Open in new window / Try shogun cloud
--- Log opened Tue Aug 30 00:00:37 2011
-!- blackburn1 [~blackburn@] has quit [Ping timeout: 240 seconds]00:39
-!- in3xes [~in3xes@] has joined #shogun03:52
-!- in3xes [~in3xes@] has quit [Quit: Leaving]05:05
@sonney2kserialhex, - you patch is not mergable06:16
@sonney2kyou have to rebase it06:16
-!- blackburn [~blackburn@] has joined #shogun10:12
-!- heiko [~heiko@] has joined #shogun10:20
@sonney2kmorning heiko - glad to see that your gf let you live and code ;-)10:21
heikomorning :)10:21
heikoyes, me too ;)10:21
heikook then, now for some hardcore coding action :)10:22
-!- blackburn [~blackburn@] has quit [Ping timeout: 260 seconds]10:24
-!- blackburn [~blackburn@] has joined #shogun10:25
@sonney2kheiko, just ping me when there is anything10:51
heikosonney2k, ok i will10:51
heikonow it the migrate method has to return by reference whether a new TParameter instance is returned, and also if its was was newly allocated10:52
heikoso all cases may be handleed10:52
@sonney2kseems like you really thought about it all then10:56
@sonney2kI would have just made the strict assumption - no messing with inputs -10:56
@sonney2kI think you will have to give me a short 'tutorial' once you are done11:00
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]11:02
-!- blackburn [~blackburn@] has joined #shogun13:40
-!- blackburn [~blackburn@] has quit [Ping timeout: 240 seconds]13:46
@sonney2kheiko, what's up in the jungle?14:11
heikosonney2k, its horrible14:11
heikoevery time i touch something, new error araise14:12
heikoI can now serialize/deserialize SGVectors14:12
heikobut the normal vectors do not work14:12
@sonney2kheiko, wait the normal serialization stuff no longer works?14:13
heikoThe most frequent problem is the problem to build a TParameter isntance that can call load, from sratch, meaning that class variables are not attached14:13
heikoyes it does, but migration does not14:13
heikonormal serialization works all the time14:13
heikobecause it is done separately14:13
heikobut migration of the old style vectors does not yet14:14
heikoits really really complicated, so many side-effects to think about14:14
@sonney2kso for nomal vectors you need extra treatment right?14:15
@sonney2klike copy doulbe* and len14:15
heikocurrently i am getting mad because i just dont get it how to create a TParameter instance to load an old style vector14:15
@sonney2kand same with matrix/string/...14:15
heikono even touched this yet14:15
heikoonly doing vector/matrix14:16
@sonney2kheiko, how about just *always* returning SGVector etc instead of normal vectors?14:16
@sonney2kI mean when you have to do migrations then SGVector exists and is a more clean structure to store things14:17
@sonney2kso then you don't need special extra treatment and the person doing migrations knows that he will only get scalars or SG* structures14:17
heikoI also thought of that some time ago but then I would have to modify things even in a deeper layer14:17
heikowhich then causes even more horrible problems14:17
heikoI am getting invalid writes/reads all the time14:18
@sonney2khmmhhh, what problems do you have in more deep layers?14:18
heikoi would have to modify Tparameter::load for that (to store old style vectors in SGVector)14:19
serialhexsonney2k: ok, i'll doo that and re-post it14:19
@sonney2kserialhex, thanks14:41
@sonney2kheiko, sorry had to answer some questions here14:41
@sonney2kheiko, that sounds a lot simpler to me14:42
@sonney2k^ modifyingt Tparameter::load to optionally store old-style stuff in new-style vectors14:42
heikosonney2k, i got to go now, sorry14:58
@sonney2kheiko, so what shall we do?14:58
heikoi dont know14:59
heikothis is simply not easy14:59
heikoI dont know if it will be finished quick, I thought I could do it today14:59
heikobut the current problem blocks it14:59
heikoI pushed all code to my github account15:00
heikoperhaps you can take a look15:00
heikohere is my test program:15:00
@sonney2kok then I guess we will have to make a release w/o this - I don't know whether you will have time to work on this later on such that we can do a proper conversion15:00
heikoand here the input file15:00
@sonney2kany thoughts?15:00
heikoI will have tome to finish it, but the problem is that i move out of my flat on friday as I already told oyu15:01
heikoI will have some time tomorrow in the morning but I really need to start on the flat15:01
@sonney2kheiko, I know - I am just asking if at all15:01
-!- f-x [75c0d906@gateway/web/freenode/ip.] has joined #shogun15:01
@sonney2kI mean I don't want to be an annoyance pushing you to do work you don't enjoy that much for too much longer15:01
@sonney2kso I meant more like in the next weeks/months then15:02
