Open in new window / Try shogun cloud
--- Log opened Wed Apr 24 00:00:28 2013
@iglesiasgwiking: around?00:13
-!- shogun-notifier- [] has quit [Quit: transmission timeout]00:25
-!- van511 [] has joined #shogun00:50
-!- van51 [] has quit [Ping timeout: 276 seconds]00:51
-!- van511 [] has left #shogun []00:59
-!- van511 [] has joined #shogun01:00
-!- kevin [820f1060@gateway/web/freenode/ip.] has joined #shogun01:23
-!- kevin is now known as Guest8773801:24
-!- Guest87738 is now known as pickle2701:24
-!- van511 [] has left #shogun []01:25
-!- vikram360 [~vikram360@] has quit [Ping timeout: 240 seconds]02:53
-!- iglesiasg [] has quit [Quit: leaving]03:06
-!- sids_aquarius [] has quit [Remote host closed the connection]03:21
-!- sids_aquarius [] has joined #shogun03:22
sijin@lisitsyn I see your submit, thx.03:55
shogun-buildbotbuild #371 of nightly_default is complete: Failure [failed test]  Build details are at
-!- pickle27 [820f1060@gateway/web/freenode/ip.] has quit [Quit: Page closed]04:36
-!- votjakovr [] has joined #shogun05:13
-!- rgirish28 [~rgirish28@] has quit [Ping timeout: 260 seconds]05:48
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun06:14
-!- rgirish28 [~rgirish28@] has joined #shogun06:17
-!- rgirish28 [~rgirish28@] has quit [Ping timeout: 255 seconds]06:31
-!- rgirish28 [~rgirish28@] has joined #shogun06:36
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun06:44
-!- votjakovr [] has quit [Quit: ChatZilla 0.9.90 [Firefox 20.0.1/20130409194949]]07:26
-!- rgirish28 [~rgirish28@] has quit [Read error: Connection reset by peer]08:11
-!- rgirish28_ [~rgirish28@] has joined #shogun08:11
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Quit: Page closed]08:16
-!- jiuding [dbe09f12@gateway/web/freenode/ip.] has joined #shogun08:36
-!- ishan [~ishan@] has joined #shogun08:58
-!- ishan [~ishan@] has quit [Ping timeout: 268 seconds]09:09
-!- vgorbati [4d2f8429@gateway/web/freenode/ip.] has joined #shogun09:19
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:33
-!- van51 [] has joined #shogun09:36
-!- van51 [] has quit [Client Quit]09:37
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun09:38
-!- van51 [] has joined #shogun09:38
vgorbatilisitsyn: hello, are you here?09:48
@lisitsynvgorbati: yes but will leave in a minute09:48
vgorbatilisitsyn: when do you return, if you do?09:49
@lisitsynvgorbati: ~20 minutes or 4 bus stops09:49
vgorbatilisitsyn: I wlll catch you then, if you don't mind09:50
@lisitsynyeah sure09:50
-!- rgirish28_ [~rgirish28@] has quit [Ping timeout: 255 seconds]09:51
-!- lisitsyn [~blackburn@] has quit [Quit: Leaving.]09:54
-!- vgorbati [4d2f8429@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:58
-!- vgorbati [2535754b@gateway/web/freenode/ip.] has joined #shogun10:14
-!- lisitsyn [] has joined #shogun10:27
-!- mode/#shogun [+o lisitsyn] by ChanServ10:27
@lisitsynvgorbati: re10:27
vgorbatilisitsyn: well, I tried to think of good data to test LLE, and one of tries was a straight 3D line. The results were kind a weird - (as usual, initial - left, embedded - right).10:29
vgorbatinot that link10:29
@lisitsynhaha the first link is kinda strange! :D10:30
vgorbatilisitsyn: yeah, I just randomly clicked at first suggestion when was looking my vid)10:31
vgorbatilisitsyn: what do you think about the second link?)10:31
@lisitsynvgorbati: well it doesn't look wrong for me10:32
vgorbatilisitsyn: could you please explain?10:32
@lisitsynand actually well with k=2 distances are preserved10:32
@lisitsynaren't they?10:32
@lisitsynnot distances10:32
vgorbatilisitsyn: not totally10:33
vgorbatilisitsyn: minimal similarity is 16/2010:33
vgorbatilisitsyn: I though dimensionality reduction should capture some dependency in data, and it does not look like the dependency was captured in that case10:34
@lisitsynvgorbati: PCA would put your line to one axis10:35
@lisitsynvgorbati: well LLE captures it10:35
@lisitsynI wouldn't say it lost the dependency10:35
-!- HeikoS [] has joined #shogun10:36
-!- mode/#shogun [+o HeikoS] by ChanServ10:36
@HeikoSlisitsyn: MultidimensionaScalingTest.distance_preserving fails10:36
@lisitsynHeikoS: yeah I have seen10:36
@lisitsynsorry forgot to check that10:36
@lisitsynwill fix tonight10:36
vgorbatilisitsyn: about 'with k=2 neighbors are preserved' - but the k was set to 20..10:36
-!- iglesiasg [] has joined #shogun10:36
-!- mode/#shogun [+o iglesiasg] by ChanServ10:36
@lisitsynvgorbati: yes I mean it doesn't break neighborhoods that much10:37
-!- foulwall [] has joined #shogun10:38
vgorbatilisitsyn: but what then should I test?:) I can try to test that when k==20 that at least 2-3 neighbors are preserved)10:38
-!- votjakovr [] has joined #shogun10:42
vgorbatilisitsyn: ok, I just tested for k==5 on the line with small noise added, it preserves pretty much all the neighbors (sometimes 1 misses)10:43
-!- jiuding [dbe09f12@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]10:44
vgorbatilisitsyn: is this ok for the fix?10:44
deerishihi everyone.i want to discuss about the gsoc proposal.i want to implement speech recognition algorithims in the shogun toolbox10:46
@iglesiasgdeerishi: aham you were thinking of it as a GSoC project10:47
@HeikoSdeerishi: hi, is this on our ideas page?10:47
@iglesiasgdeerishi: I see it difficult to find a mentor for that at this moment -- too late already I think10:47
@lisitsynvgorbati: yes10:48
