13:45
blackburnsonney2k: I won't :)13:46
blackburnI don't care now13:46
chengblackburn: if we get it fast enough, I know some people here who want to do it real time.13:46
pluskidlol13:46
@sonney2kcheng, blackburn has it fast enough13:46
@sonney2k(linear svm...)13:46
@sonney2kwith coffin13:46
@sonney2kanyways more comments to pluskid/cheng?13:47
@sonney2kif not then thanks pluskid - as usual excellent work13:47
@sonney2kpuffin444, want to continue?13:47
chengblackburn: http://www.nicta.com.au/research/projects/AutoMap/sign_geo-positioning_service13:47
blackburnI am curious whether these cool algoritmhs would beat crammer-singer13:47
puffin444sure13:47
pluskidbtw: is cutting plane related to bundle method?13:47
pluskidI'm totally ignorance of those13:47
blackburncheng: what's technology of recognition?13:47
@sonney2kpluskid, yes - vojtech is the ueberexpert here13:48
pluskidsonney2k: ok13:48
chengblackburn: No idea, coz I'm not involved.13:48
puffin444A lot has happened in the last month. I had to focus13:48
pluskidbut I'm afraid I don't even know what to ask now (before making some preliminary reading)13:48
blackburncheng: I used HOG + homogeneous kernel map + linear svm13:49
@sonney2kpuffin444, please continue13:49
blackburncheng: 97.82% on http://benchmark.ini.rub.de/?section=gtsrb&subsection=results13:49
blackburnwould be interesting to see better results with pluskid's new algos13:50
puffin444on alot of extra work in the model selection framework to support gradient selection. added also were some ARD kenels, sparse approximtion and a simple visual demo13:50
puffin444sorry about that13:50
chengblackburn: pluskid: Definitely a cool application demo for multiclass.13:50
pluskidsure!13:51
puffin444As of now all but one feature in the schedule has bee merged.13:51
@sonney2kpuffin444, so what do you still want to do for shogun 2.0?13:52
puffin444In the coming days I plan to add laplacian aproimation for inference13:52
blackburnand the main question puffin444 would you like to continue? ;)13:52
puffin444as wel as add documentation13:52
@sonney2kheiko, see - blackburn is back :D13:52
heikosonney2k, yeah13:53
heikogo blackburn, go! :)13:53
puffin444and somemore complex examples13:53
blackburnheiko: sonney2k: where I am back to?13:53
@sonney2kpuffin444, graphical too if possible13:53
heikoblackburn, in recruiting new people :D13:53
@sonney2kI need nice figures somehow produced with the aid of shogun13:53
blackburnit sounds like you had a secret conversation about my recruiting skills :D13:53
puffin444absolutely. sorry for the typing speed. it looslie talets are a horrible idea for irc13:53
@sonney2kpuffin444, ohh ok13:54
heikoblackburn, these and other skills ;)13:54
@sonney2kpuffin444, great job!13:54
blackburnpuffin444: where did you get your 86 kg notebook?13:54
@sonney2kso gsomix - please continue13:54
gsomixsonney2k, ok13:54
puffin444directly From hp 4 years go13:54
gsomixSince last meeting I and sonney2k managed to add python protocol's stuff in our swig code.13:55
blackburnpuffin444: sorry I mean where it is now :)13:55
@sonney2kpuffin444, I guess oliver has some nice examples (graphical too) - so please ask him - maybe you can easily port things over13:55
gsomixsorry, copy-past :]13:55
gsomixWe have implemented many cool features for DenseFeatures: buffer protocol for really fast memory transactions with numpy objects and for numeric operators (a=mycoolfeatures+numpy.array); mapping and sequence protocol for numpy like slicing (e.g. mycoolfeatures[ : , some_Idx ] = some_const); numeric protocol with using buffer protocol for fast numeric inplace operators (mycoolfeatures+=numpy.array).13:55
gsomixAt now I'm working on methods for zero-memory-copy working with numpy.array and protocols for Labels, SGVecotr, SGMatrix (last two for 'director classes' stuff). Next I plan work on typemaps for modelselection (hi, heiko) and nice error messages. Btw I need clean up my code and add python3 support.13:55
gsomixAfter GSoC I plan further work with modular interfaces: more director classes, 'protocoled' classes, typemaps and examples...13:55
@sonney2knot me - all gsomix's work13:55
puffin444yes he has given some great exampes13:55
gsomixand I want work in shogun because there is blackburn (mimimi, =***)13:56
gsomixthat's all13:56
heikohi gsomix :)13:56
@sonney2kcheng, that is sth you would like13:56
@sonney2k^13:56
blackburnno optics though in shogun13:56
blackburnand NEVER will be13:56
blackburn:D13:56
@sonney2kyou can do x + y now (x shogun RealFeatures, y numpy array)13:56
blackburnckwidmer must like it too13:56
@sonney2kslicing all that works and modifiactaions too13:56
@sonney2kckwidmer, too yes13:56
blackburnreal pythonista13:57
ckwidmer:)13:57
