Open in new window / Try shogun cloud
--- Log opened Mon Jul 29 00:00:12 2013
--- Day changed Mon Jul 29 2013
pickle27sonney2k: right, I don't see anything about RealFeatures in here though00:00
@sonney2k> library(modshogun)00:00
@sonney2kWelcome! This is SHOGUN version 2.2.000:00
@sonney2k> x=GaussianKernel()00:00
@sonney2k> x00:00
@sonney2kAn object of class "_p_shogun__CGaussianKernel"00:00
@sonney2kSlot "ref":00:00
@sonney2kError in slot(object, what) :00:00
@sonney2k  no slot of name "ref" for this object of class "_p_shogun__CGaussianKernel"00:00
pickle27just SGVector and SGMatrix00:00
@sonney2kthe question is what is this ref slot00:00
lisitsynsonney2k: I guess that's lispy name for method00:01
lisitsynthey follow some scheme-like thing inside R00:01
@sonney2klisitsyn, for what?00:01
lisitsynsonney2k: for ref()00:01
lisitsynmember function00:02
lisitsynwhatever00:02
@sonney2kwhat method do we need?00:02
lisitsynsonney2k: I guess it tries to call ref()00:02
@sonney2kpickle27, ahh sure typemaps are just SGVector/SGMatrix stuff00:03
@sonney2kpickle27, so whenever you have a function signature00:03
@sonney2kfoo(SGVector<double> x)00:03
@sonney2kyou can write foo(c(1,2,3)) from R00:03
@sonney2ksimilar for output00:03
@sonney2k(in and out typemaps!)00:03
@sonney2klisitsyn, IDK I tried some minimal R example where I used ref/unref and they *were* called00:04
@sonney2kno issues00:04
lisitsynsonney2k: no I was just guessing so everything can happen00:04
pickle27sonney2k: okay - so then where do I need to be looking for figuring out the RealFeatures problem00:05
@sonney2klisitsyn, bah slotNames(x)00:12
@sonney2k[1] "ref"00:12
@sonney2kwtf?00:12
@sonney2kpickle27, well lets start with some simple example first00:12
@sonney2kx=RealLabels(c(1.0,2.2))00:13
@sonney2kwhen this is called00:13
@sonney2kpickle27, go to line 36 in swig_typemaps.i00:13
@sonney2kyou will see a typecheck typemap00:13
@sonney2kand a in typemap00:14
@sonney2kfor SGVector types00:14
pickle27yeah Im there00:14
@sonney2kthe typecheck typemap well test if the type matches00:14
@sonney2kif it does it returns the equiv of true (1) otherwise 000:14
@sonney2kso my first guess is that this always returns 000:15
@sonney2kmaybe add some SG_SPRINT there to see the result00:16
@sonney2kpickle27, if the typecheck typemap returns 100:16
@sonney2kthe in typemap will be reached00:16
@sonney2kand that one then just has to assign an appropriate SGVector to $100:17
@sonney2kcurrently it doesn't  copy the memory and so the simple example (if it would pass the typecheck typemap) woudl just crash00:17
@sonney2ksince SGVector will at some point deallocate the memory when you exit R00:18
@sonney2kand R likely too00:18
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["Fallen asleep!"]00:19
pickle27sonney2k: so bascially print the result of line 4900:21
@sonney2kpickle27, questions?00:21
pickle27i mean 4000:21
pickle27(TYPEOF($input) == r_type && Rf_ncols($input)==1 )00:21
@sonney2kpickle27, I am just trying this with all fields00:21
-!- travis-ci [~travis-ci@ec2-54-243-23-104.compute-1.amazonaws.com] has joined #shogun00:22
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/958574800:22
-!- travis-ci [~travis-ci@ec2-54-243-23-104.compute-1.amazonaws.com] has left #shogun []00:22
@sonney2kpickle27, but then the big issue is to figure out why refcounting doesn't work00:22
@sonney2kpickle27, I am not even getting towards the typecheck00:23
@sonney2kso one now would have to debug the generated R code00:24
@sonney2kpickle27, hmm00:25
@sonney2kif I do00:25
@sonney2kx=RegressionLabels()00:25
@sonney2k> x$set_labels(c(1.0,2.3))00:25
@sonney2k> x$get_label(0)00:26
@sonney2k[1] 100:26
@sonney2k> x$get_label(1)00:26
@sonney2k[1] 2.300:26
@sonney2kit seems alright00:26
@sonney2kmaybe it `just' doesn't work in the constructor style syntax00:26
pickle27hmm00:27
@sonney2kyeah seems like00:28
@sonney2kwith RealFeatures I get the (expected) crasher when I do00:28
@sonney2k> x=RealFeatures()00:28
@sonney2k> x$set_features(matrix(c(1,2,3,4,5,6),2,3))00:28
pickle27sonney2k: I get this when I run that00:31
pickle27Error: evaluation nested too deeply: infinite recursion / options(expressions=)?00:31
@sonney2kpickle27, there are certainly more issues but maybe it is a start00:31
* sonney2k Zzzzz.00:31
pickle27wierd this time I got a backtrace00:31
pickle27ahg I hate when I accidently ctrl-l irc00:32
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]00:35
@iglesiasgok guys, good night!00:35
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun00:35
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]00:35
@sonney2kpickle27, memory corruption!00:43
pickle27sonney2k: I ran it again and I couldn't get another backtrace00:43
@sonney2kpickle27, you can fix it easily be copying the memory before gving it sgvector/amtrix00:43
pickle27but it did happen once00:43