deerishi@HeikoS: this is not on the ideas page.10:48
@HeikoSdeerishi: I agree with iglesiasg, you need a mentor for this10:48
sijindoes anybody know why CDenseFeatures->get_num_features is not const method?10:49
@HeikoSsijin: this probably has no reason :)10:49
@iglesiasghehe indeed10:49
@HeikoSI tried to change this once and gives many many follow up compile errors :D10:49
@HeikoSbut feel free to change :)10:49
deerishidoes shogun has any speech recognotion projects implemented?10:50
@lisitsynHeikoS: if it was not virtual it would make some great sense10:50
sijinBut when i pass CFeatures* with const, I can not use this method10:50
@lisitsynsijin: why to pass pointer with const?10:50
@lisitsynCFeatures* const?10:50
@lisitsynor const CFeatures*?10:50
sijinyes I pass it as Const CFeatures*10:51
@lisitsynwhat is the advantage?10:51
deerishi@iglesiasg : can speech recognition be a gsoc proposal?i mean is it feasible?10:52
@iglesiasgsijin: I thought you meant CDenseFeatures::get_num_features() const10:52
@iglesiasgdeerishi: I think it is late to find a mentor for that10:52
vgorbatilisitsyn: btw, I just tried to change to Isomap, and it preserves all neighbors - i even tried manipulating n_neighbors, but still - all neighbors are preserved) Maybe let's test Isomap for now?10:52
@lisitsynvgorbati: hah! yes isomap should preserve it totally10:52
@lisitsynfeel free to transform it actually10:53
@lisitsynI don't mind10:53
sijin@lisitsyn I am writing function to wrap Projectingfunction10:53
vgorbatilisitsyn: I mean, there's no test for Isomap right now, isn't it?10:53
@lisitsynvgorbati: yes10:53
@lisitsynas you can see10:53
@lisitsynsijin: what should that function do?10:53
sijinAnd the projection operator receive const Densevector10:53
@HeikoSdeerishi: its all about the mentor, and we currently dont have one. We have HMM implemented but not specially for speec recognition. If you don't know a mentor this project is unfeasible, yes10:54
sijin@lisitsyn Since I want to define a shogun::projectionfunction, so that I can use tapkee with tapkee_shogun.hpp included10:54
sijinor I will need to include all the definition in tapkee10:55
deerishiis it that late!!!! @HeikoS ,@iglesiasg: basically i am interested in ML,NLP and signal processing.i want to give a proposal that somehow combines all these elements10:55
vgorbatilisitsyn: ok, then I will transform:) And I have another question - I was looking at tapkee's embedtDistributedStochasticNeighborEmbedding method, and there's a check for perplexity: 0<=perplexity<=(n_vectors-1)/3.0 + 1e-6. I can't understand the reasons for such ranges10:56
@lisitsynvgorbati: 1e-6 or N-1/3?10:56
vgorbati N-1/310:56
@iglesiasgdeerishi: the deadline for student proposals is next week10:56
@iglesiasgyou would need10:56
@lisitsynvgorbati: it comes from the original implementation of t-sne10:56
@lisitsynwhich I integrated to my lib10:57
@iglesiasgdeerishi: 1) Find someone who can mentor this project and want to be gsoc mentor for shogun.10:57
@iglesiasgdeerishi: 2) describe the project similar to the others that appear in the list10:57
@iglesiasgdeerishi: 3) prepare your application10:57
@iglesiasgdeerishi: that's roughly all the process. Providing that, yeah I think it is a bit too late :)10:58
@iglesiasgyou may still try if you wish of course, for my part there's no inconvenient10:58
sijin@lisitsyn in my current implementation, shogun::ProjectingFunction has only one member, which is a pointer of tapkee::ProjectingFunciton.10:59
-!- foulwall [] has quit [Remote host closed the connection]11:01
deerishi@iglesiag: is  Andreas Ziehe a mentor this time?11:03
deerishihe worked on ICA last time11:03
-!- gsomix [] has joined #shogun11:03
deerishii wanna go along the lines of ICA(since it is used in signal processing)11:04
@iglesiasgdeerishi: then it seems reasonable to apply for the ICA project, don't you think so?11:05
deerishiyes : ICA is very useful in signal processing,like the cocktail party problem.11:06
sijin@lisitsyn that function will operate on (CFeatures*) and output the projected (CFeatures*)11:08
-!- Yanglittle [deb20af4@gateway/web/freenode/ip.] has joined #shogun11:15
@iglesiasgwiking: around?11:17
Yanglittlehey, i run the classifier_multiclaslogisticregression from python_modular, it says "can't import  name multiclasslogisticregression" at line 5.11:29
@lisitsynYanglittle: do you have eigen3 installed?11:30
Yanglittleoh, not yet.11:30
@lisitsynwill work once you install and recompile11:31
-!- iglesiasg [] has quit [Quit: leaving]11:32
-!- nando [] has joined #shogun11:35
deerishi@HeikoS : i want to give my gsoc proposal on Andreas Ziehe still a mentor for this year's gsoc?11:38
@lisitsyndeerishi: please check the ideas list11:39
@HeikoSdeerishi: ^11:41
deerishi@lisitsyn: i checked it.i meant is he on irc.i had sent him a mail a while ago,he did not reply11:41
@lisitsynthat happens, he will answer11:42
@HeikoSdeerishi: he should be still in, might be busy, maybe try again.11:42
@HeikoSdeershi, make sure not to just ask him to tell you what to do, but rather try to propose something11:44
deerishi@HeikoS : k.thanks.11:50
deerishiwill start working on it right away11:50
@HeikoSsome initial code specifically for your project will be very useful11:50
deerishicould you elaborate on that.i mean  i have done some coding work on ICA before.11:51
@HeikoSdeerishi: a simple thing that relates to the project would already be enough11:54