chengsonney2k: excellent! I don't know how to program in any real language anymore...13:57
@sonney2kand the no longer copying memory around for python too13:57
gsomixbtw, I need to go. heiko, when will I be able to discuss with your about typemaps and modelselection? I'll available after 5 August.13:57
@sonney2kheiko - you should give gsomix a lecture how to create the parameter / model seection tree13:58
pluskidcan I just use numpy array in place of where RealFeatures are expected?13:58
gsomix*will be available13:58
blackburnpluskid: no not yet13:58
blackburn:D13:58
@sonney2kpluskid, but the other way round!13:58
gsomix*when you will be able to13:58
gsomixawww :)13:58
heikogsomix, ok13:58
@sonney2kpluskid, so you could use e.g. PyML or mlpy or whatever python toolbox and interact !13:58
pluskidok13:58
heikogsomix, anytime that fits you13:58
gsomixheiko, thanks! :)13:59
@sonney2kalright thanks gsomix13:59
gsomixsonney2k, heiko ok, should to go, soryy13:59
blackburnsonney2k: I actually think we could add apply(SGMatrix)13:59
blackburnand train(SGMatrix)13:59
heikogsomix, ok bye, just ping me13:59
@sonney2know please some homework - send me nice figures produced with the aid of shogun13:59
@sonney2kfor the new homepage13:59
@sonney2k(next week please)13:59
@sonney2kand then finally when shall we have our last celebration - August 16 11 UTC?14:00
@sonney2kwe will then see how many warnings are left on the buildbot14:00
heikosonney2k, any interest on whats happening with the statistics-framework?14:00
@sonney2kheiko, oops14:00
@sonney2ksorry14:00
@sonney2kheiko, please continue14:00
heikono worries :)14:00
blackburnlol14:01
heikowont take long since copy/paste14:01
blackburnnobody likes statistics14:01
heikoSince last time, I implemented dependence tests using the HSIC. As for the two-sample tests these come in multiple flavours (null-distribution approximations).14:01
heikoI did some refactoring/making things more effective for bootstrapping using precomputed kernel matrices (very important for some tests, since these are computed ~250 times)14:01
@sonney2kblackburn, no no14:01
heikoHSIC tests are done -- with examples (graphical ones to come)14:01
* sonney2k *loves* statistics14:01
heikoblackburn, but everybody like statistics with kernels :D14:01
heikoFor synthetical example data, I added a data generator class (please use it everybody!), that makes artificial data available from all interfaces.14:01
@sonney2k(blackburn - I had to say that to keep heiko as core dev)14:01
heikoA cool thing that I added for linear time two-sample tests (based on MMD) is automatic kernel weight selection in an MKL style, using libqp (thanks for that vojtech) (see NIPS paper to come, currently working on review-reply experiments with Arthur)14:01
heikoApart from that, I wrote the MMD part of the tutorial. This has to be completed with HSIC and some practical stuff.14:01
heikoPlans for next weeks: adding streaming feature class to linear time MMD, and then polish: tutorial/documentation/examples, and tests (hopefully in the new framework by wiking)14:02
heikoI have some more ideas what could be done, but I think I focus on getting things stable/nicely looking now.14:02
heiko[end]14:02
blackburnI actually like some things from statistics14:02
blackburnbut I don't get much14:02
heikosonney2k, blackburn, you can do some really cool things with it14:02
wikingheiko: i'm really looking forward to test mmd!14:02
blackburnheiko: what can I do with these cool tests?14:02
wikingsounds awesome!14:03
heikocurrent experiment here is to place music data on a sine carrier wave via amplitude modulatilon, and then try to dinstinguish two songs from the same interpret14:03
wikingheiko: actually have a dataset for that and i'd like to see the weights it finds ... as i have some other methods that generated some 'optimal' weights...14:03
heikobalckburn, tell two probability distributions apart with all the gains of "kernel methods"14:03
heikowiking, what kind of data is it and what kind of problem?14:04
-!- vikram360 [~vikram360@117.192.189.69] has joined #shogun14:04
@sonney2kheiko, what kernel do you use? gaussian or some spline?14:04
wikingheiko: i have features from different domains... thus i'm using combined kernels14:04
heikogaussian currently14:04
heikokernels need to be characteristic in order to be usefull14:04
wikingheiko: and i'd like to see an optimal weight distrib for the kernels14:04
heiko(= their range covers the complete R in the fourier domain)14:05
heikowiking, cool lets try that soon!14:05
-!- gsomix [~gsomix@95.153.188.241] has quit [Ping timeout: 246 seconds]14:05
@sonney2kok thanks heiko!14:05
@sonney2kone question about the data generators14:05
@sonney2kwhat can they produce?14:05
