Open in new window / Try shogun cloud
--- Log opened Sun Jun 26 00:00:56 2016
-!- shogun-notifier- [] has quit [Quit: transmission timeout]01:21
-!- arianepaola [~ariane@unaffiliated/arianepaola] has quit [Ping timeout: 276 seconds]04:46
-!- sonne|osx [] has joined #shogun05:24
-!- Saurabh7__ [Saurabh7@gateway/shell/panicbnc/x-wlacyikiyzqrkfcs] has joined #shogun05:39
shogun-buildbotbuild #28 of memleak - valgrind is complete: Failure [failed memory check generate html]  Build details are at  blamelist: Heiko Strathmann <>, Sanuj <>, Sergey Lisitsyn <>, OXPHOS <>05:39
shogun-buildbotbuild #1029 of nightly_none is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>, Sanuj <>, Sergey Lisitsyn <>, OXPHOS <>05:43
@wikingSaurabh7__: y005:44
Saurabh7__wiking: yo05:49
Saurabh7__sry having bfast :)05:49
@wikingSaurabh7__: ok so i replied in the PR05:49
@wikingno worries05:49
@wikingbut we should discuss things about the trees05:50
@wikingas there are some stuff that needs to be fixed there05:50
@wiking(see those static analytics errors as well)05:50
@wikinglemme know when it's a good time for you05:50
Saurabh7__brb 5 min05:50
-!- sonne|osx [] has quit [Quit: sonne|osx]05:51
Saurabh7__wiking: there ?05:58
Saurabh7__so you mean the feature type itself05:58
Saurabh7__current implementation assumes some feature might be nominal , some continous05:59
Saurabh7__i.e feature1 is nominal, feature2 is continous06:02
Saurabh7__wiking: so you think we should allow stringfeatures and convert them internally ?06:07
@wikingwhat do you mean?06:13
@wikingi mean the thing is that trees should work06:14
@wikingout of box on values like hot, cold etc06:14
Saurabh7__yes thats right06:14
@wikingok so that's wha ti mean by06:15
@wikingwe shoudl support string features out of box06:15
@wikingin case of trees06:15
@wikingwe shoudl not require anybody to encode them as floats or something06:15
@wikingsince trees usually should be able to handle string features06:16
@wikingi mean the ohter thing is06:16
@wikingshoudl be supported by trees06:16
@wikingwhich could consist of StringFeatures and DenseFeatures06:17
@wikingand i've added some errors that are detected by static analysis of the code06:17
@wikingthat should be fixed as well06:18
@wikingbut more importantly06:18
@wikingin trees we should totally fix this error06:18
@wikingthat we only support dense features06:18
Saurabh7__ok , combined features look good, we already know the underling types06:19
Saurabh7__rather than the current way of passing a bool vector06:19
@wikingbut currenlty we cannot support string features06:19
@wikingok so that should be priority here06:20
@wikingto add support for that06:20
@wikingas it's a bit nonesense06:20
@wikingi mean it's great these speedup improvements06:20
@wikingbut then again this is a fundamental thing06:21
@wikingi mean you dont have to do this within this pr06:21
@wikinglets have those static analysis errors fixed in CARTree06:22
@wikingbut then i really reckon the next step should be fixing this in CART06:22
@wiking(and in all other trees)06:22
shogun-buildbotbuild #1159 of nightly_default is complete: Failure [failed test test_1 notebooks]  Build details are at  blamelist: Heiko Strathmann <>, Sanuj <>, Sergey Lisitsyn <>, OXPHOS <>06:36
-!- arianepaola [~ariane@unaffiliated/arianepaola] has joined #shogun06:51
-!- sanuj [~sanuj@] has joined #shogun07:09
@wikingSaurabh7__: still here?07:17
Saurabh7__wiking: yes07:17
@wikingok so another question07:17
@wikingshouldn't we actually make the RandomForest if not templateable07:18
@wikingbut at least to set the tree implementation07:18
@wikinghave like CART as a default tree07:18
@wikingbut have the option to use different tree implementation07:19
Saurabh7__ah so different trees07:19
@wikingyeah because you could use07:19
Saurabh7__yes that could be07:19