heikoyes, I know15:02
@sonney2kI certainly prefer to keep you in the time than to push you beyond what is fun... - so it is really up to you15:03
@sonney2kheiko, but if you commit doing it - it would certainly be great - but I don't know 1-2 months from now on ETA I guess then?15:03
heikoyes something like that15:04
heikonext two months are full15:04
heikobut again, I think it will work soon, the conecept is there15:04
heikothe problems are of technical nature15:04
heikolets do it like this:15:04
heikoi will try to finish it in the next days, i have some hours per day15:05
heikoand if it works, its great.15:05
@sonney2kthat sounds more like never then - the problem with this is that you really need to be so focussed - deep inside the code that it doing it as a side thing is close to impossible15:05
@sonney2kif not?15:05
heikoif not, either you have a look, or I will do it later15:05
heikomeaning starting again on it in one month15:05
@sonney2kthat is certainly fair15:06
@sonney2kso we will try to do the tests w/o it and just create new regression tests with the new release15:06
@sonney2kthen make a release and we will see how people react15:07
heikoperhaps wait one more day before you start on it15:07
@sonney2kwell we wanted to release tomorrow right :)15:07
heikook then ..mmh15:07
@sonney2kthen blackburn starts studying again15:07
@sonney2kI will have a new job ... so it is all difficult for us then for the next few weeks15:07
heikoits stupid: the migration basically works, scalars work, SGVectors do work, only the normal vectors do not15:08
@sonney2kwhich is the only thing we need to migrate15:08
heikoBut I will defenetly not finish it today15:08
@sonney2kyeah I mean I could not foresee that it takes so long and that you were away most of last week15:09
@sonney2kso it is like it is -15:10
heikoMe too, thought it would be done in a week or so, but now I am working on it for two weeks, and the serialization bugs also took a lot of time15:10
@sonney2kheiko, I am very happy that you intend to finish it and stay around later15:10
@sonney2kand you've done a great job anyways15:10
@sonney2kso we should be more in a party than fail mood :)15:11
heikothanks :) but a bit silly that this one does not work out15:11
heikotrue :)15:11
heikosonney2k, I got to go now, will be back tomorrow for some more talk and coding stuff15:11
@sonney2kwe will have another release w/ that feature soon15:12
@sonney2kso don't worry thanks and bye15:12
-!- heiko [~heiko@] has quit [Ping timeout: 258 seconds]15:16
-!- blackburn [~blackburn@] has joined #shogun15:25
blackburnsonney2k: any news?15:32
@sonney2kblackburn, ys16:17
@sonney2kblackburn, we have to do w/o migrations for the release16:17
blackburnis heiko out of schedule?16:20
@sonney2kblackburn, not out of schedule - it is much more difficult than we assumed16:21
@sonney2kso we should try to check things manually now16:21
@sonney2kthat is use an old shogun version somewhere  - run the python modular examples16:22
@sonney2kand then compare the output16:22
@sonney2kwith what we get when we run this on some new machine16:22
@sonney2kerr shogun16:22
blackburnhmm okay I see16:22
blackburnsonney2k: do you have time for that testing?16:26
@sonney2knot now16:27
@sonney2kcould you start a few things16:27
blackburnyeah sure, that's why I'm asking16:27
@sonney2kthanks so please start and document which examples you testes16:30
@sonney2kdid test16:30
@sonney2kbtw you could try to run tester.py16:30
@sonney2klets see how much works :)16:30
blackburnsonney2k: looks like nothing works :D16:42
blackburncompiling old shogun16:44
@sonney2kblackburn, *arg* the old shogun example also doesn't work16:58
@sonney2kerr the ocr one I meant16:58
CIA-3shogun: Soeren Sonnenburg master * rdc68e4f / (2 files): fix two more 'problematic' ruby examples -
CIA-3shogun: Soeren Sonnenburg master * r6f9f139 / src/shogun/io/BinaryFile.cpp : improve error output in binary file -
CIA-3shogun: Soeren Sonnenburg master * r4ecaa34 / (2 files): remove obsolete multiboost -
CIA-3shogun: Soeren Sonnenburg master * r312e97b / src/shogun/machine/KernelMachine.h :17:48
CIA-3shogun: Fix a bug in kernel machine introduced in SGVector transition17:48
CIA-3shogun: m_alpha has its own length and should be set and used independently from17:48
CIA-3shogun: m_svs otherwise serialization fails -
CIA-3shogun: Soeren Sonnenburg master * rf0b3121 / (applications/ocr/ applications/ocr/data/ocr.svm.gz): fix ocr example and data for new shogun -
@sonney2kblackburn, yay! progress17:49
CIA-3shogun: Soeren Sonnenburg master * r5a85eb5 / (2 files): fix sailfish application -
blackburnsonney2k: sorry had to go18:02
blackburnso are all applications operational now?18:03