wikingheiko: yep i'll update shogun on that machine where the features are and i'll give it a go... it's available in modular_* interface... or?14:05
@sonney2kis there already an example for using them for e.g. python_modular?14:05
heikosonney2k, currently its really loose, just a class  where you can add (possibly static) methods that poduce data14:06
heikoI use them from my examples14:06
heikomy methods return a matrix14:06
@sonney2kheiko, and do you properly seed the rnd number generator to get things reproducible?14:06
heikosonney2k, yeah I have started paying attention to that :)14:06
blackburnhmm actually I find the idea interesting14:06
@sonney2kk14:06
blackburnone could say14:06
heikobut unfortunately its kind of hard to get it in synch with matlab14:06
blackburnjust try 4324 seed and you will see a bug14:06
heikobut for example14:07
heikoI do a seed to compare results of x-validation on normal and precomputed kernel14:07
heikohowever, thats not data related14:07
@sonney2kalright so thanks again heiko14:07
blackburnheiko: btw14:08
@sonney2know final meeting on August 16, 11 UTC - is that ok with everyone?14:08
blackburnif we had *distributed* model selection it could be a killer feature14:08
blackburndo you think so too?14:08
@sonney2kblackburn, post shogun 2.0 I would say :D14:08
@sonney2kso blackburn note it down as wishlist in the bugtracker14:08
heikosonney2k, 16 is fine more me14:08
wikingfine by me14:08
uricamicok for me14:08
vojtechok for me14:09
blackburnfine14:09
n4nd0ok14:09
chenggood for me.14:09
heikoblackburn, I would say this is great, but has to be separate for differnt kinds of model selection14:09
@sonney2kpuffin444, alexlovesdata ?14:09
heikobut grid-search: for sure!14:09
pluskidnot sure whether I can be there14:09
blackburnheiko: yes, for grid (for now)14:09
pluskidI will be a student volunteer for KDD, which runs through Aug. 12~1614:09
puffin444There's a chance that I will be moving to Pittsburgh on that date14:09
pluskidIf it finishes early, then I guess I can attend14:09
pluskidsince it is about 19:00 in local time14:10
heikoblackburn, actually thats not too hard using threads (only  non-locked machines)14:10
heikoblackburn, ah, subsets14:10
heikonot thread safe14:10
blackburnheiko: not parallel14:10
blackburndistributed!14:10
heikobut one could start seperate processes14:10
@sonney2kblackburn, heiko - later!14:10
heikoI see14:10
ckwidmer16th ok with me14:10
@sonney2kpluskid, puffin444 well then your mentors can hopefully join and we do some mini celebration with only you later14:11
@sonney2kalexlovesdata, ^?14:11
alexlovesdataye14:11
@sonney2kok14:12
@sonney2kalright then - thanks everyone for attending again14:12
@sonney2kcontinue the nice work14:12
@sonney2kand send me *pictures*14:12
@sonney2knice ones for the new website14:12
@sonney2kn4nd0, and send bern4rd kudos for his work!14:12
n4nd0sonney2k: sure14:12
@sonney2k(on the new website)14:12
-!- cronor [~cronor@141.23.76.80] has quit [Remote host closed the connection]14:12
@sonney2kcase dismissed!14:12
wikingshall we start a short discussion on unit testing now for max 10 mins?14:13
@sonney2kwiking, go14:13
-!- cronor [~cronor@fb.ml.tu-berlin.de] has joined #shogun14:13
wikingok so i've said everyting what i think about how we should include unit testing into shogun in the discussion of the PR14:13
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed]14:13
heikogive me 5 mins14:13
heikophone ringing14:13
wikingbut basically i'm not supporting the idea of using submodules14:13
wikingi rather go with a separate branch and when we have enough code14:14
wikingwe just merge14:14
wikingthis way i can 'commit' all the stuff we'll need for unit testing14:14
wikingand then we'll just have one big PR :P14:14
wikingbut then again i guess14:14
wikingthere's no discussion anymore14:14
wikingabout whether we shoudl use google's c++ unit testing and mocking framework for this, or?14:15
wikingbut anyhow we go14:15
wikingi think what would be the best that if possible everybody who's now working on a gsoc project14:15
wikingat least add unit testing to their own part14:15
@sonney2kwiking, I am thinking of splitting up shogun even a bit more into shogun / interfaces / data / tutorial / examples&tests14:15
wikingsince you know the best what should happen with your code14:16
wikingcreating unit tests and mocking tests are rather simple14:16
@sonney2kwiking, the biggest advantage of having a submodule would be that you can directly work on that stuff and merge at your own will14:16
wikingand i've added a very simple example into the PR14:16
wikingsonney2k: yep i know :)14:16
wikingsonney2k: that's very tempting14:16
wikingsonney2k:  but my concern is more about how a PR would go14:16