@wikingor C5.0 for a random forest07:19
@wikingor chaid for that matter07:19
@wikingi mean the BaggingMachine is flexible07:19
@wikingit can do bagging with *any* machine07:19
@wikingnow RandomForest should be a bagging machine with for all trees07:20
@wikingmaybe have a default tree it would use07:20
@wikingbut then again the user should be able to override this easily07:20
@wikingsince there are sometimes cases when other trees are more appropriate07:21
Saurabh7__that makes sense, can be done by setting machien parameters differently07:22
Saurabh7__for random forest depending on the tree07:22
-!- sanuj [~sanuj@] has quit [Ping timeout: 260 seconds]07:57
-!- sanuj [~sanuj@] has joined #shogun09:54
sanujBZDMG, yo10:00
-!- HeikoS [] has joined #shogun11:15
-!- mode/#shogun [+o HeikoS] by ChanServ11:15
-!- shogun-notifier- [] has joined #shogun11:31
shogun-notifier-shogun-data: Sanuj :master * 11a34f8 / testsuite/meta/regression/ (2 files):
shogun-notifier-shogun-data: rename mkl regression cookbook data11:31
shogun-notifier-shogun-data: Heiko Strathmann :master * 7087f26 / testsuite/meta/regression/ (2 files):
shogun-notifier-shogun-data: Merge pull request #109 from sanuj/mkl_regression11:31
shogun-notifier-shogun-data: rename mkl regression cookbook data11:31
sanujHeikoS, hallo11:44
shogun-notifier-shogun-data: Sanuj :master * 60692de / testsuite/meta/neural_nets/feedforward_net_classification.dat:
shogun-notifier-shogun-data: add data for nn cookbook11:51
shogun-notifier-shogun-data: Heiko Strathmann :master * 494032f / testsuite/meta/neural_nets/feedforward_net_classification.dat:
shogun-notifier-shogun-data: Merge pull request #110 from sanuj/nn_cookbook11:51
shogun-notifier-shogun-data: add data for nn cookbook11:51
shogun-notifier-shogun-data: OXPHOS :master * c6f46ec / testsuite/meta/gaussian_processes/gaussian_process_classifier.dat:
shogun-notifier-shogun-data: gp classifier cookbook integration test dataset11:54
shogun-notifier-shogun-data: Heiko Strathmann :master * 5bcd88d / testsuite/meta/gaussian_processes/gaussian_process_classifier.dat:
shogun-notifier-shogun-data: Merge pull request #108 from OXPHOS/master11:54
shogun-notifier-shogun-data: gp classifier integration test dataset11:54
shogun-notifier-shogun: OXPHOS :develop * 7dccabb / src/shogun/multiclass/ShareBoost.cpp:
shogun-notifier-shogun: update error msg  in shareboost11:59
shogun-notifier-shogun: Heiko Strathmann :develop * 3dbeb8a / src/shogun/multiclass/ShareBoost.cpp:
shogun-notifier-shogun: Merge pull request #3298 from OXPHOS/shareboost11:59
shogun-notifier-shogun: update shareboost error msg / update shareboost nonzero feat num boundary11:59
shogun-buildbotbuild #741 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>, OXPHOS <>12:00
shogun-notifier-shogun: OXPHOS :develop * a1fb210 / / (42 files):
shogun-notifier-shogun: separate classifiers in meta examples12:00
shogun-notifier-shogun: Heiko Strathmann :develop * d6353fe / / (42 files):
shogun-notifier-shogun: Merge pull request #3297 from OXPHOS/cookbook_index12:00
shogun-notifier-shogun: separate classifiers in cookbook12:00
shogun-buildbotbuild #742 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: OXPHOS <>12:00
shogun-buildbotbuild #743 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>12:01
sanujHeikoS, hallo12:04
sanuji was not able to solve the any.h destructor bug12:04
@HeikoSsanuj: did you debug it?12:04
sanujyes, i did that12:05
@HeikoSso our question was: why is it called, from where12:05
sanujneed to share a link12:05
sanujHeikoS, when this happens
sanujthis is called
sanujand i think the destructor was called from here12:07
@HeikoSthis is because the old one is overwritten?12:07
sanujnot sure, but how is it overwritten?12:08
@HeikoSsanuj: map[tag]12:08
@HeikoSmight contain another any12:09
@HeikoSsanuj: so we have to solve this12:09
