Open in new window / Try shogun cloud
--- Log opened Sat Jun 25 00:00:54 2016
-!- bazdmeg is now known as BZDMG01:47
-!- sonne|osx [] has joined #shogun05:13
-!- sonne|osx_ [] has joined #shogun05:18
-!- sonne|osx [] has quit [Ping timeout: 240 seconds]05:20
-!- sonne|osx_ is now known as sonne|osx05:20
-!- sonne|osx [] has quit [Quit: sonne|osx]05:28
-!- Saurabh7__ [Saurabh7@gateway/shell/panicbnc/x-wlacyikiyzqrkfcs] has left #shogun []07:02
-!- sonne|osx [] has joined #shogun07:03
-!- sonne|osx [] has quit [Quit: sonne|osx]07:14
-!- sanuj [~sanuj@] has joined #shogun07:41
sanujHeikoS_web, there?07:41
sanujwiking, there?07:42
-!- HeikoS [] has joined #shogun07:55
-!- mode/#shogun [+o HeikoS] by ChanServ07:55
sanujHeikoS, yo07:55
@HeikoSsanuj: jo, sorry gotta come back later07:56
-!- HeikoS [] has quit [Client Quit]07:56
-!- sanuj [~sanuj@] has quit [Ping timeout: 244 seconds]08:01
-!- sanuj [~sanuj@] has joined #shogun08:06
-!- Saurabh7__ [Saurabh7@gateway/shell/panicbnc/x-wlacyikiyzqrkfcs] has joined #shogun09:40
-!- GandalfTheWizard [~Eva@] has joined #shogun10:20
-!- HeikoS [] has joined #shogun10:23
-!- mode/#shogun [+o HeikoS] by ChanServ10:23
@HeikoSsanuj: jo10:24
@HeikoSnow I have some itme10:24
sanuji have a problem10:24
sanujwait i'll push the code10:25
@HeikoSok !10:25
@HeikoSsanuj: btw did you see my email for midterm update?10:27
@HeikoSsanuj: can you cc me in the daily update email?10:27
@HeikoSSaurabh7__: jo10:27
Saurabh7__HeikoS: yo10:29
sanujHeikoS, the error is caused in this region10:30
sanujHeikoS, here is the error
@HeikoSsanuj: so what shappening?10:31
sanujHeikoS, yeah, i'll start sending mails from monday....10:31
@HeikoSsanuj: cool thanks for that!10:32
sanujHeikoS, did you see the error?10:32
@HeikoSsanuj: I see it, but can you point me the the things important?10:33
sanuji think any destructor deletes member variables10:33
sanujany.h destructor10:33
@HeikoSwhats a minimal example?10:34
@HeikoSis there one in the PR?10:34
@HeikoSthis line deletes memory10:34
@HeikoSBZDMG: jo10:35
sanujBZDMG is away i think10:35
sanujHeikoS, why is the destructor being called10:36
@HeikoSsanuj: can you put the test that segfaults into its own main10:36
@HeikoSand then run again, share the output10:36
@HeikoSof valgrind10:36
@HeikoSthe current one is convoluted so much with gtest10:36
sanujokay, let me do that10:36
@HeikoSSaurabh7__: how are things?10:37
Saurabh7__HeikoS: I have kept the duplicate indices,10:37
-!- shogun-notifier- [] has joined #shogun10:37
shogun-notifier-shogun-data: Sanuj :master * 39551ae / testsuite/meta/regression/multiple_kernel_learning_regression.dat:
shogun-notifier-shogun-data: add data for mkl regression cookbook10:37
shogun-notifier-shogun-data: Heiko Strathmann :master * 5e64d2a / testsuite/meta/regression/multiple_kernel_learning_regression.dat:
shogun-notifier-shogun-data: Merge pull request #107 from sanuj/mkl_regression10:37
shogun-notifier-shogun-data: Mkl regression10:37
Saurabh7__I believe they are justified10:37
Saurabh7__I checked10:37
@HeikoSSaurabh7__: ok10:38
@HeikoScan you share someevidence for that with me? :)10:38
sanujHeikoS, what is HeikoS_web10:38
@HeikoSsanuj: thats my work computer logged in via webinterface10:38
@HeikoSbut it is in london10:38
@HeikoSand I am in France10:38
@HeikoSescaping the brexit10:38
Saurabh7__HeikoS: HeikoS for e.g
Saurabh7__so sampling with replacement10:39
Saurabh7__am i right?10:39
@HeikoSSaurabh7__: can you also check sklearn impl?10:40
Saurabh7__HeikoS: I checked10:41
Saurabh7__its very different in terms of approach, has been changed with time10:41
Saurabh7__dups are allowed10:41
Saurabh7__but they handle it with sample weights10:41
Saurabh7__like bincount(samples)10:42
Saurabh7__and use it somehow10:42
Saurabh7__so the idea is same tho10:42
@HeikoSso lets continue with things then10:43