wikinglet's say after we get unit testing into a good state14:17
wikingand we could require that each new module/class14:17
wikingmust have unit testing added as well14:17
wikingthen it'll be very complicated14:17
wikingto do a simple PR14:17
heikore14:17
wikingsonney2k: because it'd be great to have one of those pre-commit hooks working in shogun repo for developers... of course we cannot enforce developers to use that, but still we could ask them nicely in a README.developer14:18
wikingor something14:18
@sonney2kwiking, I cannot follow... what do you want to discuss?14:18
@sonney2kwiking, yes that makes sense14:18
@sonney2kbut that is just extra candy14:18
wikingsonney2k: the submodule vs directly into shogun repo14:19
blackburnI do not think hooks are so useful14:19
n4nd0all right, time to get on board14:19
n4nd0see you guys, thank you for the meeting!14:19
blackburnn4nd0: have a good flight14:19
n4nd0blackburn: thank you :)14:19
@sonney2kn4nd0, have a nice flight14:19
n4nd0thanks14:19
-!- n4nd0 [~nando@192.36.80.8] has quit [Quit: leaving]14:19
wikingsonney2k: and that to require a little bit of unit tests from all the gsoc developers14:19
wikingif possible14:19
heikon4nd0 have a good flight!14:19
wikingn4nd0 safe tripe!14:19
wiking*trip14:19
wikingsonney2k: so for example could we agree to ask all the gsoc-ers in shogun to create unit test, except gsomix? :)14:20
wikingsonney2k: nothing so fancy... just something really basic14:20
@sonney2kwiking, I think we are a little late for that now (requiring contributions from everyone - maybe one example or so from each gsocer)14:20
heikowiking, I would agree on that, at least some14:20
@sonney2k... is the only thing we can possibly manage14:21
wikingsonney2k: i mean really i dont want any full blown mocking stuff14:21
wikingjust a very basic unit testing14:21
wikingfor at least one class then :)14:21
@sonney2kwiking, for some part of the code - ok14:21
wikingsonney2k: yeah that'd be more than enough i think14:21
wikingsonney2k: just to keep it a little flowing14:21
wikingso maybe somebody could pick up the thread later...14:21
@sonney2kyeah14:22
@sonney2kok so back to submodule / not submodule14:22
wikingok14:22
wikingi'm really14:22
@sonney2kwhat is the advantage of having it not as submodule?14:22
wikingup for getting examples out to a submodule14:22
wiking:))14:22
@sonney2k1 commit, 1 PR right?14:22
wikingsonney2k: yes14:22
@sonney2kanything else?14:22
wikingsonney2k: this way you can really see that unit testing is at least handled somehow...14:23
wikingsonney2k developers must use it14:23
wikingsonney2k: as git submodule --init is optional ;)14:23
@sonney2kanyone else have an opinion on that?14:24
wikingsonney2k: i know that currently almost everybody is using the git HEAD of the master when using shoung14:24
@sonney2kwiking, most people just need the source14:24
wikingbut idealy only developers would do that14:24
heikoI like this approach, this way, only developers see it14:24
@sonney2kthey don't care about tests :)14:24
wikingsonney2k: yes... but what i'm saying is that actually most people wouldn't need the source at all :)))14:24
wikingsonney2k: just the binary release :P14:25
@sonney2kwiking, we are still not there yet...14:25
wikingsonney2k: i know of course...14:25
wikingi'm just saying that as an end goal14:25
wikingthat would be the ideal14:25
@sonney2kblackburn, pluskid, uricamic opinions?14:25
wikingyeah but let's just do a quick vote on this, or? :)14:26
wikingwith sonney2k having a veto on it ;)14:26
@sonney2kwiking, I don't have strong feelings on that14:26
blackburnsorry I am ab it lost14:26
pluskidno preference here :-/14:26
wikingblackburn: submodule or no submodule for unit testing14:26
wikingblackburn: or dontcare ;)14:27
@sonney2kjust thought that it is more clean and convenient to have it all in one tests/ dir and let you take control over it14:27
blackburnwiking: so what is the question? :)14:27
wikingblackburn: :DDDDDDDDDDDDDDDDDDDDDDDDDD14:27
wikingblackburn: VODKA?: )14:27
blackburnwrong14:27
* wiking failed big time :)14:27
@sonney2kwiking, vodka is the answer!14:27
@sonney2knot the question14:27
blackburnexactly14:28
wikingblackburn: more vodka?14:28
wikingok14:28
blackburnit is the answer again14:28
wikingok anyhow :>> as i'm failing here14:28
wikingsubmodule then?14:28
wikingi mean it'll be a little bit of a hassle14:28
@sonney2kblackburn, heiko, pluskid - submodule y/n/<don't care>14:28
blackburnn14:29
blackburnhmm14:29
alexlovesdataI do not understand the problem really, I hope none is angry about it14:29
blackburnn14:29
pluskiddon't care14:29
wikingheiko: ?14:29
blackburnalexlovesdata: I am crazy angry about that14:29
blackburn:D14:29
wikingblackburn: :D14:29