shogun-buildbotbuild #1103 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1103  blamelist: Soeren Sonnenburg <sonne@debian.org>00:48
pickle27sonney2k: fix it in the typemap func?01:01
shogun-buildbotbuild #1420 of deb3 - modular_interfaces is complete: Failure [failed install test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1420  blamelist: Soeren Sonnenburg <sonne@debian.org>01:11
pickle27okay thats enough for me today, see you all tomorrow01:36
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]01:36
-!- nube [~rho@49.244.47.72] has joined #shogun02:37
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]02:55
-!- nube [~rho@49.244.47.72] has quit [Quit: Leaving.]03:29
shogun-buildbotbuild #472 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/47204:12
-!- nube [~rho@116.90.239.3] has joined #shogun05:25
-!- nube [~rho@116.90.239.3] has left #shogun []05:28
-!- lisitsyn [~lisitsyn@92-240-133-94.clients.tlt.100megabit.ru] has quit [Quit: Leaving.]08:06
gsomixgood morning08:37
-!- van51 [~van51@ppp-94-66-56-238.home.otenet.gr] has joined #shogun09:28
van51hello09:30
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun10:18
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun10:19
-!- mode/#shogun [+o iglesiasg] by ChanServ10:19
@iglesiasggood morning to all10:20
van51iglesiasg: good morning :)10:22
-!- van51 [~van51@ppp-94-66-56-238.home.otenet.gr] has quit [Quit: Leaving.]10:41
-!- van51 [~van51@94.66.56.238] has joined #shogun10:49
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has joined #shogun10:59
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun11:27
lisitsynvgorbati: hey haven't seen you for a while?11:28
lisitsynhow is it going? are you done with your thesis?11:28
-!- gsomix_ [~gsomix@95.67.137.114] has joined #shogun11:50
-!- gsomix [~gsomix@109.169.249.121] has quit [Ping timeout: 264 seconds]11:53
vgorbatilisitsyn: hello there) I basically rested all the time:) How is Berlin?12:01
vgorbatilisitsyn: do you mean master's thesis?12:02
lisitsynvgorbati: berlin is nice indeed :)12:03
lisitsynvgorbati: well some thesis you were working on12:03
vgorbatilisitsyn: I was working on monograph for my university supervisor (is it a correct word for it?), and yes, I believe it is done now:)12:05
-!- foulwall [~user@116.114.141.87] has joined #shogun13:15
-!- van51 [~van51@94.66.56.238] has quit [Quit: Leaving.]13:37
-!- foulwall [~user@116.114.141.87] has quit [Ping timeout: 240 seconds]13:48
-!- gsomix_ [~gsomix@95.67.137.114] has quit [Quit: Leaving]14:24
-!- foulwall` [~user@116.114.141.87] has joined #shogun14:58
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 256 seconds]15:58
@wikingmmmm16:33
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun16:43
-!- mode/#shogun [+o iglesiasg] by ChanServ16:43
-!- van51 [~van51@ppp-94-66-56-238.home.otenet.gr] has joined #shogun16:47
-!- foulwall` [~user@116.114.141.87] has quit [Ping timeout: 240 seconds]17:45
-!- foulwall` [~user@116.114.133.251] has joined #shogun17:46
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun17:54
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun18:04
shogun-notifier-shogun: Roman Votyakov :develop * c4b0bb6 / tests/unit/machine/gp/ (4 files): https://github.com/shogun-toolbox/shogun/commit/c4b0bb66f22d8895de7bf3a71554ad573c762ee218:04
shogun-notifier-shogun: add some unit-tests for likelihood models18:04
shogun-notifier-shogun: Roman Votyakov :develop * f8f1b75 / src/shogun/classifier/GaussianProcessBinaryClassification.cpp: https://github.com/shogun-toolbox/shogun/commit/f8f1b751b8e046d9bd9afe3b787abb8bb78763d018:04
shogun-notifier-shogun: replace eigen3 exp() call with shogun exp() call18:04
shogun-notifier-shogun: Heiko Strathmann :develop * b2727fc / / (5 files): https://github.com/shogun-toolbox/shogun/commit/b2727fc0b20e54d9b6ab6fb9ca556c759fa5b59c18:04
shogun-notifier-shogun: Merge pull request #1325 from votjakovr/feature/gp_refactoring18:04
shogun-notifier-shogun:18:04
shogun-notifier-shogun: Add some unit-tests for likelihood models18:04
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]18:07
shogun-buildbotbuild #1104 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1104  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com>18:09
@sonney2kfoulwall`, hey there - any news on the README?18:25
foulwall`Hey sonney2k, already finished, and now working on migration for tapkee demos.18:27
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]18:27
@sonney2kfoulwall`, please send a PR then18:27
foulwall`ok sonney2k18:28
van51heh I thought that the ` after foulwall` was dust on my screen :p18:29
@sonney2kvan51, heh!18:30
@sonney2kvan51, so how is it going now?18:30
van51sonney2k: well first of all, the CHashedDocDotFeatures and the converter have identical results18:31
@sonney2kvan51, btw regarding benoits email - we already use bitmasks right?18:31