@HeikoSor a preliminary plan on how you want to implement it11:54
@HeikoSbest to discuss with the mentor11:55
-!- erlenda [] has joined #shogun11:58
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun11:59
-!- vgorbati [2535754b@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]11:59
sijin@lisitsyn the Projectingfunction.implentation is still 012:21
-!- abinash [75ef5e6e@gateway/web/freenode/ip.] has joined #shogun12:33
-!- makokal [~bokal@] has joined #shogun12:37
@lisitsynsijin: still? what did you expect?12:41
-!- makokal [~bokal@] has quit [Quit: makokal]12:46
@lisitsynI didn't add a projection function for kpca yet ;)12:47
saketlisitsyn: there?12:55
@lisitsynsaket: yes13:01
saketHi in the KNN code, if you don't pass testfeatures while apply_multiclass is using the original train data on;ly!13:02
saketinstead of distance->rhs13:02
saketI tested it by printing SG_PRINT(num_lab) and also results->get_num_labels()13:04
saketCMulticlassLabels* result = knn->apply_multiclass();13:05
saketprintf("testing size : %d\n", result->get_num_labels()) ;13:05
saketlisitsyn: can you please check?13:07
nandosaket: what do you mean?13:07
nandoif you don't pass test data, it only uses the train data yeah13:07
-!- makokal [~bokal@] has joined #shogun13:08
nandoupps wrong nick13:08
-!- nando is now known as iglesiasg13:08
-!- mode/#shogun [+o iglesiasg] by ChanServ13:08
saketiglesiasg: it should use the distance->rhs as per the code13:09
@iglesiasgsaket: mmm I don't understand sorry13:09
saket /* create distance */ 13:11
saketCEuclideanDistance* distance=new CEuclideanDistance(features, testfeatures);  13:12
saket // create KNN  CKNN* knn = new CKNN(3, distance, labels ) ;13:12
saketCKNN* knn = new CKNN(3, distance, labels ) ;13:12
saketknn->train(); 13:12
saketCMulticlassLabels* result = knn->apply_multiclass(); 13:12
saketprintf("testing size : %d\n", result->get_num_labels()) ;13:12
saketI am expecting testing size = |testfeatures|13:13
saketBut I get testing size = |features|13:13
@iglesiasgok, I understand what you mean13:15
@iglesiasglet me check13:15
saketiglesiasg: :)13:15
saketiglesiasg: do you want me to send this testing code ?13:16
@iglesiasgsaket: the result is right13:16
@iglesiasgyou should pass the test features to KNN if you want to apply them to the test features13:16
@iglesiasgreason why?13:17
saketiglesiasg: reason why? I didn't get you13:17
@iglesiasga classifier has *one* set of features I believe13:17
@iglesiasgindependently that it is associated or not with distance or kernels13:18
@iglesiasgwhich do have two set of features13:18
@iglesiasgtherefore you have to either set the features you want to test before calling apply13:18
@iglesiasgor give them to apply13:19
saketI wil do so13:20
saketiglesiasg: also you had asked me to send a PR to develop in
saketiglesiasg: I had sent one to develop a few days ago ( ) ...please take a look :)13:22
-!- iglesiasg [] has quit [Read error: Operation timed out]13:23
-!- van51 [] has left #shogun ["QUIT :Leaving."]13:32
-!- ishan [~ishan@] has joined #shogun13:36
-!- ishan_ [~ishan@] has joined #shogun13:38
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]13:38
sijin@lisitsyn sorry, i went out for a while. i am back13:40
sijin@lisitsyn  I intend to warp Kernel PCA using the projecting function13:41
-!- shogun-buildbot [] has quit [Remote host closed the connection]13:41
-!- ishan__ [~ishan@] has joined #shogun13:42
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]13:42
-!- iglesiasg [] has joined #shogun13:43
-!- mode/#shogun [+o iglesiasg] by ChanServ13:43
@iglesiasgsaket: about the PR13:44
saketiglesiasg: yes13:44
@iglesiasgsaket: as I told you in the other PR13:45
@iglesiasgI cannot merge PRs so another person will take a look :)13:45
@iglesiasgsaket: I remember though that you had some problems with multiple definitions of v_array13:45
@iglesiasgdid you solve that?13:45
saketKinda yes,13:46
saketI switched to vector from STL13:46
@iglesiasghehe ok13:46
saketinstead of v_array from /lib/v_array.h13:46
@iglesiasgso JLCoverTree is using now stl::vector?13:46
saketMine (BallTree) is13:47
@iglesiasgah ok, got it13:47
@iglesiasgsaket: how is the ball tree working?13:47
@iglesiasgis it already finished?13:48
saketIts working good13:48
saketfor small dimensions13:48
sijin@lisitsyn The current Kernel PCA calculate the transformation matrix using "computer_eigenvector", and preserve the transformation_matrix.13:48
@iglesiasgsaket: how fast is it?13:48
saketit has very high relative speedup13:48
saketbut as dimensions increase13:48
@iglesiasgsaket: compare to cover tree?13:48
saketit decreases to 1.2 near 4013:48
saketI will compare to cover tree ..13:48
saketcurrently comparing with naive knn13:48
@iglesiasgaham ok13:49
@iglesiasgour naive knn is pretty slow IIRC13:49
-!- ishan [~ishan@] has joined #shogun13:49
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]13:49
saketshall i use sort from algorithms STL to check then ?13:50
saket*algorithm STL13:50
@iglesiasgit is not a matter of the quicksort13:50
@iglesiasgsonney2k even claims his is faster than STL's one :)13:50
@iglesiasgit is a matter of KNN13:51
@iglesiasglet me take a look to remember what was it13:51
sijin@lisitsyn I just felt that so little work is done with tapkee if tapkee lib is only used for calculationg the transformation(embeded data).13:51
