Open in new window / Try shogun cloud
--- Log opened Sun May 29 00:00:41 2011
@sonney2kmore involved than I thought...00:25
CIA-32shogun: Soeren Sonnenburg master * r52ce59f / (82 files in 9 dirs): rename preproc -> preprocessor -
@sonney2know for the CClassifier rename...00:32
-!- alesis-novik [~alesis@] has quit [Quit: Leaving]01:59
-!- blackburn [~qdrgsm@] has joined #shogun10:18
@sonney2kblackburn, around?10:52
@sonney2kI've done the preproc PreProc etc transition10:52
@sonney2kdoes it that look good to you or do you have any other ideas for improvement?10:52
@sonney2kI am doing the machine transition now10:53
@sonney2kCurrent plan is to create a directory machine and then  move the KernelMachine DistanceMachine in there10:53
@sonney2kin addition Classifier and rename it to Machine10:53
@sonney2kand also LinearClassifier - and name it LinearMachine10:54
splovingsonney2k, I debug one day for lua MatrixTest.lua example and now it works. I pushed it. when you have time please take a look at it. l8r11:31
@sonney2ksploving, nice11:32
@sonney2ksploving, not sure why but this thing cannot be applied11:35
@sonney2kI think you need to rebase your patch against git master11:35
CIA-32shogun: Soeren Sonnenburg master * r54150e8 / (65 files in 10 dirs):11:41
CIA-32shogun: Transistion: rename CClassifier to CMachine11:41
CIA-32shogun: correspondingly rename CLinearClassifier -> CLinearMachine11:41
CIA-32shogun: and move them to the machine/ subdirectory -
@sonney2kblackburn, ok we have Machines now :D11:41
@sonney2ksploving, btw thanks for the status email to the mailinglist!11:42
splovingsonney2k, you are welcome. git fetch && git merge upstream/master. fatal: Entry 'src/configure' would be overwritten by merge. Cannot merge12:01
-!- Tanmoy [56ce19a2@gateway/web/freenode/ip.] has joined #shogun12:10
-!- Tanmoy [56ce19a2@gateway/web/freenode/ip.] has quit [Client Quit]12:13
blackburnsonney2k: now yes12:35
blackburnsonney2k: I have only one idea about preproc - to structurize it some another way, i.e. now dim reduction processor are all inherited from SimplePreproc<float64_t> - looks strange12:41
blackburnI'd like to study more about MDS for genomic data sets :) will try a bit later12:47
@bettybooblackburn, smart!12:48
blackburnbettyboo: just like you12:48
@bettybooblackburn: well then send another request: btw I also noted that you don't have newlines after and before loops ... that is also sth I would like to have for readability12:48
splovingblackburn, /root/shogun/src/libshogun/classifier/DistanceMachine.cpp:32: multiple definition of `shogun::CDistanceMachine::~CDistanceMachine()'12:50
splovingwhat does that mean? I  reset my repo to the trunk, but when compile it shows the following error12:51
blackburnsploving: will try it now too12:51
splovingsonney2k, I reset it to the uptream, change my files need commit and pushed it!12:58
splovingblackburn, what is going?13:06
blackburnsploving: compiling :)13:07
splovingblackburn,so slow? how many cors of you machine?13:10
blackburnit is 2 core T4300 processor in my notebook13:11
splovingyou could use make -j413:11
blackburnwhat it is?13:11
blackburnjobs, see now13:12
blackburnsploving: done, everything is ok13:13
blackburnno error you mentioned before13:13
splovingyou mean libshogun and python?13:15
splovingI compile lua13:15
splovingI will have a try again13:16
blackburnanyway I think there is no difference13:16
blackburnbecause the error is in libshogun13:16
blackburndamn it is fast!13:16
blackburnthank you for -j4 advice13:16
-!- sploving [~root@] has quit [Ping timeout: 240 seconds]13:20
@sonney2kblackburn, so looks ok for you ?14:36
blackburnsonney2k: preproc?14:38
@sonney2kblackburn, preproc and machine14:55
blackburnsonney2k: yeah it is ok14:55
@sonney2khmmhh sploving above didn't do make distclean or git clean -dfx before14:55
@sonney2kblackburn, I am not sure if the machine directory shouldn't have the regression etc stuff as subdirs14:55