van51sonney2k: in the doc features18:31
@sonney2kvan51, excellent!18:31
@sonney2kvan51, and speed?18:31
van51sonney2k: in dense/sparse I had switched to just mod18:31
van51sonney2k: for speed I'm not sure, it depends on the normalization18:32
van51sonney2k: I mean unnormalized dot features take a lot time18:32
@sonney2kvan51, well of course with same normalization18:32
van51sonney2k: but the sparsefeatures returned from the converter are not normalized18:32
@sonney2kvan51, oh you should add that to the converter too!18:33
van51sonney2k: I tried but it would be the same normalziation. In the dot features we normalize the result, while there we create the entire vector so we would normalize each feature18:33
van51wouldn't18:34
van51dot product result*18:34
@sonney2kvan51, parse error18:34
van51sonney2k: heh18:34
van51sonney2k: in the dot features we compute the dense dot for instance18:35
van51sonney2k: and we normalize it18:35
@sonney2kvan51, you jsut normalize each feature by dividing with 1/sqrt(count)18:35
@sonney2kvan51, sure but in dotfeatures add_to_dense_vec you do the same normalization per feature18:36
van51sonney2k: I tried it with 1/sqrt(doc.size()) and it was giving poor results18:37
van51sonney2k: maybe I missed something18:37
van51sonney2k: I can try it again though18:37
van51sonney2k: btw I wanted to ask you about the CCommUlongKernel with svmLight that you had told me18:37
van51sonney2k: I tried to compare their speed first, but when I used the kernel it segfaulted in a radix_sort method18:39
van51sonney2k: I'm guessing it ran out of memory18:39
@sonney2kvan51, maybe forgot that you need to use a double?18:40
@sonney2kvan51, it should give the exact same results18:40
@sonney2kvan51, hmmhh radix sort should work just fine (at least it did)18:41
van51sonney2k: I'll reproduce it now18:43
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun18:45
shogun-buildbotbuild #1421 of deb3 - modular_interfaces is complete: Failure [failed install test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1421  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com>18:46
-!- travis-ci [~travis-ci@ec2-54-226-172-147.compute-1.amazonaws.com] has joined #shogun18:53
travis-ci[travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/961036318:53
-!- travis-ci [~travis-ci@ec2-54-226-172-147.compute-1.amazonaws.com] has left #shogun []18:53
-!- foulwall [~user@116.114.133.251] has joined #shogun18:53
@sonney2kpickle27, yes18:55
@sonney2kpickle27, let me just add it18:55
-!- foulwall [~user@116.114.133.251] has quit [Remote host closed the connection]18:56
-!- foulwall` [~user@116.114.133.251] has quit [Remote host closed the connection]18:56
-!- foulwall [~user@116.114.133.251] has joined #shogun18:56
van51sonney2k:  this is my valgrind output18:56
van51http://pastebin.com/wcT9rLpj18:56
shogun-buildbotbuild #1105 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1105  blamelist: Roman Votyakov <votjakovr@gmail.com>18:58
@sonney2kvan51, hmmhh looks like we broke it :/19:01
van51sonney2k: let me show you my code in case there is something wrong there19:01
van51sonney2k: https://gist.github.com/van51/610583419:07
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.]19:13
shogun-buildbotbuild #1422 of deb3 - modular_interfaces is complete: Failure [failed install test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1422  blamelist: Roman Votyakov <votjakovr@gmail.com>19:21
votjakovrsonney2k: hi! i've found, that since doxygen 1.7.2 we can use MathJax instead of using prerendered bitmaps for rendering formulas. I think, that with MathJax formulas will be prettier. What do you think about using it?19:40
@sonney2kvotjakovr, sure!19:56
@sonney2kvotjakovr, that would speed up things too19:56
@sonney2kvotjakovr, only issue is that it won't detect errors in the file - I mean if any of us screws up a formula19:58
votjakovrsonney2k: hmm, yep, unfortunately20:00
shogun-notifier-shogun: Soeren Sonnenburg :develop * eccafc8 / src/ (4 files): https://github.com/shogun-toolbox/shogun/commit/eccafc84132f31fe5bb2e03372200482710f87a920:11
shogun-notifier-shogun: add get_copy function to lib/memory to be used from typemaps for clones20:11
@sonney2kpickle27, that fixes the crashers etc for me20:12
@sonney2kpickle27, I could do20:12
@sonney2kx=Regressionlabels()20:12
@sonney2kx$set_labels(c(1,2,3))20:12
@sonney2kx$set_labels(c(1,2))20:12
@sonney2k...20:12
@sonney2kand also20:12
@sonney2kx=RealFeatures()20:13
@sonney2kx$set_feature_matrix(matrix(c(1,2,3,4,5,6),2,3))20:13
@sonney2kvotjakovr, btw looks like doxygen also supports markdown syntax nowadays20:18
votjakovrsonney2k: oh, yeah, i see20:20
pickle27sonney2k: okay i will merge and see if it works for me soon20:21
-!- iglesiasg [~iglesias@2001:6b0:1:1da0:9ce1:9054:335:9d36] has joined #shogun20:21
-!- mode/#shogun [+o iglesiasg] by ChanServ20:21