@iglesiasglines 155 to 18513:53
saketiglesiasg: looking ..13:53
-!- abinash [75ef5e6e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]13:55
saketiglesiasg: its not *that* slow either... perhaps the only speed up i can think here is somehow making labels and distances as pair in line 17413:56
@iglesiasgsaket: why would that be faster?13:57
saketiglesiasg: saving cpu operations13:57
saketfaster in cpu utilisation by saving the loop not order (big-O ) wise13:58
@iglesiasgsaket: mmm I don't see why it would save cpu operations13:59
-!- shogun-buildbot [] has joined #shogun14:00
saketiglesiasg: because you won't need the executions in lines 176 && 17714:01
@sonney2kiglesiasg, than qsort in libc - I don't know about std::sort14:01
saketsay you have an array of nodes { float64_t dist, int32_t label }14:01
saketand the distance class could directly fill this array14:01
saketthen lines 176 && 177 would not be needed14:02
-!- shogun-notifier- [] has joined #shogun14:03
shogun-notifier-shogun: Soeren Sonnenburg :develop * aa9f749 / src/NEWS:
shogun-notifier-shogun: update news14:03
@iglesiasgsaket: mmm14:04
@wikingsonney2k: have u managed to check buildbot?14:04
@iglesiasgsaket: I think you need the train_lab array for every example14:04
@iglesiasgsaket: so my guess is that you actually need to do that14:04
@iglesiasgsaket: maybe you can take lines 176 and 177 out of the loop14:04
@iglesiasgsaket: but you would need to create n arrays like that in any case14:05
@iglesiasgI may be wrong though :)14:05
@iglesiasgsaket: anyway, I think I remembered now why I was saying our KNN is slow14:05
@sonney2kwiking, just trying it now14:06
@iglesiasgsaket: so what is the complexity of our naive KNN? O() of what?14:06
saketiglesiasg: Yes it won't work for multiple train examples..14:06
@iglesiasgsaket: for each tet example isn't it?14:07
@iglesiasgand we are doing *K*NN not *N*NN :)14:07
@iglesiasgwhat I mean is K << N normally14:07
@iglesiasgsaket: agree?14:08
saketgenerally it is14:08
@iglesiasgsaket: indeed14:08
@iglesiasgsaket: now this is the part I like14:08
@iglesiasgsaket: do you need O(N log(N)) to get the first K elements of an array of N elements?14:08
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]14:08
@iglesiasgiglesiasg: with the first I mean that you need to order them according to something14:09
@iglesiasgsaket: indeed, how much do you need?14:09
-!- ishan [~ishan@] has joined #shogun14:09
saketyou can get them in a method to that of quick sort in O(n)14:09
saket*method similar14:09
@iglesiasgsaket: in O(N)?? how14:09
saketusing Order statistics14:09
saketit is similar to the way in which we get median in O(n)14:10
@iglesiasgmmm not sure what that is14:10
@iglesiasgbut my claim is that you can't do that faster than O(N log(K))14:10
@iglesiasgO(N) is too optimistic hehe14:10
saketI think the algo is more commonly known as "median of medians"14:10
saketso basically using "median of median" technique one can get the k largest elements in an array in O(n)14:11
saketand then O(klogk) to sort them ...but k << N so it is O(n) only14:11
-!- shogun-buildbot [] has quit [Remote host closed the connection]14:11
-!- shogun-buildbot [] has joined #shogun14:12
@lisitsynI guess that?14:12
saketiglesiasg: the name of algorithm is "Selection algorithm"14:12
-!- shogun-buildbot [] has quit [Remote host closed the connection]14:13
saketlisitsyn: yes exactly that one14:13
@iglesiasgsaket: yeah, I am checking14:13
-!- shogun-buildbot [] has joined #shogun14:13
@iglesiasgsaket: I am not sure but I think they are just forgetting about the log(K) part14:13
@iglesiasgwell forgetting14:14
@iglesiasgone may say that N log(K) \approx N for N >> K14:14
saketno..there is no log(k)14:14
saketif you want the k largest/smallest elements sorted then its O(N + klogk)14:14
sijinthe algorithm is just O(log(n)) + O(n)14:15
-!- shogun-buildbot [] has quit [Remote host closed the connection]14:15
-!- shogun-buildbot [] has joined #shogun14:15
sijinYou use log(n) to find the k-th largest ,after that , you only need to select the numbers that is smaller than the k-th largest14:15
@iglesiasgit looks promising14:17
saketI can do this as an extension to the project LMNN :-)14:18
@sonney2ksaket, idea was that we want kNN results for all possible k=1...K14:18
@iglesiasgsaket: the approach I knew is the one in data structure-based solutions in the last link14:19
-!- ishan_ [~ishan@] has joined #shogun14:19
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]14:19
saketsonney2k: once we have the top K {distance, label} pairs then the code in current KNN is good enough for knn results for all possible 1...k14:20
sijin@lisitsyn do you have time now?14:23
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has joined #shogun14:24
-!- travis-ci [] has joined #shogun14:24
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:
-!- travis-ci [] has left #shogun []14:24
@wikingtravis needs fix :(14:24
@iglesiasgwiking: yes, I was looking around for you :)14:25
@wikingiglesiasg: here14:25
@sonney2kwiking, I am totally clueless why it doesn't work14:25
@iglesiasgwiking: so it seems that what is failing are the tests14:26
@wikingiglesiasg: which, now new ones?14:26
@iglesiasgno idea, let me see14:26
@iglesiasgsome in converter IIRC14:26
@iglesiasgand yesterday something in serialization14:27