Saurabh7__HeikoS: antoher thing, presortin I have kept optional still10:44
Saurabh7__because in sinlge tree we train on subset10:44
Saurabh7__so cant presort inside it10:44
Saurabh7__so I do it in RF10:44
@HeikoSthats ok then10:45
Saurabh7__HeikoS: ok I will add some doc in t current PR and some tests are needed10:46
sanujHeikoS, i'm not able to compile this10:52
sanujtemp.cpp:(.text+0x5): undefined reference to `shogun::init_shogun_with_defaults()'10:52
sanujerror ^10:52
@HeikoS :)10:52
@HeikoSso lets see10:53
@HeikoSyou know what an undefined reference means?10:53
@HeikoSI mean the error, what causes it=?10:53
sanujnot able to link it10:54
@HeikoSsanuj: so that happens10:54
@HeikoSif you can compile a program10:54
@HeikoSthat is, turn the program.c into program.o10:54
@HeikoSso all header files were available10:55
@HeikoSin the linking step, all the .o files are put together into the binary10:55
@HeikoSif there is a method that does not appear in one of the .o files linked together (that is, no implementation of the defined function)10:56
@HeikoSthen you get linker error10:56
@HeikoSif the method is in a third party lib, like your init_shogun_with_defaults10:56
@HeikoSthen, this means you linker cannot see any library that contains an implementation of the method10:56
@HeikoSsanuj: I suggest you read up on how c/c++ compiler work10:56
@HeikoSin practice, it means your forgot -lshogun10:56
sanujHeikoS, i compiled with => g++ --std=c++11 -I/home/sanuj/Projects/shogun/src/ -L/home/sanuj/Projects/shogun/buildpy/src/shogun -lshogun temp.cpp10:57
@HeikoSthis tells the linker where to look for .so files10:57
@HeikoSif you put the -lshogun but still get the error10:57
@HeikoSthis means that you LD_LIBRARY_PATH does not contain the file10:57
@HeikoSso either your -L option is wrong10:57
@HeikoSor there is no file there10:58
@HeikoSit is one of the most basic errors when compiling c++10:58
sanujmy LD_LIBRARY_PATH is /home/sanuj/Projects/shogun/buildpy/src/shogun:/home/sanuj/torch/install/lib:10:58
@HeikoSso no reason to ask more than once ;)10:58
sanujand the path in -L is also correct10:58
@HeikoSsanuj: nope10:58
@HeikoSbecause if it was, it would wortk10:58
@HeikoSdoes your -L path contain
sanujthat path has libshogun.so10:59
@HeikoSah yeah10:59
@HeikoSwell, I dont know, there are 100 mini things that can be set wrongly11:00
@HeikoSbut the rationale is usper simple11:00
@HeikoSlinker cannot find init_shogun_with_defaults11:00
@HeikoSthats it11:00
@HeikoSyou have to tell it where it is11:00
@HeikoSand it will work11:00
sanujHeikoS, this worked11:00
sanujg++ --std=c++11 -I/home/sanuj/Projects/shogun/src/ -L/home/sanuj/Projects/shogun/buildpy/src/shogun temp.cpp -lshogun11:00
sanujswapping temp.cpp and -lshogun11:00
@HeikoSsanuj: ah yeah that also is a thing to consider11:01
@HeikoSsanuj: try to read up on why11:01
@HeikoSso next time someone asks here, you can explain them11:01
sanujHeikoS, okay thanks11:01
sanujbtw i know c++ linker and compiler basics11:01
sanujeverything was correct but it was not working that's why i asked here11:02
@HeikoSsanuj: sure11:02
@HeikoSsanuj: I only explained11:02
@HeikoSsince the question11:03
@HeikoSto make sure you are on track with this11:03
@HeikoSsince there is only one answer to the question: the linker doesnt see libshogun.so11:03
@HeikoSbut all good11:04
@HeikoSthe mkl cookbooks look nice btw11:04
shogun-notifier-shogun: OXPHOS :develop * 9a8de01 / doc/ (3 files):
shogun-notifier-shogun: - update ecoc intro - remove python modular example11:04
shogun-notifier-shogun: Heiko Strathmann :develop * a39216f / doc/ (3 files):
shogun-notifier-shogun: Merge pull request #3296 from OXPHOS/cookbook_ecoc11:04
shogun-notifier-shogun: ECOC cookbook update11:04
sanujHeikoS, thanks11:05
shogun-buildbotbuild #738 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>, OXPHOS <>11:05