* wiking just migrating to llvm 4.0... yey :)14:29
blackburnwiking: will you be available 2 hrs later?14:29
@sonney2kalexlovesdata, russians always are (when they lack the right level of vodka in blood)14:29
wikingblackburn: i hoped not to be14:29
blackburnargh14:29
-!- ckwidmer [45c9b18c@gateway/web/freenode/ip.69.201.177.140] has quit [Quit: Page closed]14:29
wikingblackburn: maybe 5-6 hours later?14:29
blackburn5-6 would work too14:30
wikingheiko: pinguuu14:30
blackburnthe more the better14:30
wikingblackburn: that's ok by me i guess ;P14:30
blackburnwiking: not sure if alexlovesdata would accept it though14:30
blackburnso just let me know what you've decide14:30
heikowiking, I also dont really have a preference as both approaches have their downsides14:30
alexlovesdataaccept what?14:30
wikingok14:30
blackburnalexlovesdata: 5-6 hrs later :)14:30
wikingsonney2k: your call :)14:30
alexlovesdatayou mean to talk in 5-6 hours?14:30
blackburnyeah14:30
wikingalexlovesdata: yes14:30
@sonney2kwiking, I prefer submodule14:31
wikingsonney2k: ok then we go with that14:31
alexlovesdatathen lets chat at 9 PM ?14:31
@sonney2kwiking and you in code?14:31
blackburnwhat's timezone?14:31
@sonney2kno idea what the vote is now14:31
wikingsonney2k: well i'm for n14:31
alexlovesdatamiddle european ... too late for you (+1 I think) ...14:31
wikingsonney2k: so 2 non-submodule, 3 dontcare14:31
alexlovesdataot are you living in Tshukshcia?14:31
blackburnnot too late I live in UTC-514:31
blackburnvirtually14:31
blackburn:D14:31
@sonney2kwiking, so then no submodule14:31
wikingsonney2k: but u have the veto :))))14:31
blackburnalexlovesdata: tshukshcia :D14:32
wikingsonney2k: u shall use it wisely ;P14:32
@sonney2kas I said I don't have strong feelings about it14:32
wikingsonney2k: ok... then no submoodule14:32
wikingsonney2k: then can we merge?14:32
heikowiking, how to proceed now?14:32
@sonney2kand one can at any time convert a dir into a submodule14:32
alexlovesdatayeee. wiking got it ;)14:32
blackburnalexlovesdata: it is a name of people so you should rather ask 'are you tshukcia?'14:32
blackburn:D14:32
wikingsonney2k: or is there something you want it to be change still in that PR?14:32
@sonney2kargh puffin left already14:33
alexlovesdataI did not know how the landscape would be named ;)14:33
wikingheiko: i think now it's time to ask nicely gsocers to try to add some simple unit testing for their code, after the PR has been merged14:33
blackburnalexlovesdata: chukotka14:33
alexlovesdataok!!14:33
heikowiking, Ill merge it and fly over it, see how its works etc14:33
wikingheiko: and for you and me to crazy produce unit testing :)))14:33
wikingheiko: GREAT!14:33
alexlovesdatawiking, blackburn ... so at what time we like to talk about VOC?14:34
wikingok so i think then unit testing is settled14:34
heikoI will write a mail later this week, nicely asking people to add tests for the most iportant methods, and how to design these etc14:34
alexlovesdata9 PM local time ok for wiking?14:34
wikingheiko: sounds like a plan!14:34
wikingalexlovesdata: for sure not14:34
wikingalexlovesdata: i've just realised14:34
heikook Ill merge ..14:34
wikingalexlovesdata: i have family errands to attend to :(14:34
blackburnalexlovesdata: I live in place where lada crap is produced http://autoteile-blog.daparto.de/wp-content/themes/Lada-Niva.jpg :D14:34
CIA-18shogun: Heiko Strathmann master * r5847502 / (272 files in 36 dirs): Merge pull request #667 from vigsterkr/utest (+5 more commits...) - https://github.com/shogun-toolbox/shogun/commit/5847502840dc0b3c67c55360d20612e8460ba71614:34
alexlovesdataI will also work this weekend14:34
wikingcan we have a little chate14:34
wikingheiko: WOOOHOOOO!!!14:34
wikingalexlovesdata: me too14:35
wikingblackburn: weekend?14:35
heikowiking, that was a long one :)14:35
blackburnalexlovesdata: wiking: the time is ok for me14:35
blackburnanytime14:35
blackburn:D14:35
wikingso anybody interested in pascal VOC this year?14:35
wikingblackburn: ok cool14:35
heikoI might be, but not before september14:35
alexlovesdatalada niva is produced in UTC -5 ???14:35
heikoAnd I gotta go now, buy guys!14:35
blackburnalexlovesdata: you skipped 'virtually' word14:35
wikingheiko: cya! and thanks14:35
wikingok so anyhow14:35
blackburnalexlovesdata: I live in UTC+4 but virtually live like in UTC-5 :D14:35
wikingjust to keep alexlovesdata updated14:35
wikingand then i'll have to leave as well14:36
alexlovesdataso half-way chukotka14:36
-!- heiko [~heiko@host86-181-156-213.range86-181.btcentralplus.com] has left #shogun []14:36
wikingalexlovesdata: we are currently trying to participate in this actually14:36
alexlovesdatain what of the pascal VOC categories?14:36
blackburnexactly14:36