@iglesiasghello!20:22
@sonney2kiglesiasg, helllo!20:22
-!- lisitsyn [~lisitsyn@213.87.133.71] has joined #shogun20:23
lisitsynkevin is a good boy today20:24
lisitsynpickle27, ;)20:24
@iglesiasglisitsyn,  I will become a good boy soo, I promise :P20:24
lisitsynvan51, votjakovr20:24
lisitsynhaha20:25
pickle27haha20:25
lisitsynfoulwall, hushell20:25
lisitsyndamn this phone typing is not so fast20:25
@sonney2kvan51, is the normalization working in the converter now?20:26
lisitsyniglesiasg, how is your marginal deals?20:27
foulwallHi lisitsyn20:27
lisitsynlargely marginal20:27
@iglesiasglisitsyn, LMNN?20:27
lisitsynfoulwall, hey I just called to say20:28
lisitsynweekly report ;)20:28
foulwallok lisitsyn :)20:28
lisitsyniglesiasg: ja this one20:28
lisitsyniglesiasg, is it better performance wise now?20:29
@iglesiasglisitsyn, so so. My implementation currently sucks a bit compared to the author's (and it is in Matlab, although with some parts in C)20:29
@iglesiasglisitsyn, no, not yet20:29
@iglesiasglisitsyn, I realized I have to change a sort of big thing now20:29
@iglesiasglisitsyn, well, not that big actually20:29
@iglesiasglisitsyn, you are with phone now, right?20:29
lisitsyniglesiasg: do you know the bottleneck?20:29
@iglesiasgotherwise I can show you20:30
-!- foulwall` [~user@116.114.133.251] has joined #shogun20:30
lisitsyniglesiasg: ueah, i can read things though20:30
lisitsynbut not compile shogun20:30
@iglesiasglisitsyn, haha indeed20:30
@iglesiasglet me show you then, on moment20:30
-!- foulwall` [~user@116.114.133.251] has quit [Client Quit]20:30
@iglesiasgone*20:30
@iglesiasglisitsyn, https://github.com/iglesias/shogun/blob/feature/lmnn/src/shogun/metric/LMNNImpl.cpp#L13520:31
@iglesiasglisitsyn, so I have that method right now that is computing the difference between every pair of feature vectors, and then computing the outer product of the diff vector with itself20:32
@sonney2kfoulwall, and where is the README PR ?20:32
@iglesiasglisitsyn, right now that computes N^2 matrices of D^2 elements each one20:33
@iglesiasglisitsyn, it is possible to make that smaller20:33
foulwallsonney2k: sending20:34
@iglesiasglisitsyn, since (x_i-x_j)*(x_i-x_j)' is the same as (x_j-x_i)*(x_j-x_i)'20:34
@iglesiasglisitsyn, but I think that even if I take that into account, it is still something too big to keep in memory (it will still be the order of N^2 matrices)20:35
lisitsyniglesiasg: yeah quite big thing20:36
@iglesiasglisitsyn, so I think that it is better to compute the outer products on demand; every time one is needed I compute it, even if that migh entail computing the same outer product more than once20:36
@iglesiasglisitsyn, what do you think of doing it on demand?20:37
@iglesiasglisitsyn, I think the author's implementation does it in this way20:37
lisitsyniglesiasg, would be slower I guess but takes less memory20:37
lisitsynif you dont have any N2 matrices more it totally makes sense20:38
@iglesiasglisitsyn, then the bottleneck is this method https://github.com/iglesias/shogun/blob/feature/lmnn/src/shogun/metric/LMNNImpl.cpp#L31620:38
lisitsynare there any more N2 matrices?20:39
@sonney2kiglesiasg, sounds like kernel caching20:39
@iglesiasgsonney2k, yeah, I thought that caching could help20:39
@iglesiasgso I think that for the moment I will modify the code to compute outer products on demand, and later we can see if it pays off to have caching20:40
@iglesiasglisitsyn, what do you mean if there are more N2 matrices?20:40
lisitsyniglesiasg: I think caching makes sense here20:40
lisitsyniglesiasg: if thats the only NxN matrix here20:41
@iglesiasglisitsyn, aah ok got it. Yes it is20:41
@iglesiasglisitsyn, but note that it is not only a NxN matrix20:41
lisitsynthen kill it yeah20:41
@iglesiasglisitsyn, it is NxN matrices of DxD elements each (N=number of vectors, D=feature dimension)20:42
lisitsynoh20:42
lisitsynheh20:42
@iglesiasgbecause you compute the outer product of the difference of feature vectors20:43
@iglesiasgfor each pair of feature vectors, you make its difference, and compute the outer product of the difference with itself20:43
lisitsynyeah thats not something to be in memory20:43
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has joined #shogun20:44
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/961508320:44
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has left #shogun []20:44
lisitsynby outer product you mean dger from blas20:44
lisitsyneight?20:44
lisitsynright?20:44
@iglesiasglisitsyn, no idea :D20:44
lisitsyn:D20:44
@iglesiasglisitsyn, there are two ways you can multiply a vector with itself20:45
lisitsynyeah heard sth20:45
@iglesiasglisitsyn, inner product -> you get an scalar (something like x'*x)20:45
lisitsynthis one produces matrix20:45
@iglesiasgouter product -> you get a matrix (x*x')20:45
@iglesiasgexactly20:45
lisitsyndger :p20:46