sanujHeikoS, why did you get out of UK because of brexit?11:05
@HeikoSsanuj: I was afraid that idiotism is infectious11:06
sanujHeikoS, haha, i meant is that the only reason you are in france?11:07
sanujHeikoS, updated error from main
@HeikoSsanuj: I also wanted to use the channel tunnel for the last time before it is blown up11:07
sanujthey will blow up the channel tunnel?11:08
shogun-notifier-shogun: OXPHOS :develop * a4e49b3 / / (4 files):
shogun-notifier-shogun: relaxed_tree cookbook11:08
shogun-notifier-shogun: Heiko Strathmann :develop * e001ab5 / / (4 files):
shogun-notifier-shogun: Merge pull request #3286 from OXPHOS/cookbook_relaxedtree11:08
shogun-notifier-shogun: Cookbook - relaxedtree classifier11:08
@HeikoSsanuj: :D11:08
@HeikoSjust in case11:08
@HeikoSchecking the error11:08
@HeikoScan you put this as a gist, and also put the example code11:08
shogun-buildbotbuild #739 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>, OXPHOS <>11:09
@HeikoSsanuj: gist allows you to have multiple files11:09
@HeikoSand no ads11:09
@HeikoSsanuj: ok thanks, thats much easier to look at11:11
sanujHeikoS, did you guys go to deepmind?11:12
@HeikoSnot yet11:12
@HeikoSprobably next week11:13
sanujokay :D11:13
@HeikoSis the code in your PR exactly the same as when you ran the example?11:13
sanuji read some of their research stuff11:13
@HeikoSyeah they do interesthing work11:13
@HeikoSand lots of really good people11:13
sanujHeikoS, code is same as PR11:14
@HeikoSsanuj: what happens if you remove the line 16 in your example?11:14
@HeikoSsame error?11:14
sanujit segfaults11:14
sanujHeikoS, i think the error is in init_params() of mock object11:15
@HeikoSsanuj: ^11:15
sanujHeikoS, any destructor deletes the member variable11:16
sanujany.h destructor11:16
sanujnot sure though11:16
sanuj==15743==    at 0x4026FD: shogun::Any::~Any() (in /home/sanuj/Projects/shogun/buildpy/a.out)11:16
@HeikoSsanuj: can you remove the set and tell me what happens?11:16
@HeikoSline 1611:17
sanuj5 mins11:17
sanujremoving all sets11:19
@HeikoSjust line 16 in example11:20
@HeikoSand add it to the gist11:21
@HeikoSdont overwrite the old11:21
@HeikoSso that we can compare11:21
sanujoh i had removed set in MockObject and built it11:29
sanujundoing and rebuilding11:29
sanujHeikoS, do you know someone in UCL who works on deep learning related stuff and can take interns like me for 6 months or so11:32
sanujor anywhere else also apart from UCL11:32
@HeikoSsanuj: if you wanna do deeplearning, better go to deepmind11:32
@HeikoSthey took all good people11:32
@HeikoSbut I know a few folds at UCL11:32
@HeikoSwhod o ml11:32
sanujHeikoS, deepmind wont take people with just undergraduate degrees11:33
@HeikoSi see11:35
@HeikoSinternships are good11:35
@HeikoSneed to find some good group11:36
@HeikoSthat matches you11:36
sanujHeikoS, really?11:37
sanuji have been mailing profs but no one replies back :P11:37
@HeikoSsanuj: haha!11:38
@HeikoSpeople are busy11:38
@HeikoSand they get mails like this all the time11:38
@HeikoSwho did you write?11:38
sanujwell i say that11:38
sanuji have read your recent work11:38
sanujrespect what you do and would like to work with you to be a part of it11:39
sanujand send my CV11:39
@HeikoSwho, not what11:39
sanuji updated the gist11:39
sanujDavid hsu from NUS - works on robotics and reinforcement learning11:40
sanujNando de freitas11:40
sanujhonglak lee11:41
-!- GandalfTheWizard [~Eva@] has quit [Quit: Leaving.]11:41
@HeikoSI see11:41
@HeikoSthey are all too busy11:41
sanuj^ univ of michigan11:41
sanujsatinder singh from univ of michigan too11:42
sanujand some other profs who work on deep reinforcement learning and related stuff11:42
sanujnow i have switched to mailing PhDs working in good labs to ask if they have a vacancy11:43
@HeikoSI see11:43
@HeikoSthey usually cannot decide11:43
@HeikoSah it is hard11:43
@HeikoSto stick out of the noise11:43