wikingalexlovesdata: http://www.image-net.org/challenges/LSVRC/2012/14:36
blackburnsome dogs IIRC14:36
alexlovesdataso the large scale stuff14:36
wikingalexlovesdata: mainly task1 and task214:37
wikingalexlovesdata: but of course VOC woudl be great as well14:37
wikingi have a cluster of 9 nodes14:37
wikingrunning hadoop14:37
wikingand have 7tb space on a hdfs14:37
alexlovesdatahow many CPUS?14:37
alexlovesdatanode = ?14:37
wikingalexlovesdata: 9*414:37
wikingor sorry!!14:37
alexlovesdatagood14:37
wikingonly 8 nodes14:37
blackburnguys just discuss it now if you have time now and let me know what can I help with14:37
wikingso 8*4 cpu14:38
alexlovesdatafor imageclef2011 I had only 32 32 bit cpus for feature comp and that was enough14:38
wikingalexlovesdata: we thought about starting with this paper's ideas:14:38
wikingjust a sec14:38
wikingi have to find it14:38
wikingLarge-scale Image Classification:14:39
wikingFast Feature Extraction and SVM Training14:39
wikinghttps://www.linyq.com/~linyqcom/0694.pdf14:39
alexlovesdatahow many train and test data precisely ?14:39
-!- vikram360 [~vikram360@117.192.189.69] has quit [Ping timeout: 244 seconds]14:39
wikingalexlovesdata: There are a total of 1281167 images for training. The number of images for each synset ranges from 732 to 1300.14:40
wikingThere are 50,000 validation images, with 50 images per synset.14:40
alexlovesdatayou know the CVPR 2012 paper by vedaldi and zisserman on how to gett tht coding right?14:40
wikingalexlovesdata: exactluy14:40
alexlovesdatathe problem in this coding is to get the parameters ... and it is very slow14:40
alexlovesdatayou will not have time to recompute the coding 10 times14:41
wikinganyhow14:41
wikingthe thing is14:41
wikingthat i have to go now :)))14:41
wikingso i just wanted to give an update what we thought about doing14:41
alexlovesdataso when will we talk about it? Saturday afternoon?14:41
wikingand to spread the word14:41
wikingthat we are trying to do something here with blackburn14:41
wiking:)))14:41
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has quit [Quit: Leaving.]14:41
wikingand that we need ideas and manpower :)14:42
wikingbut yes14:42
shogun-buildbot_build #144 of bsd1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/14414:42
wikingsaturday would work for me in the afternoon14:42
wikingso let's try that14:42
wikingalexlovesdata: ok?14:42
wikingalexlovesdata: and send me hog please when u can :)))14:42
alexlovesdataok, then I will send around a mail to you ... when I have finished the first draft of my diss until Sat afternoon/evening14:42
alexlovesdatayes, you will get hte hog14:42
alexlovesdataI have ideas how to get a reasonable result on that14:42
wikingokeeey14:43
blackburnthat's pretty cool14:43
wikingbut now i really gotta go14:43
alexlovesdatapls go14:43
wikingfamily errands started 30 mins ago already :)))14:43
alexlovesdatalets talk another time14:43
wikingso tomorrow14:43
alexlovesdatatomorrow14:43
blackburnalexlovesdata: did you try FREAK  already?14:43
wikingand thnx14:43
alexlovesdatawhat is FREAK?14:43
blackburnalexlovesdata: some cool descriptor14:44
@sonney2kblackburn, is that from opencv?14:44
blackburnsonney2k: yes already in14:44
blackburnhttp://infoscience.epfl.ch/record/175537/files/2069.pdf14:44
blackburnhere is the paper14:44
blackburnsonney2k: there is a standalone code at github too as well14:45
shogun-buildbot_build #145 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/145  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Viktor Gal <viktor.gal@maeth.com>14:45
CIA-18shogun: Soeren Sonnenburg master * rfe8dddc / examples/undocumented/libshogun/regression_gaussian_process_fitc.cpp : ifdev have_eigen3 regression_gaussian_process_fitc.cpp example - https://github.com/shogun-toolbox/shogun/commit/fe8dddc971f98eb0bf8200ef9de50f1276262b3c14:49
@sonney2kblackburn, can we directly use opencv for such stuff?14:55
blackburnsonney2k: what do you mean?14:55
blackburnuse for what exactly?14:55
CIA-18shogun: Soeren Sonnenburg master * rd02abd4 / (2 files): ifdef examples requiring NLOPT with HAVE_NLOPT - https://github.com/shogun-toolbox/shogun/commit/d02abd49cc050596dd9e1bcc09e14236d798427514:59
@sonney2kfor any kind of coffin based features if required15:00
blackburnsonney2k: well in case of only one keypoint I think wrapper can be easily done15:00
blackburnI don't really know what is better to wrap15:01
blackburnshogun machine to opencv or opencv stuff to shogun's features15:02
blackburnokay will be back in ~3 hrs15:02
shogun-buildbot_build #248 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/24815:17
shogun-buildbot_build #206 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/206  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Viktor Gal <viktor.gal@maeth.com>15:19