CIA-3shogun: Soeren Sonnenburg master * rfaf1d9c / applications/vw/Makefile : add makefile for vw -
@sonney2kblackburn, looks like yes18:05
blackburnsonney2k: oh we got nice email with suggestion to implement some thing18:05
@sonney2kwhat's missing now is check tests and make release script / debian package18:05
@sonney2kblackburn, I would suggest you check the examples and I do the release scripts18:05
@sonney2kblackburn, when ?18:06
@sonney2kfrom whom18:06
blackburnsonney2k: take a look on the mailing list18:06
@sonney2kahh ok18:06
blackburnsonney2k: will you answer?18:07
@sonney2knice - please reply that this is warmly welcome18:07
@sonney2kcan you do that?18:07
@sonney2kjust now18:07
blackburnsonney2k: yeah okay18:07
-!- in3xes [~in3xes@] has joined #shogun18:27
shogun-buildbotbuild #171 of octave_modular is complete: Success [build successful]  Build details are at
blackburnI think there is no reason to check all distances/kernels18:57
blackburnwill check some of them now18:57
@sonney2kblackburn, I think you can even script this or?19:02
blackburnsonney2k: need your help to automate this :)19:02
blackburnsonney2k: surprise! gaussian kernel changed result19:07
blackburndistance too!19:09
blackburneven euclidian distance changed?!19:12
blackburnI guess I know19:15
blackburnsonney2k: our saved results are totally useless: we used wrong data reading19:17
-!- in3xes [~in3xes@] has quit [Ping timeout: 240 seconds]19:19
blackburnokay distances and kernels should be ok19:42
blackburnwe haven't changed any compute routines so I'm pretty sure all OK19:42
blackburnI had some doubts about parallel things but it is working smoothly19:43
@sonney2kblackburn, btw let me check if I regenerate the tests if they work or not19:57
@sonney2kkernel machines were broken before my last commit19:57
@sonney2kblackburn, please try the predict example in applications/ocr19:57
@sonney2kit's cool :)19:57
blackburnsonney2k: I tried it days before19:59
@sonney2kyou mean years?19:59
blackburnsth like 0.3 year before hah19:59
@sonney2knew crashers20:02
@sonney2ktry python in python_modular testsuite20:03
blackburn==22401==  Access not within mapped region at address 0x020:08
blackburn==22401==    at 0x9C14FF1: shogun::TParameter::save_ptype(shogun::CSerializableFile*, void const*, char const*) (Parameter.cpp:1987)20:08
blackburn==22401==    by 0x9C1545F: shogun::TParameter::save_stype(shogun::CSerializableFile*, void const*, char const*) (Parameter.cpp:2054)20:08
blackburn==22401==    by 0x9C15E1C: shogun::TParameter::save(shogun::CSerializableFile*, char const*) (Parameter.cpp:2193)20:08
@sonney2kyeah and other issues...20:09
@sonney2kyeah isn't it. it will never be boring20:11
@sonney2kblackburn, there are other valgrind errors before even20:17
@sonney2kwell actually it starts in the first file20:17
CIA-3shogun: Soeren Sonnenburg master * rc52988d / (20 files in 3 dirs):20:28
CIA-3shogun: Merge pull request #320 from serialhex/master20:28
CIA-3shogun: Many more ruby examples (+6 more commits...) -
@sonney2kI am not really making progress on this bug20:46
CIA-3shogun: Soeren Sonnenburg master * r03ee3f8 / (doc/pages/ModularTutorial.mainpage src/README.developer): mention apply() instead of classify() in tutorial and README -
@sonney2kblackburn, are you making progress?21:02
shogun-buildbotbuild #157 of ruby_modular is complete: Failure [failed test_1]  Build details are at  blamelist:, serialhex@gmail.com21:10
blackburnsonney2k: sorry got disturbed there in dormitory21:25
blackburnsonney2k: what's with ruby?21:27
shogun-buildbotbuild #160 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist:, serialhex@gmail.com21:28
blackburnsonney2k: ruby examples in python?!21:29
blackburnserialhex: hey you probably shouldn't have to commit python examples modifications ;)21:40
CIA-3shogun: Sergey Lisitsyn master * r9580eb8 / (3 files): Restore wrong modified python modular examples -
-!- f-x [75c0d906@gateway/web/freenode/ip.] has quit [Ping timeout: 252 seconds]22:28
blackburnsonney2k: got fixes for these crashes22:31
shogun-buildbotbuild #2 of nightly_default is complete: Failure [failed test_1]  Build details are at
shogun-buildbotbuild #162 of python_modular is complete: Success [build successful]  Build details are at
CIA-3shogun: Sergey Lisitsyn master * r21b24b3 / src/shogun/machine/DistanceMachine.cpp : Fixed segmentation fault error emerged with DistanceMachine serialization -
shogun-buildbotbuild #163 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist: sonne@debian.org23:09
blackburnshogun-buildbot: what the fuck are you compiling?23:10
CIA-3shogun: Sergey Lisitsyn master * raf951c1 / (2 files): Added modelselection examples' description -
--- Log closed Wed Aug 31 00:00:06 2011