@iglesiasgwiking: anyway, can we set a different exit status if tests fail than if the build fails?14:28
@wikingiglesiasg: i'll rebuild now shogun locally14:28
@wikingiglesiasg: and tell u what's happening here with me14:29
@wikingsince in osx a lot of integration tests failed14:29
@iglesiasgwiking: the problem that I see is that now all the pull requests to develop appear as travis build failed14:29
@wikingiglesiasg: yes14:29
@wikingi know14:29
@iglesiasgwiking: even though sometimes it is "only" because I test failed14:29
@wikingiglesiasg: not only14:29
@wikingit should fail if it's really something bad14:29
@wikingthe thing is that there's this serialization error14:29
@wikingok i see there's now even more14:30
@iglesiasgwiking: mmm I have checked a bit some traces and didn't find anything but tests broken14:30
@iglesiasgmaybe I missed sth14:30
@wiking[ FAILED ] MultidimensionaScalingTest.distance_preserving14:30
@wikingwho broke this?14:30
@wikingor rather, which commit broke this?14:30
@iglesiasgno idea14:30
@wikingok well it's time to find out14:30
@iglesiasgwiking: indeed14:31
@wikingwell it's easy14:31
@wikingand find a build where only python_modular fails14:31
@wikinghere it was still ok14:31
@wikinghere it broke14:31
@wikingso lisitsyn  ;)14:31
@wikinglisitsyn: your tapkee update brok MultidimensionaScalingTest.distance_preserving unit test14:32
@iglesiasgwiking: python_modular is broken in both there, isn't it?14:32
@wikingiglesiasg: ignore python_modular error14:32
@wikingnow just let's fix first MultidimensionaScalingTest.distance_preserving unitttest14:32
@iglesiasgwiking: all right, why?14:32
@wikingiglesiasg: i have no idea14:33
@wikingdo a git bisect14:33
@wikingand you'll know why14:33
@wikingi mean for MultidimensionaScalingTest.distance_preserving failure14:33
@iglesiasgwiking: I mean why to ignore python_modular14:33
@wikingfor python_modular i dont know why that exact serialization test fails14:33
@wikingiglesiasg: because that's broken ever since i've enabled python_modular on travis14:33
@iglesiasggot it14:33
@wikingnone of us knows why that doesn't work14:33
@wikingbut it just doesn't14:33
@wikingbut everything else was fine14:34
@wikingi.e. libshogun build both with gcc and clang compiled well and all the unit tests ran as well w/o errors14:34
@wikingand it got broken with lisitsyn's tapkee upgrade14:35
@wikingso it's somewhere there... i mean the error14:35
@wikingso please try to fix that14:35
@iglesiasgI understand it now :)14:35
@wikingi'll try to debug now that integration test problem14:35
@iglesiasgwiking: it is sort a fucking huge commit :)14:35
@wikingas i had much more troubles with them on osx14:35
@iglesiasglet's better wait until lisitsyn can take a look14:35
@wikingiglesiasg: ask lisitsyn then wtf has happened :)14:35
@iglesiasgwiking: nah no worries14:36
@iglesiasgit will be fixed soon probably14:36
@wikingyeah right but until then we don't know anything for sure about the PRs14:37
@wikingso it's rather crucial that code that is commited into the repo14:37
@wikingis actually running w/o errors14:37
@iglesiasgwiking: yes, I totally agree14:37
@iglesiasglet's not merge stuff until that is fixed14:37
@wikingthat's a lot of help when finally GSoC starts14:37
@iglesiasgwiking: mmmm one thing14:38
@wikingand then PRs starts to land14:38
@iglesiasgwiking: introducing the fix in the develop14:38
@iglesiasgwiking: wont't change anything in the current state of the PRs, will it?14:38
@wikingiglesiasg: unfortunately no14:38
@wikingit won't try to rebuild it :(14:38
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]14:38
* iglesiasg wonders if rebasing and pushing might be a solution14:39
-!- ishan_ [~ishan@] has joined #shogun14:39
@lisitsynwiking: iglesiasg: re14:41
@lisitsynwiking: I know it fails will fix today14:41
@iglesiasglisitsyn: :)14:42
@lisitsynI totally forgot to run tests yesterday14:43
@iglesiasgwiking: I think a possible solution for the PRs after the fix is in develop is14:43
@iglesiasgrebase upsatram develop14:43
@iglesiasgand push to origin develop/feature14:44
@iglesiasgalthough I think the second step would require forcing the push or deleting the branch in the repo first14:44
sijinlisitsyn: would you talk about your expectation of kpca warper?14:44
@lisitsynsijin: yes, what's up?14:47
sijinI would like to hear your expecation of this warper so that I can develop the warper which can agree with your interface in tapkee14:49
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]14:49
@iglesiasgsonney2k: how is it going in Paris BTW? nice talks?14:49
-!- ishan_ [~ishan@] has joined #shogun14:49
@lisitsynsijin: alright - as I see that14:49
@lisitsynwe can just store the embedding matrix and later use it in apply14:50
@lisitsynIIRC kpca is just K_{train,test} * embedding_{train}14:50
sijinyeah, i know that. in current implementation, nxn matrix is stored, which is of course wasting.14:51
@lisitsynsijin: what tapkee does is lines 72-9614:55
sijinYes. I have read this code14:55
vgorbatilisitsyn: 'will fix today' - I guess it would be better if I send my PR's after your fix?14:55
@lisitsynvgorbati: your pr with isomap test?14:56
vgorbatilisitsyn: both: with isomap test and with t-SNE wrapper14:57
@lisitsynfeel free to send anytime - should not break my fix14:57
sijinThe reason why I start from Projecting function is that I found many code are doing the same thing. I would like to have a simple function to do that14:57