-!- uricamic [~uricamic@2001:718:2:1634:39be:d411:577:692b] has quit [Quit: Leaving.]15:27
@sonney2kblackburn, license wise we can use / link to opencv's stuff but not the other way round15:36
shogun-buildbot_build #249 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/249  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Viktor Gal <viktor.gal@maeth.com>15:43
shogun-buildbot_build #146 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/146  blamelist: Soeren Sonnenburg <sonne@debian.org>15:48
-!- vikram360 [~vikram360@117.216.149.224] has joined #shogun15:58
shogun-buildbot_build #250 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/250  blamelist: Soeren Sonnenburg <sonne@debian.org>16:07
shogun-buildbot_build #207 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/207  blamelist: Soeren Sonnenburg <sonne@debian.org>16:09
-!- vikram360 [~vikram360@117.216.149.224] has quit [Ping timeout: 244 seconds]16:12
-!- vikram360 [~vikram360@117.192.174.5] has joined #shogun16:12
-!- vikram360 [~vikram360@117.192.174.5] has quit [Ping timeout: 255 seconds]16:19
-!- vikram360 [~vikram360@117.192.166.45] has joined #shogun16:19
-!- tiger_eye [~tiger_eye@rrcs-24-106-116-254.central.biz.rr.com] has joined #shogun16:22
-!- vikram360 [~vikram360@117.192.166.45] has quit [Ping timeout: 240 seconds]16:32
CIA-18shogun: Soeren Sonnenburg master * r61ff3da / (2 files): fix include path: base/config.h -> lib/config.h - https://github.com/shogun-toolbox/shogun/commit/61ff3daa6366c65fa45426f19486789764e6bde816:44
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Quit: Page closed]16:49
shogun-buildbot_build #147 of bsd1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/14716:50
shogun-buildbot_build #251 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/251  blamelist: Soeren Sonnenburg <sonne@debian.org>17:10
shogun-buildbot_build #208 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/208  blamelist: Soeren Sonnenburg <sonne@debian.org>17:12
-!- cronor [~cronor@fb.ml.tu-berlin.de] has quit [Quit: cronor]17:13
-!- pluskid [~pluskid@108.171.196.83] has quit [Quit: Leaving]18:21
blackburnalexlovesdata: thanks for sharing some thoughts19:05
alexlovesdata:)19:05
alexlovesdatafunny I am about to go now19:05
alexlovesdatainstead of PCA we could use sift with 3x319:05
alexlovesdatais probably more sane for reducing dimension load eg for fisher vectors if you like to go for them19:06
-!- alexlovesdata [~binder@goldenezahl.ml.tu-berlin.de] has left #shogun []19:06
@sonney2kwiking, seen the buildbot breakage?20:10
@sonney2kdue to test move stuff20:10
-!- sonney2k_ [~shogun@7nn.de] has joined #shogun20:18
-!- zxtx [~zv@173-228-88-191.dsl.dynamic.sonic.net] has joined #shogun20:18
-!- sonney2k [~shogun@7nn.de] has quit [Ping timeout: 265 seconds]20:19
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun20:21
-!- tiger_eye [~tiger_eye@rrcs-24-106-116-254.central.biz.rr.com] has left #shogun ["Leaving"]20:27
-!- Netsplit *.net <-> *.split quits: sonney2k_20:36
-!- Netsplit *.net <-> *.split quits: shogun-buildbot_, octopine, blackburn20:36
-!- Netsplit *.net <-> *.split quits: zxtx20:36
-!- Netsplit *.net <-> *.split quits: naywhayare20:36
-!- Netsplit over, joins: zxtx, blackburn, shogun-buildbot_, naywhayare, octopine20:37
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds]20:40
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun21:02
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds]21:39
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun21:42
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun22:04
KMcQuistenHello again!22:04
blackburnhey22:04
blackburnwhat is problem today? ;)22:05
KMcQuistenThanks for the help yesterday.  I think I have another small problem.  I think something is getting freed twice when I use DirectorKernels in MKL.  The process runs fine and I get my results, but the script segfaults upon completion.  Once again I have a script to send you to reproduce.22:05
blackburnsure send me22:06
blackburnthank you for improving user experience :)22:06
KMcQuistenNo problem.  I use this package every day, so i'm glad to help make it better22:09
blackburnKMcQuisten: what is you are applying it for btw?22:09
KMcQuistenModeling sequence and target data to predict the efficacy of RNAi gene knockdown constructs.22:10
blackburnI think we should track it down somehow22:11
blackburnso you are using string features widely, right?22:11
KMcQuistenYes.  I use StringCharFeatures, StringWordFeatures to implement Spectrum Kernel, and RealFeatures to model thermodynamic properties of nucleotide sequences22:13
blackburnsounds cool22:13
KMcQuistenReproduction script sent22:13