@iglesiasglisitsyn, I am checking dger20:46
lisitsynddot and dger20:46
@iglesiasglisitsyn, yeah it seems to be like that20:46
@iglesiasglisitsyn, although they same in the doc that it is x*y' where x has m elements and n elements20:46
lisitsyniglesiasg, eigen has symmetric rank update20:46
lisitsyni used that thing in tapkees pca20:47
@iglesiasglisitsyn, symmetric rank update? Do you think that could help me to make it faster?20:47
@iglesiasglisitsyn, right now I get two eigen vectors, make their difference, and then the outer product20:47
@iglesiasglisitsyn, https://github.com/iglesias/shogun/blob/feature/lmnn/src/shogun/metric/LMNNImpl.cpp#L15420:48
@iglesiasgthat line and the next one20:48
lisitsynno probably wont help here20:48
@iglesiasglisitsyn, aham. Why do you think so?20:49
@iglesiasgjust trying to understand it better :)20:49
lisitsyniirc that rank update is better for multiple calls20:49
@iglesiasglisitsyn, I guess it is a matter of testing it anyway, so I will keep in mind that there is another way of getting the outer product20:50
lisitsyniglesiasg: it employs symmetricity better somehow20:51
-!- lisitsyn [~lisitsyn@213.87.133.71] has quit [Remote host closed the connection]20:52
-!- lisitsyn [~lisitsyn@213.87.133.71] has joined #shogun20:52
lisitsynoh I get disconnected once turning my phone off20:53
-!- lisitsyn [~lisitsyn@213.87.133.71] has left #shogun []20:57
shogun-buildbotbuild #1106 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1106  blamelist: Soeren Sonnenburg <sonne@debian.org>21:03
foulwallls21:05
@iglesiasgDocuments Desktop Dropbox21:06
@iglesiasgfoulwall, :P21:06
foulwallhaha iglesiasg21:06
@sonney2kvan51, minimalchange: https://gist.github.com/sonney2k/610678021:06
pickle27hahaha I do that too21:06
pickle27except I have this thing where I like to clear my terminals using ctrl-l all the time21:06
van51sonney2k: hey sorry I was helping my grandmother with something21:06
pickle27and then I clear IRC and its annoying21:06
@iglesiasghehe21:07
van51sonney2k: what did you change?21:07
van51the 8-1?21:07
@sonney2kvan51, yeah21:07
@sonney2klike in the python example21:07
van51sonney2k: what is the meaning of start then?21:07
@sonney2kvan51, where to start21:08
van51sonney2k: :p21:08
@sonney2kvan51, well this is doing some bitheavy stuff21:08
@sonney2kand basically it goes back 8 chars from the start so starting at index 7 is what you want to get 7...021:10
van51sonney2k: ah I see21:11
van51sonney2k: well I'm going to get some food and work on some stuff later and overnight21:11
van51sonney2k: btw I've also started the document example21:11
van51sonney2k: and I was thinking of making a simple web ui, just to be able to write some lines and test it21:12
@sonney2kvan51, use the python_modular interface instead and some simple input or so and foulwall can later nicely integrate this21:13
van51sonney2k: ok21:14
van51sonney2k: so I'm off for about an hour21:15
van51cu later21:15
@sonney2kvan51, btw it might be good to do that for your example too. prototyping with python really is much faster (and debugging too!)21:17
van51sonney2k: I don't know python so I'm trying to avoid it21:17
van51sonney2k: but I guess it's a good time to get to know as well :021:18
van51:)*21:18
-!- van51 [~van51@ppp-94-66-56-238.home.otenet.gr] has quit [Quit: Leaving.]21:21
pickle27sonney2k: I tried the new R modular and I ran those samples you posted and they work for me21:23
pickle27sonney2k: it still seams like none of the methods are "binding" though21:23
pickle27whenever I try and use a method I get a not found error21:23
@sonney2kpickle27, that must be some swig issue - we have ton's of constructors21:23
pickle27yeah21:23
@sonney2kso figuring out which one to use is the task of swig21:23
shogun-notifier-shogun: Soeren Sonnenburg :develop * 95def24 / src/shogun/machine/gp/InferenceMethod.h: https://github.com/shogun-toolbox/shogun/commit/95def246a179d7a0065c85cd1252e2f6cae136ef21:26
shogun-notifier-shogun: fix typo in doxygen formula21:26
shogun-notifier-shogun: Soeren Sonnenburg :develop * 3f113b7 / / (16 files): https://github.com/shogun-toolbox/shogun/commit/3f113b7a74d2a0ad54a9a484e93fca6ba9afde4f21:26
shogun-notifier-shogun: add some get_strdup function to avoid compile errors (strdup is not ansi C!)21:26
@sonney2kpickle27, back then I totally understood how that R stuff worked but I forgot21:26
@sonney2kpickle27, the only thing I remember is that from the R code there are some tests run selecting the appropriate constructor21:27
pickle27sonney2k: and I have no idea how it works21:27
pickle27it sounds like its pretty tough to fix - maybe I'll just push an r_static example for now21:28
-!- hushell [~hushell@c-24-21-169-136.hsd1.or.comcast.net] has quit [Ping timeout: 276 seconds]21:29
-!- foulwall` [~user@116.114.133.251] has joined #shogun21:31
-!- foulwall [~user@116.114.133.251] has quit [Ping timeout: 246 seconds]21:32