@lisitsynsijin: just implement it as kernel pca member function14:58
sijinFine. I will do that.14:58
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:59
vgorbatilisitsyn: I meant that if I wait for your fix, then send a PR and then a travis build fails it would be clear that's because of my code. Otherwise it wont be that clear)15:00
sijinlisitsyn: btw, i have implemented the interface to use ProjectingFunciton. If any other embedding method need that I can commit that file also15:01
@lisitsynsijin: I don't know - lets see15:01
-!- vikram360 [~vikram360@] has joined #shogun15:01
@lisitsynvgorbati: I'd say ETA 20-21 kiev time15:02
vgorbatilisitsyn: :D15:02
sijinI mean I have written a interface in "tapkee_shogun.cpp", It can return the projecting function as output15:02
vgorbatilisitsyn: thanks for localization)15:03
@lisitsynsijin: alright that may appear useful15:03
sijinFine. I will commit them together with the warper. Or if you want to see now, I can commit that file right now, since the warper will not use Projecting funciton.15:06
@lisitsynsijin: just do a PR once you are ready15:07
sijinlisitsyn: :), ok15:08
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]15:09
-!- ishan_ [~ishan@] has joined #shogun15:09
-!- abinash [75ef5e6e@gateway/web/freenode/ip.] has joined #shogun15:16
-!- ishan__ [~ishan@] has joined #shogun15:20
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]15:20
-!- abinash [75ef5e6e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]15:27
-!- ishan [~ishan@] has joined #shogun15:39
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]15:39
-!- shogun-buildbot [] has quit [Remote host closed the connection]15:41
@wikingiglesiasg: ok15:45
@wikingiglesiasg: i've recompiled a lot of things15:46
@wikingiglesiasg: but still15:46
@wikingiglesiasg: the same things fail as yesterday15:47
@wikinglisitsyn: yo15:47
@iglesiasgwiking: the tapkee one?15:47
@wikingiglesiasg: no, i mean integration test of python_modular15:48
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]15:49
-!- ishan [~ishan@] has joined #shogun15:50
@lisitsynwiking: yes15:50
@wikinglisitsyn: your tapkee update broke a unitttest15:51
@lisitsynwiking: bahh! really? I said I'll fix it at least two times :D15:51
@wikingfix it!15:54
@wikinglisitsyn: [ RUN ] MultidimensionaScalingTest.distance_preserving15:55
@wikingconverter/ Failure15:55
@wikingThe difference between euclidean_distance_matrix(i,j) and euclidean_distance_for_embedding_matrix(i,j) is 275.89303860676392, which exceeds 1e-9, where15:55
@wikingeuclidean_distance_matrix(i,j) evaluates to 4.5450052246886221,15:55
@wikingeuclidean_distance_for_embedding_matrix(i,j) evaluates to 280.43804383145255, and15:55
@wiking1e-9 evaluates to 1.0000000000000001e-09.15:55
@wiking[ FAILED ] MultidimensionaScalingTest.distance_preserving (6 ms)15:55
@lisitsynwiking: come on I said I'll fix it tonight15:56
@wikinglisitsyn: :)15:56
@lisitsynwiking: I do not know the reason it fails but should be some easy fix15:58
@wikingcan we disable integration test?16:00
@wikingat least with python16:00
@lisitsynwiking: all integration tests?16:00
@lisitsynor what?16:00
@wikingbecause it fails big time on my OS16:00
@wikinglisitsyn: at least python_modular16:00
@lisitsynwhich test?16:00
@lisitsyndisable where?16:01
@wikinglisitsyn: these:
@lisitsynwiking: this should be called reverse engineering - when you engineer a solution that always fails16:02
@wikinglisitsyn: so obviously with integration test for python something is very weird16:02
@wikingit works for some default setups16:02
@lisitsynwiking: kinda strange it always fails16:03
@wikingbut for example on osx it's just fails big time16:03
@wikingand travis build machine create some weird shit as well16:03
@wikingand now because of faulty integration tests we are not having successful builds on travis16:03
@wikinglisitsyn: but of course it might be that the seeded random for osx is tooootaly different than the one on linux16:04
@wikingas obviously the random generator in osx is different from linux16:04
@wikingand this way16:04
@wikingof course we'll have some troubles...16:04
@wikingrelying on seed+random16:05
@wikinghmmm i guess the intergration test would fail as well on freebsd16:06
@lisitsynwiking: but we don't really on that that much?16:07
@lisitsynnot really16:07
@wikinglisitsyn: apart from those tests that are failing on osx :P16:07
@lisitsynwiking: have you checked or?16:08
@wikinglisitsyn: just checking whether it's consistent on freebsd ;P16:08
@wikingi would guess so16:08
@wikingas the random machine on osx and freebsd are the same16:08
@wikingbut different from linux16:08
-!- ishan_ [~ishan@] has joined #shogun16:10
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]16:10
@wikinglisitsyn: and indeed the one failing on travis as well based on a seeded random input16:10
@lisitsynwiking: argh - any idea how to resolve that?16:10
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:11
@wikingwell ... dont rely on seeded random input? :)16:13
@lisitsynwiking: sounds like 'write high-quality code'16:13
@wikinglisitsyn: indeed :D16:14
@lisitsynwiking: you've been to industry for years haven't you?16:14
@wikinglisitsyn: yes :D16:14
@wikingi wanted to say 'unfortunately yes' :D16:14
@lisitsynwiking: we don't differ that much16:15
@lisitsynso we need real solution :)16:15