@sonney2kblackburn, should machine's classify function be renamed to apply ?14:56
@sonney2kor do you have a better name?14:56
blackburnsonney2k: train - apply, right?14:56
@sonney2kblackburn, good for me14:56
@sonney2ktrain we have14:56
blackburnsonney2k: for me too14:56
@sonney2kbut classify needs a rename to apply() then14:57
@sonney2kthat has to be done manually though14:57
blackburnit makes sense for clustering and regression issues14:57
@sonney2kcurrently we have14:57
blackburnabout CDistanceMachine14:57
blackburnit have nice thread based distance calculation routines14:58
blackburnwhy we have not ones in distance?14:58
@sonney2kblackburn, IIRC distance has those already14:58
blackburnsonney2k: for example get_distance_matrix do it without threads14:59
@sonney2kblackburn, but I would say you only comment on the structure of shogun and do the remaining things afterwards14:59
@sonney2kblackburn, i think it does with threads14:59
@sonney2kI mean you have exams which are more important for now14:59
blackburnno problem with exams, will study it this evening :)15:00
blackburnsonney2k: classifier, clustering might be subdirs of machine15:00
@sonney2kif distance matrix is not yet pthreaded - submit patches sure15:00
@sonney2kblackburn, yeah - thought so too15:01
@sonney2kregression too15:01
blackburnsonney2k: I am not very familiar with threads used here, will be think about just after finishing some important steps in my dimreduction15:02
blackburnsome of algos should be parallelized for sure15:02
@sonney2kblackburn, yeah15:02
@sonney2kvery welcome...15:02
blackburnand I would implement kd-tree or balltree15:02
blackburnseems working with instance-learning (or neighborhood-based, etc) algos is my fate :D15:03
@sonney2kblackburn, there is a lightning fast kdtree implementation available already15:03
blackburnbecause I started shogun developing with kNN and my project uses it too15:03
@sonney2kby john langford15:03
@sonney2kso we could borrow his code :)15:04
blackburnwell it simplifies work :)15:04
@sonney2kbut structure first15:04
blackburnjust adapt it to shogun15:04
@sonney2kso what do you think should I move the machine derived things into machine ?15:04
blackburnyeah, of course15:05
@sonney2kand should I make subdirs there or not15:05
@sonney2kprobably or?15:05
blackburnsubdirs sounds nice15:05
@sonney2kI mean regression -> machine/regression15:05
blackburnbtw, may be some more namespaces should be nice too15:05
@sonney2kblackburn, but how do we reflect that change in e.g. python_modular?15:05
blackburnwhat do you mean?15:06
@sonney2kcurrently it is from shogun.Classifier import *15:06
blackburnshogun.Machine.Classifier or shogun.Classifier?15:06
@sonney2kwe don't want shogun.Machine.Classifier I think15:06
blackburnyeah sounds cumbersome15:06
@sonney2kso I am not sure if we should actually move things already15:07
@sonney2kshogun isn't that big yet...15:07
blackburnis it possible to make shogun.Classifier just a shorthand name for libshogun/machine/classifier?15:07
@sonney2kif we were numpy then it makes sense15:07
@sonney2kblackburn, sure but it is not consistent then15:07
@sonney2kI would prefer the same structure in C++ as in *_modular15:07
@sonney2kotherwise it will be confusing15:07
blackburnsounds reasonable15:07
blackburnokay so let it be the way it is now15:08
@sonney2kI would then add a shogun.Machine for the modular interfaces15:08
CIA-32shogun: Baozeng Ding master * r6688b5c / (4 files in 2 dirs): use lua_objlen to count table size, improve error handing, and add lua vector/matrix test -
blackburnthe only issue - looks confusing then classes in some 'package' are inherited from the other one15:10
blackburnbut no other way now15:10
blackburnsonney2k: btw 'make -j4' rocks, it makes shogun compile fast!15:10
@sonney2kblackburn, I use that all the time of course15:11
@sonney2k(I have it set in MAKEFLAGS='-j4')15:11
@sonney2kso I don't have to specify it15:11
blackburnwhere I can set it?15:11
blackburnah. i see15:12