sanujmap[tag] should be empty right12:09
@HeikoSsanuj: yeah it should12:09
sanujit is the first time something is set12:09
@HeikoSwhy would the destructor then be called?12:09
@HeikoSis the assignment operator maybe doing it?12:10
sanujHeikoS, shall i share the logs of ddd with you?12:10
@HeikoScausing it12:10
@HeikoSsanuj: we have two options12:10
@HeikoSyou figure it out, or I have a look12:10
@HeikoSI prefer you doing it :D12:10
@HeikoSbut for that, you need to dig a bit12:10
sanujHeikoS, setting was working fine before i made these changes....destructor was not being called before12:11
@HeikoSsanuj: these changes as in?12:11
sanujlet me call ddd again, i did this yesterday so forgotten some stuff12:11
@HeikoSwhat are "these changes"12:11
sanujHeikoS, new protected set and calling set in init_params() with member variables12:12
-!- oxphos [c68f0c0c@gateway/web/freenode/ip.] has joined #shogun12:12
@HeikoSsanuj: I see12:12
@HeikoSsanuj: ok then, I suggest:12:12
@HeikoSsanuj:  you (partly) revert them for bug hunting12:12
@HeikoSmake it like this12:12
@HeikoSsanuj: or actually, one more question12:12
oxphos@HeikoS hey I'm on my cellphone12:12
@HeikoSso registering *member* variables inside a class worked before you made "these changes"?12:13
@HeikoSoxphos: I see, hi! :)12:13
@HeikoSoxphos: no worries then, just wanted to discuss a bit12:13
@HeikoSoxphos: but no point on cellphone12:13
sanujHeikoS, i was never registering member variables before12:13
@HeikoSsanuj: why did you say that it worked then?12:13
sanujand there is one more thing that worries me12:13
sanujHeikoS, it was able to set random parameters12:14
@HeikoSsanuj: we need to be a bit more systematic if we want to solve this12:14
@HeikoSsanuj: so let's do this then: make a backup of the changes you did12:14
@HeikoSthen revert them12:14
@HeikoSthen try to register a member variable12:14
@HeikoSI am pretty sure you will get the same error12:14
@HeikoSbut we need to check12:14
sanujokay, i also think so12:14
@HeikoSand then finding it, we are not distracted by your latest changes12:14
sanujlet me do it12:14
@HeikoSsanuj: we need to reduce the search space of potential error here12:15
@HeikoStoo many things at once and it is not clear what causes what12:15
@HeikoSthis statemeht "it worked before these changes", is neither checked nor true at this moment12:15
@HeikoSso careful when thinking about things12:15
@HeikoSneed to really be sure whats going on12:16
sanujHeikoS, sure sure12:16
@HeikoSif one convolutes things in ones mind, then finding errors becomes very hard12:16
@HeikoSright now we are not sure whether it is your update or a general problem that causes this member registering error12:16
@HeikoSor maybe even the framework is not designed for registering member variables?12:17
oxphosHeikoS okay will revise the schedule in 6h and will be back to laptop client tmr12:17
@HeikoSnot clerar12:17
@HeikoSoxphos: cool! perfect12:17
@HeikoSsanuj: cool then, Ill see you later, lets keep going on this, try to think about what good next steps could be, dont stop if there is a problem12:18
@HeikoSfinally, if nothing else works, you can push everything to a local feature branch in your own shogun fork12:18
@HeikoSsanuj: and then I can clone that and debug locally12:18
-!- oxphos [c68f0c0c@gateway/web/freenode/ip.] has quit [Quit: Page closed]12:19
sanujHeikoS, yes12:19
sanujHeikoS, danke :)12:19
@HeikoSbitte :)12:19
-!- HeikoS [] has quit [Remote host closed the connection]12:20
-!- sanuj [~sanuj@] has quit [Quit: Leaving]13:43
BZDMGHeikoS_web: tldr what error?14:06
-!- sonne|osx [] has joined #shogun14:12
-!- sanuj [~sanuj@] has joined #shogun14:36
BZDMGwhat's the bug you're talking about14:40
sanujsee error_2.txt14:41
sanujBZDMG, it happens when _set is called in init_params() of MockObject14:42
BZDMGok where is the actual one14:42