@wikingi'd like to propose that we remove integration testing for the time being when calling make tests16:15
@lisitsynconverting all examples would take some time16:15
@wikingyeah i know16:15
@wikingbut until then16:15
@wikinglet's not try to run them16:15
@lisitsynI don't mind, but ask sonney2k16:16
@wikingsonney2k: ^16:16
@wikingbut let me double-check my assumption about random here16:16
@wikingjust trying to do integration test on fbsd16:16
@wikinglet's hope the same tests fails16:17
@wikinglisitsyn: i'm sure we can even break integration tests within linux as well16:17
@wikingit's just a matter of tweaking the random machine16:17
@wikingshould i add 'rewrite integration tests' as an entrance bug in github? :)))16:20
@wikingif somebody does that, imo he passed :P16:20
-!- ishan__ [~ishan@] has joined #shogun16:20
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]16:20
@lisitsynwiking: heh well feel free16:20
@lisitsynbut not like rewrite16:21
-!- iglesiasg [] has quit [Read error: Operation timed out]16:21
@lisitsynget rid of random tests I think16:21
@wikinganybody seen this baby before:
-!- gsomix [] has quit [Quit: Miranda IM! Smaller, Faster, Easier.]16:26
@wikinglol man somebody still uses miranda IM16:26
@wikingamaaaazing gsomix! :)16:26
@lisitsynwiking: old school :D16:27
@lisitsynwiking: I used it in 2006 I think16:28
@lisitsynand 2007 probably16:28
@wikingsomewhere there16:28
-!- erlenda [] has quit [Quit: Leaving]16:37
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]16:39
-!- ishan__ [~ishan@] has joined #shogun16:40
-!- ishan [~ishan@] has joined #shogun16:50
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]16:50
-!- Kirsh [~kirshanth@] has joined #shogun16:57
-!- gsomix [~Miranda@] has joined #shogun17:01
-!- shogun-notifier- [] has quit [Quit: transmission timeout]17:03
-!- lisitsyn [] has quit [Quit: Leaving.]17:09
-!- ishan [~ishan@] has quit [Ping timeout: 252 seconds]17:11
-!- ishan [~ishan@] has joined #shogun17:19
-!- Yanglittle [deb20af4@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:36
* wiking hates recompiling atlas...17:43
-!- gsomix [~Miranda@] has quit [Quit: Miranda IM! Smaller, Faster, Easier.]17:45
-!- ishan_ [~ishan@] has joined #shogun17:50
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]17:50
-!- ishan__ [~ishan@] has joined #shogun18:20
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]18:20
-!- lisitsyn [~blackburn@] has joined #shogun18:35
-!- mode/#shogun [+o lisitsyn] by ChanServ18:35
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun18:37
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]18:50
-!- ishan__ [~ishan@] has joined #shogun18:50
-!- gsomix [~Miranda@] has joined #shogun18:59
gsomixgood evening18:59
-!- sids_aquarius [] has quit [Ping timeout: 245 seconds]19:04
-!- votjakovr [] has quit [Quit: ChatZilla 0.9.90 [Firefox 20.0.1/20130409194949]]19:11
-!- ishan [~ishan@] has joined #shogun19:20
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]19:20
-!- sids_aquarius [~sidi_@] has joined #shogun19:37
sids_aquariusjoin #tux4kids19:37
-!- saket [0e8b5206@gateway/web/freenode/ip.] has joined #shogun19:39
deerishi@HeikoS: hi19:49
@HeikoSdeerishi: hi19:49
deerishijust sent my pull request19:49
@HeikoSah nice!19:49
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]19:49
-!- ishan [~ishan@] has joined #shogun19:50
deerishihope this one has everything you asked for . :)19:50
deerishii reduced the number of loops.19:50
deerishialso now19:50
deerishithe data from tools19:50
deerishiis being imported directed19:51
@HeikoSdeerishi: ok nice19:51
@HeikoSstill the while loops19:51
deerishikindly give me your feedback . :)19:51
@HeikoSI dont like while loops in python19:51
@HeikoSyou can easily replace them by vector operations19:51
deerishijust the two because , sqrt function takes in only 1 element at a time19:51
deerishirest all are the vector operations19:52
@HeikoS[sqrt(x) for x in bla]19:52
@HeikoSno need for loops19:52
@HeikoSalso those guys here:19:53
@HeikoSfor i in range(lx0):+    modified_d1[i][:]=X2[i][0:number_of_points_for_circle1]19:53
@HeikoSif you use the variable you are iterating over only for indexing, you dont need a loop19:53
@HeikoSesp no while loop19:54
@HeikoSslow - danger of non-termination etc19:54
-!- pararth [0e8b614d@gateway/web/freenode/ip.] has joined #shogun19:57
@HeikoSdeerishi: no big deal to remove them, everything else is fine. Once this is done we can renew the integration test :)20:04
deerishijust give me a couple of minutes, i will remove them20:05
@HeikoSdeerishi: no worries20:05
deerishiso its good enough to merge ?20:07
@HeikoSdeerishi: once the loops are hone, we can do the integration test which is a minor thing, and then merge20:08
-!- ishan_ [~ishan@] has joined #shogun20:20
-!- ishan [~ishan@] has quit [Read error: Connection reset by peer]20:20
@lisitsynwiking: bug in custody20:26
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]20:26
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:29
-!- saket [0e8b5206@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:35
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:36
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun20:40
deerishi@HeikoS: just sent another PR20:40
@HeikoSdeerishi: saw it!20:41
@HeikoSnow this looks nicer! :)20:41
deerishican  you check it?20:41
deerishifinally.... :)20:41