@sonney2ki have my son now on my left arm...15:12
@sonney2kslight handicap15:12
blackburnsonney2k: btw PreProc was wrong indeed15:13
blackburnlooks like wrong camel-case15:13
blackburncause word is Preprocessor not pre processor and so on15:13
blackburnso I like it more :)15:13
@sonney2kme too15:14
@sonney2kmore readable15:14
blackburnjust adapted LLE for new naming15:15
@sonney2kand we avoid unneeded shortcuts anyway15:15
@sonney2kLLE should also have a proper name :)15:15
blackburnanyway it is not in trunk so I have much time to think about how to name it15:16
blackburnbut LocallyLinearEmbedding is bad too15:16
@sonney2kit says more than LLE I think15:16
blackburnsonney2k: I have a very simple idea about codestyle15:17
blackburnjust modify it
@sonney2kwhich link exactly?15:21
blackburnI just mean if we will have an 'official' codestyle15:21
blackburnnew developers should not make some not-shogun-style code15:21
blackburnit could simplify your work15:21
@sonney2kwe have that though not as erfect15:23
@sonney2kcookie time15:23
blackburnwell then I will have some game theory time :D15:26
-!- blackburn [~qdrgsm@] has left #shogun []15:58
-!- blackburn [~qdrgsm@] has joined #shogun20:13
serialhexOOH!!!! I WANNA POKE SOMEONE IN THE EYE!!!!21:29
serialhexsonney2k: so i keep getting this error when i compile shogun, i have the configure string i'm using & the last little bit of the output.  if you have any ideas of things you want me to look at let me know, i've got a few things i'm doing and will get back to you asap!21:37
@sonney2kserialhex, look in the mirror ;-)21:38
serialhexin the mirror??21:38
@sonney2kwhich error?21:38
@sonney2k^^eye poking :D21:38
serialhexone sec :P21:38
@sonney2kbut only the left eye please ;-)21:38
serialhexwhy?  left eye wonky or somethin??  can you see better out of your right eye?21:39
@sonney2kserialhex, do a git clean -dfx21:39
serialhexanything else before i try the make?21:40
* serialhex is configurin & makin!21:41
serialhexi've got some ppl coming over in a minute sonney2k so i've got to handle that & i'll let you know how things go, all right?21:43
serialhex(and besides, the compiling takes FOR FREAKING EVAR!!!)21:43
@sonney2kserialhex, heh21:43
@sonney2kdeal with them - let them choose the weapons ;-)21:44
serialhexthey're here!  i'll bbl!!21:44
blackburndamn stochastic game with shity bellman equations :E21:51
@sonney2kblackburn, what do you think should I *add* a apply() and apply_to_example() function to all the classifiers and make that one call the classify() / classify_example() one that already exists in classifiers or should I rename the classify() etc functions in classifier's?22:44
blackburnsonney2k: to _add_ sounds easily :)23:18
@sonney2kblackburn, I did the other thing but now I am not convinced that this was the right deision23:19
@sonney2kdiaper time...23:19
blackburnyou renamed it?23:19
blackburnnot bad too23:19
@sonney2kblackburn, I don't know... it will break all scripts..23:26
blackburnyeah a little revolution23:27
@sonney2kCMachine / CClassifier rename is not intrusive for the interface side23:27
@sonney2kso I don't like it23:27
blackburnwell time for some sleep23:30
blackburnthree days - three exams to go :D23:30
blackburnsonney2k: see you23:31
@sonney2kI tend to revert the changes ...23:31
blackburnwhich changes have you done?23:31
@sonney2kand only add the apply function23:31
@sonney2kblackburn, all the replacement classify -> apply23:31
@sonney2kclassify_example -> apply_example23:32
blackburnapply_to_example looks better23:32
@sonney2kbut for classifiers I think we should keep these functions23:32
@sonney2kblackburn, yeah sure but that is not the problem23:32
blackburnsonney2k: btw what about no apply_to*? apply(features) & apply(vector)?23:33
@sonney2kand apply(idx) ?23:34
blackburnsome kind of polymorphism :)23:35
blackburnwell, time to bed :) now really23:37
blackburnsee you23:37
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]23:37
--- Log closed Mon May 30 00:00:44 2011