blackburnI am asking because I am not really into that stuff22:14
KMcQuistenIt's quite fun.  We're going to use it to predict RNAi sequences that will effectively knockdown genes critical to cancer growth22:14
blackburnI am collaborating with c. widmer on some cancer as well22:14
KMcQuistenLet the computer figure it out instead of spending millions in a lab22:14
blackburnbut for now I only have worked on microarray data22:15
KMcQuistenRight on22:15
blackburnokay thanks for the script let me check now22:15
KMcQuistenThat's great for the large scale ON/OFF switching to see what genes are on in tumors, then we do specific screens against those genes to build datasets for modelling knockdown of those genes.22:16
blackburnI feel stupid now :D22:16
KMcQuistenHa!  Don't feel bad.  I knew nothing about molecular genetics when I started work on this project.  Just a lowly mathematician among the bio nerds :)22:17
blackburnI wish I could take some course on that but that country is just totally outdated22:18
KMcQuisten?22:19
blackburnI mean nobody in my university and probably even city knows anything about molecular biology :)22:20
-!- zxtx [~zv@173-228-88-191.dsl.dynamic.sonic.net] has quit [Ping timeout: 248 seconds]22:20
KMcQuistenAh, I see22:20
blackburnand no optional courses in educational system at all22:21
blackburnKMcQuisten: where editdist comes from?22:25
KMcQuistenGah!  It's another package.  I forgot that it was in there22:25
KMcQuistenLet me edit the definition22:25
KMcQuistenand resend22:25
blackburnKMcQuisten: what can I use instead?22:26
KMcQuistensum([1 for x,y in zip(s1,s2) if x!=y])22:27
KMcQuistenthat's hamming distance, a bit simpler22:27
KMcQuistenbut it should work without needing a reference to the editdist package22:28
blackburnI used sequence matcher but it fails to converge22:28
blackburnlet me try hamming-like22:28
puffin444blackburn, is the build still broken?22:29
blackburnokay I reproduced it22:29
blackburnI don't know :D22:29
KMcQuistenNope, I rebuilt about 20 min ago and it's fine22:29
blackburnpuffin444: well check that http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/251/steps/test%20python_modular/logs/stdio22:29
blackburngaussian_process_modular seems to die or so22:30
KMcQuistenI always do a fresh pull and rebuild to make sure my bugs are still bugs before I come in here for help :)22:30
blackburnKMcQuisten: btw you can use apply now22:30
blackburnwith libsvr22:30
KMcQuistenYes I noticed :)  got it all sqaured away in my code22:30
KMcQuistenthanks blackburn22:30
blackburnnot the best way we did22:31
blackburnso one need to track properly these renames22:31
KMcQuistenYeah, making swig magic happen is easily broken22:31
puffin444uh oh, the buildbot doesn't have pylab?22:32
blackburnpuffin444: is it a graphicla examplE?22:32
blackburnput it into the graphical then ;)22:32
puffin444yes that is the one22:32
blackburnKMcQuisten: interesting bug22:33
blackburnKMcQuisten: did director kernel work for you before?22:33
KMcQuistenYes, we did.  I think some of the recent changes with ref counting may have broken this.22:34
KMcQuistenIt had worked, and it's still working, save for not properly deallocating at the end.22:35
KMcQuistenThe machine trains fine, I can extract alphas and support vectors, and make predictions on unseen functions, it just crashes when the object is destroyed.22:35
blackburnyeah I see22:36
blackburnactually reference counting works bad here22:37
KMcQuistenoh?22:39
blackburnKMcQuisten: each get_lhs and get_rhs increases reference counter but it doesn't get down hmmmm22:40
KMcQuistenodd22:42
blackburnKMcQuisten: in your example it goes to 124747 :D22:43
KMcQuistenGah!22:43
KMcQuistenThat's no good22:43
KMcQuistenOk, I have to run to a research group meeting.  Feel free to email me if you discover anything22:48
blackburnsure22:49
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Quit: Page closed]22:49
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed]23:11
CIA-18shogun: puffin444 master * rafc77dc / (2 files in 2 dirs): Put gaussian process regression demo in proper folder - https://github.com/shogun-toolbox/shogun/commit/afc77dc3c6c54ca8f210b6546117c89675040a6423:27
CIA-18shogun: Sergey Lisitsyn master * r13785f4 / (2 files in 2 dirs): Merge pull request #689 from puffin444/master - https://github.com/shogun-toolbox/shogun/commit/13785f408b3d1d4cfe3e37a5f269703617d96aab23:27
shogun-buildbot_build #252 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/252  blamelist: puffin444 <walke434@msu.edu>23:56
shogun-buildbot_build #209 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/209  blamelist: puffin444 <walke434@msu.edu>, Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:58
--- Log closed Sat Aug 04 00:00:17 2012