@HeikoSdeerishi: do the results still make sense?20:41
@HeikoSif you run it?20:41
deerishiresults are still the same20:41
@HeikoSso then lets make the integration test work20:41
@HeikoSwhat you need to do is to go to tests/integration/python_modular20:42
@HeikoSthen run python preprocessor_kpca.py20:42
@HeikoSthis will generate two files20:42
@HeikoSwhich contain the things that your example returns, one file for each eleent of the parmeter list20:42
@HeikoSthey are in tests/data20:42
@HeikoSyou  have to commit those20:43
@HeikoSonce this is done, you have to commit the data version in the development repository20:43
@HeikoSthen push everything20:43
@HeikoSto update PR20:43
@HeikoSthen the travis build will work20:43
@HeikoSand one more small thing, have a look in examples/documented/ there are descriptions for the examples, please add one for yours, just a few lines that exaplain what is being done in the example20:44
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:45
-!- ishan__ [~ishan@] has joined #shogun20:50
-!- ishan_ [~ishan@] has quit [Read error: Connection reset by peer]20:50
-!- iglesiasg [] has joined #shogun20:54
-!- mode/#shogun [+o iglesiasg] by ChanServ20:54
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun20:59
-!- pararth [0e8b614d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:02
-!- shogun-notifier- [] has joined #shogun21:03
shogun-notifier-shogun: Sergey Lisitsyn :develop * 2fcf2ce / src/shogun/lib/tapkee/tapkee_methods.hpp:
shogun-notifier-shogun: Fixed bug in MDS21:03
@iglesiasglisitsyn: yaaay bug fixed :)21:04
@iglesiasglisitsyn: do you think we should do sth about the pull requests with travis?21:09
@iglesiasg LLE neighbors preserving21:11
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:12
-!- iglesiasg [] has quit [Quit: leaving]21:15
-!- iglesiasg [] has joined #shogun21:18
-!- mode/#shogun [+o iglesiasg] by ChanServ21:18
-!- iglesiasg [] has quit [Client Quit]21:19
-!- ishan [~ishan@] has joined #shogun21:21
-!- ishan__ [~ishan@] has quit [Read error: Connection reset by peer]21:21
-!- travis-ci [] has joined #shogun21:23
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []21:23
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has joined #shogun21:24
-!- ishan [~ishan@] has quit [Remote host closed the connection]21:49
-!- zxtx [] has quit [Ping timeout: 264 seconds]22:02
-!- deerishi [73f88294@gateway/web/freenode/ip.] has joined #shogun22:04
deerishi@HeikoS: i ran the integration test but it gave the following error 'module' object has no attribute 'preprocessor_kpca'22:05
deerishi ERROR generating '../../../tests/integration/tests/preprocessor_kpca0.txt' using ''22:07
deerishi@HeikoS : what does it mean?22:07
@HeikoSpls show me the code again22:07
-!- gsomix [~Miranda@] has quit [Quit: Miranda IM! Smaller, Faster, Easier.]22:09
@HeikoSdeerishi: and the call you did?22:11
@HeikoSdeerishi: please pastebin the command and the output22:11
deerishipython preprocessor_kpca.py22:12
@HeikoSdeerishi: sorry, but you have to ask somebody else on this, I will go home now,22:14
deerishican we discuss this tomorrow?22:14
-!- zxtx [] has joined #shogun22:14
@HeikoSdeerishi: busy tomorrow, but we can try22:14
@HeikoSyou can also ask lisitsyn on this22:15
@HeikoSbut please create a full pastebin on the error22:15
@HeikoShave a good day! :) bye22:15
deerishibye. :)22:15
deerishi@lisitsyn : der?22:16
-!- HeikoS [] has left #shogun []22:18
deerishi@lisitsyn : could you help me in the integration test22:20
deerishii am getting an error ERROR generating '../../../tests/integration/tests/preprocessor_kpca0.txt' using ''22:20
deerishiwhat does it mean ? and how can i solv eit22:20
deerishi*solve it ?22:21
deerishithe code is working fine22:22
shogun-notifier-shogun: :develop * 43dea2e / tests/unit/converter/ (2 files):
shogun-notifier-shogun: Switched the unit test for any neighborhood preserving DR method from testing Locally Linear Embedding method to testing Isomap (since LLE seems to be not very robust to noise).22:26
shogun-notifier-shogun: :develop * ffc40a3 / tests/unit/converter/
shogun-notifier-shogun: Fixed missing #include's in test
shogun-notifier-shogun: Sergey Lisitsyn :develop * c9153ac / tests/unit/converter/ (2 files):
shogun-notifier-shogun: Merge pull request #1016 from vladislav-horbatiuk/lleTest-fix22:26
shogun-notifier-shogun: Switched the unit test for any neighborhood preserving DR method from sting Locally Linear Embedding method to testing Isomap.22:26
-!- deerishi [73f88294@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:30
-!- sids_aquarius [~sidi_@] has quit [Read error: Connection reset by peer]22:36
-!- sids_aquarius [~sidi_@] has joined #shogun22:39
-!- sids_aquarius [~sidi_@] has left #shogun []22:42
-!- vgorbati [d4029f22@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]22:42
-!- sids_aquarius_ [~sidi_@] has joined #shogun22:47
-!- travis-ci [] has joined #shogun23:04
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []23:04
-!- makokal [~bokal@] has quit [Quit: makokal]23:06
-!- sids_aquarius_ [~sidi_@] has quit [Read error: Connection reset by peer]23:13
-!- sids_aquarius_ [] has joined #shogun23:13
-!- iglesiasg [] has joined #shogun23:17
-!- mode/#shogun [+o iglesiasg] by ChanServ23:17
-!- Kirsh [~kirshanth@] has quit [Ping timeout: 256 seconds]23:19
-!- makokal [] has joined #shogun23:34
--- Log closed Thu Apr 25 00:00:29 2013