-!- sonney2k [~shogun@7nn.de] has quit [Ping timeout: 246 seconds]21:33
-!- sonney2k_ [~shogun@7nn.de] has joined #shogun21:33
-!- sonney2k_ is now known as sonney2k21:34
-!- mode/#shogun [+o sonney2k] by ChanServ21:35
@sonney2kpickle27, well you have to read swig code and the generated wrapper code modshogun.R etc21:35
@sonney2knot fun21:35
pickle27sonney2k: doesn't sound like it!21:35
@sonney2kiglesiasg, any progress on the failing tests on the buildbot?21:36
@iglesiasgsonney2k, unfortunately not, sorry. I have focused on LMNN again since yesterday's eve21:36
@sonney2kfoulwall`, ohh please don't put images in the repository21:38
@sonney2kfoulwall`, please put them again under shogun-data/tapkee21:39
foulwall`ok sonney2k , I'll resend21:39
@sonney2kfoulwall`, it might make sense to not have >1000 files too btw21:40
@sonney2kfoulwall`, maybe better use a single .zip file with the images in there21:40
@sonney2klike faces.zip21:40
@sonney2kand mnist.zip21:40
foulwall`I see21:41
@sonney2kfoulwall`, btw do you have a live demo already for that?21:41
foulwall`let me upload it on nn.7nn.de21:41
foulwall`sonney2k: check this, http://nn.7nn.de:8000/tapkee/entrance .21:46
@sonney2kfoulwall`, nice!21:47
shogun-buildbotbuild #1107 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1107  blamelist: Soeren Sonnenburg <sonne@debian.org>21:55
shogun-buildbotbuild #1108 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1108  blamelist: Soeren Sonnenburg <sonne@debian.org>22:01
@iglesiasgfoulwall`, looking cool!22:01
@iglesiasglisitsyn's gf will become worldwide famous eventually thanks to the faces embedding :)22:02
@sonney2kiglesiasg, yeah the new lena :)22:02
@iglesiasgsonney2k, haha yeah22:02
@iglesiasgsonney2k, have you actually seen the original of the Lena's picture?22:03
foulwall`haha :) iglesiasg, I've seen.22:03
@iglesiasgit is, let's say, not precisely the one they show in class or in college textbooks :P22:03
foulwall`sonney2k: thanks:D22:04
@iglesiasgLena is Swedish btw :)22:04
@iglesiasgpoeple are pretty proud in here when they use the picture in class22:04
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has joined #shogun22:11
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/961821122:11
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has left #shogun []22:11
-!- gsomix [~gsomix@95.67.137.114] has joined #shogun22:11
gsomixgood evening22:11
gsomixreport time? :)22:11
@sonney2kgsomix, where is your PR?22:12
gsomixsonney2k, with quoting and libsvm reader? not ready yet, I need little time for it.22:14
@sonney2kgsomix, no the one that you submitted ~21hrs ago22:15
@sonney2kwhere I wanted some tiny changes22:15
shogun-notifier-shogun: Soeren Sonnenburg :develop * a6f59dd / src/configure,src/shogun/io/Parser.cpp: https://github.com/shogun-toolbox/shogun/commit/a6f59dd8e159a0945b75eb2aca45c1e6a11e24b322:16
shogun-notifier-shogun: use strtod if strtold is not available22:16
gsomixsonney2k, ah, just a moment!22:16
gsomixsonney2k, forgot it. =___= today was little holiday, sorry.22:17
gsomixsonney2k, btw I haven't any warnings related to format on my machine.22:17
@sonney2kgsomix, I need to merge yours before I can start to convert examples22:17
@sonney2kgsomix, yeah sure 32bit vs 64bit22:17
@sonney2kgsomix, so use the 'standard' ones and we will be fine22:18
gsomixsonney2k, aha, indeed.22:18
gsomixsonney2k, btw, can I help with examples&22:18
gsomix*?22:18
@sonney2kiglesiasg, wiking, votjakovr, pickle27 anyone here up to doing a doc sprint at google in october?22:19
@iglesiasgsonney2k, sure!22:19
@sonney2kgsomix, well sure but we are quite a bit behind schedule :/22:19
pickle27probably no for me because I'm hoping to be doing some travelling in Oct22:20
@sonney2kit is during oct14-18 http://www.booksprints.net/2013/07/2013-doc-camp-announced/22:20
@iglesiasgsonney2k, what should we do to raise our odds of getting in?22:21
@sonney2kiglesiasg, IDK22:21
@sonney2kwe certainly need 4 people that can do a stunt22:22
pickle27sonney2k: looks very cool though, I'd be happy to help I just won22:22
@iglesiasgsonney2k, would you be in?22:22
pickle27't be able to be there22:22
@sonney2kiglesiasg, HeikoS, me + ??22:22
@iglesiasglet's see if wiking and lisitsyn could make it22:22
@sonney2kI am not sure I would go ipython-notebook for that22:23
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has joined #shogun22:23
@iglesiasgsonney2k, mmm I don't see why not22:23
@sonney2kiglesiasg, they aim for a real book22:24
@iglesiasgsonney2k, people are doing real books these days based on ipython I think22:24
@sonney2kcopy & pasting parts from people's thesis would not be that easy then22:25
@iglesiasgsonney2k, lol22:25
@sonney2kiglesiasg, I think that is our only chance to *finish* a book in 4 days22:25
@sonney2kthat is what they aim fo22:25
@sonney2kare22:25