sanujapparently Any destructor is called and deletes the member variable14:42
sanujsending link14:43
BZDMGyou didn't call CSGObject ctor14:44
BZDMGin the MockObject ctor14:44
BZDMGI think the map is not really initialized14:44
sanujso in this commit i have added a protected _set() in SGObject14:45
sanuji undid this commit on my local14:45
sanujand did the same thing with the old public set() in SGObject14:46
sanujthat seems to be working without SGObject ctor14:46
BZDMGyou should call it anyway14:46
BZDMGI am not sure I understand what you did14:48
BZDMGwhat commit? :)14:48
sanujBZDMG, this commit
BZDMGehrr that's too many changes14:53
BZDMGdid ctor help?14:53
sanujwait let me switch branches and build it14:54
sanujBZDMG, it's building, will take time14:59
sanujBZDMG, how's work?15:00
-!- shogun-notifier- [] has quit [Quit: transmission timeout]15:00
BZDMGsanuj: bit busy as you can see haha15:01
BZDMGstarting some project soon15:01
BZDMGwill be better soon15:02
sanujBZDMG, do you read deepmind's papers?15:02
sanujjust asking15:02
BZDMGthey are doing some atari reinforcement learning :)15:03
sanujyou are interested in deep learning15:03
sanujbut some recent work is different15:03
-!- sonne|osx [] has quit [Quit: sonne|osx]15:04
sanujthings like15:04
sanujneural turing machine (2014)15:04
sanujneural programmer-interpreters15:04
sanujand memory augmented nn15:05
sanujare cool15:05
sanujBZDMG, shogun doesn't have RNNs and LSTMs15:05
BZDMGactually I am not a fan of that synthetic stuff like neural turing machine15:05
sanujBZDMG, were you ever talking about using tensorflow in shogun?15:06
BZDMGyeah I thought of that15:06
BZDMGbut heiko said optimizer is in python15:06
BZDMGso we can't really do anything useful just from C++15:06
sanujBZDMG, still segfaults15:18
BZDMGsanuj: is it the same error?15:21
sanujBZDMG, yes15:22
BZDMGsanuj: I am pretty sure something is not initialized15:25
BZDMGyou should probably check it with gdb15:25
sanuji did that15:25
sanujgives you lots of stuff15:26
sanuji used ddd15:26
BZDMGsanuj: well may be you can print `self` in init_params15:28
BZDMGare you sure you've added base class ctor call and recompiled it?15:28
sanujBZDMG, i did this15:29
sanujCMockObject() : CSGObject()15:29
sanuj    {15:29
sanuj        init_params();15:29
sanuj    }15:29
BZDMGlooks valid15:29
BZDMGit should not work w/o15:29
BZDMGor work by chance15:29
BZDMGwhich is not any better15:29
sanujBZDMG, did you see protected _set that i added in SGObject?15:30
BZDMGyeah but I am a bit lost in _set set and _set15:30
sanuji think protected _set() might be calling itself instead of private _set()15:31
BZDMGoh my god15:31
BZDMGare they of the same name ?15:31
BZDMGprivate and protected?15:31
sanujyes :/15:31
BZDMGplease dont do this :D15:31
BZDMGbetter call it kawabanga15:32
sanuji'll call this add15:32
BZDMGcall it set_init15:32
sanuji thought it would differentiate with arguments15:32
sanujlet me see if this is causing the error15:32
BZDMGeven if it does it is way too dangerous15:33
sanujbuilding will take time again15:34
sanujBZDMG, btw is "add" fine or shall i call it something else?15:34
BZDMGadd is fine to me15:34
sanujBZDMG, if this is going to be the problem then i might bang my head on the wall :P15:35
sanujBZDMG, does it happen to you that you made some really silly mistake and spent lots of time debugging it ? :D15:38
-!- sonne|osx [] has joined #shogun15:47
sanujBZDMG, there?16:16
-!- sanuj [~sanuj@] has quit [Ping timeout: 260 seconds]16:23
-!- sanuj [~sanuj@] has joined #shogun17:51
sanujBZDMG, there?18:17
-!- sanuj [~sanuj@] has quit [Ping timeout: 240 seconds]18:45
-!- sanuj [~sanuj@] has joined #shogun18:49
-!- sonne|osx [] has quit [Quit: sonne|osx]18:56
-!- sanuj [~sanuj@] has quit [Remote host closed the connection]18:57
--- Log closed Mon Jun 27 00:00:57 2016