@HeikoSI get lots of emails from indian students actually11:43
sanujHeikoS, haha11:43
@HeikoSbut takes too much time to answer all11:43
sanujHeikoS, tell me if you find a good group, i would like to join after gsoc11:44
@HeikoSsanuj: what do you want to do?11:44
@HeikoSthat is the biggest question11:44
sanujHeikoS, i like what deepmind is doing11:45
sanujanything related to that11:45
sanujbut i'm flexible11:45
@HeikoSI see11:46
@HeikoSand which place?11:47
@HeikoSyou like?11:47
sanujUSA, Canada, UK, Europe, Singapore11:47
sanujHeikoS, will it be difficult in germany if i don't know german? :D11:48
@HeikoSsanuj: some unis have english speaking groups11:48
@HeikoSsanuj: ok so we get the same error when you remobed the set11:49
sanujit is from init_param()11:49
@HeikoSso my hunch is: your tag implementation tries to delete the member variable11:49
@HeikoSsince its happening in Any destructor11:49
sanujyes, i think the same11:50
sanujbut we need a way around it11:50
sanuji thought about removing that from any.h destructor and using something like SG_REMOVE11:50
@HeikoSI dont understand the problem fully yet11:51
@HeikoScan you maybe re-build your example without optimizations?11:54
@HeikoSso it doesnt say11:54
@HeikoSat 0x4024A9: shogun::Any::~Any() (in /home/sanuj/Projects/shogun/buildpy/a.out)11:54
-!- GandalfTheWizard [~Eva@] has joined #shogun11:54
@HeikoSbut gives the line number at the end?11:54
@HeikoSI dont understand why the desctructor is called11:54
sanujHeikoS, debug mode in cmake?11:54
@HeikoSsanuj: check the error output11:55
@HeikoSshogun's line numbers are there11:55
@HeikoSbut not the ones of your example11:55
@HeikoSthere if only sais a.out11:55
@HeikoSbut no line number11:55
sanujoh i see11:55
@HeikoSsanuj: so the problem is: registering member variables in the parameter map11:56
@HeikoSwe are already down to that11:56
sanujHeikoS, i used g++ -O012:01
sanujnot getting the line numbers12:01
@HeikoSthere is a debug option12:01
@HeikoSthat includes line number infos12:01
sanujHeikoS, updating12:03
sanujHeikoS, updated12:04
@HeikoSthe example line numbers dont matcht he listing you put12:06
@HeikoSline 17 in error message is nothing12:06
@HeikoSI think since you remove a line12:06
sanujHeikoS, it's line 12 in the error message for temp.cpp12:07
@HeikoSah yeah12:08
@HeikoSok then12:08
@HeikoSso why is desctructor called?12:08
@HeikoSof any12:08
@HeikoSwhy is that done?12:09
@HeikoSdid you maybe update the "overwritign" machanism and forget to consider the case where something is newly added to the parameter map?12:10
sanujHeikoS, erase_type creates an object of Any12:10
sanujusing the value from the argument12:11
sanujbecause parameter map stores string --> any12:11
sanujHeikoS, sorry, it stores BaseTag ---> any12:11
sanujHeikoS, the diff between public set and protected set is that public set only sets if the provided basetag already exists in the map12:13
@HeikoSsanuj: maybe a better name should be chosen12:13
@HeikoSlike "add"12:13
@HeikoSor so12:13
@HeikoSand "set" is for public use12:13
sanujcurrently it is _set12:13
sanuji'll use add12:13
@HeikoSso why is any destructor called?12:14
sanujeven i'm asking the same question12:14
@HeikoSfrom erase_value?12:15
sanujHeikoS, i had used the exact same method in the old implementation12:15
sanuji just made it protected12:15
@HeikoSthere you return by value12:16
@HeikoSsanuj: so the way to solve this12:16
@HeikoSis to use a debugger12:16
@HeikoSand to go through the code step by step12:16
@HeikoSunderstand what it does12:16
@HeikoSI cannot do that from the PR12:16
@HeikoSsince I cannot brwose the code12:16
sanuji'll do that12:17
@HeikoStoo many redirections for my mind :)12:17
@HeikoSsanuj: you know how to use gdb?12:17
@HeikoSor ddd maybe?12:17
sanujnow i'll find learn it from somewhere12:17
@HeikoSif you already compiled with debug symbols it is easy12:17