@sonney2kiglesiasg, look at this http://www.booksprints.net/about/22:26
@iglesiasgsonney2k, check thhis one: http://camdavidsonpilon.github.io/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers/22:26
@iglesiasgsonney2k, http://nbviewer.ipython.org/urls/raw.github.com/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers/master/Chapter1_Introduction/Chapter1_Introduction.ipynb22:26
@iglesiasgchapter 1 ^22:27
@iglesiasgto me that looks like a nice real book22:27
@iglesiasgsonney2k, what do you think?22:28
@sonney2kiglesiasg, looks *very* cool22:29
@iglesiasgindeed, we agree :)22:29
@sonney2kiglesiasg, can this be converted to some epub /pdf etc?22:30
@sonney2kor is it 'just' web?22:30
@iglesiasgsonney2k, no idea, honestly22:30
-!- van51 [~van51@ppp-94-66-56-238.home.otenet.gr] has joined #shogun22:30
@iglesiasglet me have a quick look22:30
@sonney2kvan51, hey again22:31
van51sonney2k: hi22:31
@iglesiasgthey say so in the page, no idea how it looks like, let's see22:31
@sonney2kiglesiasg, ok then it is pretty clear how to increase our chances - we should have some ideas what our target audience is, structure, running examples and methods we want to use22:32
@sonney2kiglesiasg, but issue still remains that we are just a very small org (compared to e.g. KDE)22:32
@iglesiasgsonney2k, when is the deadline?22:32
@sonney2kiglesiasg, in 3 days22:32
@iglesiasgsonney2k, I am thinking, if it is after GSoC ends22:32
@iglesiasgshit...22:32
@sonney2kor 5?22:32
@sonney2kiglesiasg, last year it was like this22:32
@iglesiasgI drop the idea then22:33
@sonney2kbefore august 7!22:33
@sonney2kif heiko is not around for discussion yeah then indeed22:33
@iglesiasgsonney2k, https://dl.dropboxusercontent.com/u/11020840/June82013.pdf22:34
@iglesiasgsonney2k, pdf version of the book22:34
@iglesiasgsonney2k, IMHO it looks nicer in the browser, but it is still cool I think22:34
@sonney2kiglesiasg, I think we have a chance this year22:35
@sonney2kiglesiasg, jsut because big data etc is such a big hype22:35
gsomixsonney2k, compiling...22:35
@sonney2kso we should try!22:35
@iglesiasgsonney2k, yep!22:35
@sonney2kiglesiasg, very nice still!!!22:36
@iglesiasgso I vote sticking to ipython-notebook :)22:36
shogun-buildbotbuild #1424 of deb3 - modular_interfaces is complete: Failure [failed compile octave_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1424  blamelist: Soeren Sonnenburg <sonne@debian.org>22:42
@sonney2kiglesiasg, sure then we just have to figure out how to integrate big chunks of latex22:44
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 268 seconds]22:45
@iglesiasgsonney2k, yes22:45
@sonney2kiglesiasg, no idea whether stuff like toc / index / references are possible22:47
gsomixsonney2k, testing...22:49
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["Fallen asleep!"]22:49
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun' --branch=develop22:51
shogun-buildbotbuild forced [ETA 17m23s]22:51
shogun-buildbotI'll give a shout when the build finishes22:51
shogun-notifier-shogun: Soeren Sonnenburg :develop * e6788a0 / src/interfaces/ (2 files): https://github.com/shogun-toolbox/shogun/commit/e6788a07fdb6845a38f92bc55489e80f1c33f7c122:53
shogun-notifier-shogun: drop get_copy from octave and perl22:53
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has joined #shogun22:54
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/962030222:54
-!- travis-ci [~travis-ci@ec2-72-44-42-216.compute-1.amazonaws.com] has left #shogun []22:54
gsomixsonney2k, done. https://github.com/shogun-toolbox/shogun/pull/132322:59
gsomixcompiling takes many time22:59
shogun-notifier-shogun: Evgeniy Andreev :develop * 5798840 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/5798840ae3254defe789403e3eac001cc665be3123:01
shogun-notifier-shogun: fixed invalid read in Parser, CSVFile constructor is shortened23:01
shogun-notifier-shogun: Soeren Sonnenburg :develop * f0ccda8 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/f0ccda88694ea364c7e1b021c7dc2256648b001b23:01
shogun-notifier-shogun: Merge pull request #1323 from gsomix/feature/parser23:01
shogun-notifier-shogun:23:01
shogun-notifier-shogun: Fixed invalid read in Parser, CSVFile constructor is shortened23:01
gsomixsonney2k, so what to do with examples?23:02
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]23:02
@sonney2kgsomix, maybe we should talk how to continue in general?23:02
@sonney2kgsomix, it would be nice to have the other ascii formats working23:03
gsomixsonney2k, yep.23:03
@sonney2kgsomix, and the stringfeatures/sparsefeatures/densefeatures/matrixfeatures etc cleaned up / supported23:03
gsomixsonney2k, so I'm working on libsvm file format now.23:03
@sonney2kI mean this is the first and most important thing to finish23:04
@sonney2kgsomix, sounds like I should rather convert the examples23:04
@sonney2kwhile you continue your work23:04