@HeikoSwhich OS?12:17
sanujwhich is ubunut12:17
@HeikoSok install "ddd"12:17
-!- GandalfTheWizard [~Eva@] has quit [Quit: Leaving.]12:19
@HeikoSsanuj:  and then start it, and just load your a.out12:19
@HeikoSthen set a breakpoint before the line 1212:19
@HeikoSand then run12:19
@HeikoSand then to "single steps"12:19
sanujHeikoS, btw can you merge
@HeikoSif it goes to far in, you can always "step return"12:20
sanujthis is holding my neural net cookbook since 2 weeks12:20
@HeikoSsanuj: did you ever use any debugger?12:20
@HeikoSi.e. you know "step next" "step return" etc?12:20
@HeikoSand breakpoints12:20
sanuji know12:20
@HeikoSso ddd does that for your c12:20
@HeikoSgiven it is compiled with debug mode12:20
sanujoh wow, it has a gui12:21
@HeikoSsanuj: so use that to find out whats wrong12:22
@HeikoSyou can just read what all the variables are12:22
@HeikoSgood exercise to understand c memory errors ;)12:23
@HeikoSsanuj: I am here for a bit more, let me know your findings12:23
@HeikoSsanuj:  and?12:29
sanuji set a breakpoint on line 1112:30
sanuj10 actually12:31
sanujthen i do next12:31
sanuj and it reaches the any destructor12:31
sanujHeikoS, ^12:33
@HeikoSyou gotta do single steps12:33
@HeikoSor put the breakpoint earlier, sometimes ddd is confusing12:33
@HeikoS"step in"12:34
@HeikoSnot "step over"12:34
@HeikoSstep over exectues all code by the active line12:34
@HeikoSbut you want to jump  inside the call12:34
sanujHeikoS, there is "step" and "next"12:34
@HeikoSsanuj: I dont know how they are called12:34
sanuj"next" says = proceeding through subroutine calls12:34
@HeikoSbut if it goes to destructor, it obvioulsy tries to execute the whole line12:34
sanujoh okay12:35
@HeikoSgotta go12:39
@HeikoSsanuj: will check back later tonight12:39
@HeikoSfeel free to send an email12:39
@HeikoSshould be easy to track the problem with the ddd12:39
@HeikoSsanuj: see you!12:39
sanujHeikoS, cool12:39
@HeikoSthanks to you too! :)12:39
-!- HeikoS [] has quit [Quit: Leaving.]12:39
-!- sanuj [~sanuj@] has quit [Ping timeout: 244 seconds]12:44
-!- lambday [569de6b3@gateway/web/freenode/ip.] has joined #shogun12:46
-!- mode/#shogun [+o lambday] by ChanServ12:46
-!- sanuj [~sanuj@] has joined #shogun13:13
-!- lambday [569de6b3@gateway/web/freenode/ip.] has quit [Ping timeout: 250 seconds]13:15
-!- travis-ci [] has joined #shogun13:41
travis-ciit's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []13:41
sanujBZDMG, hey14:03
-!- shogun-notifier- [] has quit [Quit: transmission timeout]14:08
Saurabh7__wiking: yo14:44
shogun-buildbotbuild #569 of deb1 - libshogun - PR is complete: Failure [failed cookbook]  Build details are at  blamelist: sanuj15:41
shogun-buildbotbuild #570 of deb1 - libshogun - PR is complete: Success [build successful]  Build details are at
-!- sanuj [~sanuj@] has quit [Ping timeout: 264 seconds]16:49
-!- sanuj [~sanuj@] has joined #shogun17:21
-!- Saurabh7__ [Saurabh7@gateway/shell/panicbnc/x-wlacyikiyzqrkfcs] has left #shogun []18:57
sanujBZDMG, there?19:27
-!- sanuj [~sanuj@] has quit [Remote host closed the connection]19:37
-!- shogun-notifier- [] has joined #shogun22:21
shogun-notifier-shogun: Sanuj :develop * 4375d99 / src/interfaces/modular/NeuralNets.i,src/shogun/neuralnets/NeuralLayers.h:
shogun-notifier-shogun: add %newobject in neuralnet interface22:21
shogun-notifier-shogun: Sergey Lisitsyn :develop * 8cb801e / src/interfaces/modular/NeuralNets.i,src/shogun/neuralnets/NeuralLayers.h:
shogun-notifier-shogun: Merge pull request #3283 from sanuj/nn_cleanup22:21
shogun-notifier-shogun: add %newobject in neuralnet interface22:21
shogun-buildbotbuild #740 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: Sanuj <>, Sergey Lisitsyn <>22:21
-!- sonne|osx [] has joined #shogun23:36
-!- sonne|osx [] has quit [Quit: sonne|osx]23:57
--- Log closed Sun Jun 26 00:00:56 2016