@sonney2kgsomix, afterwards I would like us to have some binary format based on protobuf and well there still is reading / writing matlab files23:05
@sonney2kwhich likely is most important for users23:05
shogun-buildbotbuild #1109 of cyg1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/110923:06
gsomixsonney2k, aha23:06
@sonney2kgsomix, but seriously I think most people will just use ascii and matlab format23:07
gsomixsonney2k, other ascii formats. which?23:07
@iglesiasgmatlab format please!23:07
@iglesiasggsomix, when are we getting this? Really looking forward to it23:08
gsomixsonney2k, matlab7 file format is last version, right?23:09
@sonney2kgsomix, arff then van51 would say vw23:09
@sonney2kgsomix, no23:10
@sonney2kgsomix, matlab has a couple of formats23:10
gsomixsonney2k, not compatible?23:10
@sonney2kgsomix, as I wrote on the ideas page octave supports reading IIRC up to 7.223:10
@sonney2kafterwards they switched to sth hdf5 based23:10
@sonney2kso you it would take some time23:11
@sonney2kto do this23:11
gsomixsonney2k, I should look into octave source? or there is specs on 7.2?23:11
gsomixI found only 7.0.123:11
@sonney2kgsomix, http://www.mathworks.com/help/pdf_doc/matlab/matfile_format.pdf23:12
gsomixsonney2k, huh! I never imagined that I could find it on mathworks site.23:14
gsomixI mean MATLAB very closed and have not many docs about how it works inside.23:15
gsomix*is very23:15
@sonney2kgsomix, it seems not to be the latest format for sure23:15
shogun-buildbotbuild #1110 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1110  blamelist: Soeren Sonnenburg <sonne@debian.org>23:15
@sonney2kgsomix, maybe you can use octave's reader somehow23:16
@sonney2kgsomix, only issue is we cannot test this23:16
@sonney2kwe have no matlab23:16
@sonney2kso maybe we just get some .mat file from sb.23:16
@sonney2kand also the ascii representation or sth23:16
@sonney2kand then you can read/compare23:16
gsomixsonney2k, ok, I can test it on my local machine.23:17
gsomixand I can make .mat with ascii representation too23:17
gsomixI have licensed matlab 2013a at lab.23:18
gsomixsonney2k, I'm not full clear with protobuf now.23:19
gsomixsonney2k, we plan to develop our file format based on protobuf? or so?23:20
shogun-notifier-shogun: Soeren Sonnenburg :develop * 5733a65 / src/shogun/mathematics/Math.h: https://github.com/shogun-toolbox/shogun/commit/5733a65137c48909c95272fbbc6df57af5a1bb6723:21
shogun-notifier-shogun: define _USE_MATH_DEFINES to get non-standard constant M_PI23:21
@sonney2kgsomix, yes it is basically just a binary format / fast & size efficient to store values23:23
@iglesiasghave you guys started doing the mid-term evaluation?23:23
* sonney2k gets hungry23:24
* sonney2k off -> food23:24
gsomixiglesiasg, nope. I have many work that I promised to finish before midterm.23:24
@iglesiasggsomix, you will make on time! :)23:25
shogun-buildbotbuild #1423 of deb3 - modular_interfaces is complete: Failure [failed compile octave_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1423  blamelist: Soeren Sonnenburg <sonne@debian.org>23:25
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun' --branch=develop23:28
shogun-buildbotbuild forced [ETA 16m24s]23:28
shogun-buildbotI'll give a shout when the build finishes23:28
shogun-buildbotbuild #1111 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/111123:28
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun' --branch=develop23:29
shogun-buildbotbuild forced [ETA 16m24s]23:29
shogun-buildbotI'll give a shout when the build finishes23:29
shogun-buildbotbuild #1112 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/111223:30
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun' --branch=develop23:31
shogun-buildbotbuild forced [ETA 16m24s]23:31
shogun-buildbotI'll give a shout when the build finishes23:31
-!- travis-ci [~travis-ci@ec2-54-226-172-147.compute-1.amazonaws.com] has joined #shogun23:36
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/962166523:36
-!- travis-ci [~travis-ci@ec2-54-226-172-147.compute-1.amazonaws.com] has left #shogun []23:36
@sonney2kiglesiasg, I guess failing unit tests will be the only thing to show why matrixlist is unhappy23:38
@iglesiasgsonney2k, yeah, maybe. But valgrinds seems to complain about stuff going on when unref-ing23:39
@sonney2kiglesiasg, yeah exactly23:40
shogun-buildbotbuild #1425 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/1425  blamelist: Soeren Sonnenburg <sonne@debian.org>23:41
@iglesiasgsonney2k, that confuses me though23:41
@sonney2kiglesiasg, why?23:41
@iglesiasgsonney2k, where can the bug be if the error arises in the destructor or when un-referencing?23:41
@sonney2ksome kind of double free23:41
shogun-buildbotbuild #1113 of cyg1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/111323:47
@iglesiasgevalution submitted23:49
--- Log closed Tue Jul 30 00:00:53 2013