Open in new window / Try shogun cloud
--- Log opened Thu Nov 08 00:00:17 2012
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]01:12
-!- zxtx_ [] has quit [Ping timeout: 246 seconds]01:59
-!- zxtx_ [] has joined #shogun02:51
-!- zxtx_ [] has quit [Ping timeout: 255 seconds]06:24
-!- zxtx_ [~zv@] has joined #shogun07:48
-!- zxtx_ [~zv@] has quit [Ping timeout: 245 seconds]08:03
-!- n4nd0 [] has joined #shogun09:28
n4nd0sonney2k: hi, no news about doc camp yet?09:29
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun10:09
-!- blackburn_ [5bdfb203@gateway/web/freenode/ip.] has joined #shogun10:12
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]10:15
wikingwhere's heiko? :)10:18
-!- n4nd0 [] has quit [Quit: leaving]12:24
-!- zxtx_ [] has joined #shogun15:50
blackburn_wiking: who's that for? :)16:08
wikingwell it's about streaming features16:08
blackburn_first rule of streaming features16:09
-!- sonne|work [~sonnenbu@] has joined #shogun16:09
wikingno wonder it leaks16:09
blackburn_no, first rule is never talk about streaming features obviously ;)16:09
sonne|workbetter leaks than GC :)16:09
wikingwho's Shashwat Lal Das?16:10
blackburn_our former gsoc16:10
blackburn_I'd say alumni16:10
sonne|workwiking: btw you can find the git tag in any shogun 2.0 .tar.bz2 in src/lib/versionstring.h16:10
blackburn_wiking: if you have any questions to shashwatted code ask heiko16:11
blackburn_he is shashwat now16:11
wikingsonne|work: do u have it hand?16:12
wikingdownloading ...16:12
wikingstay tuned16:12
blackburn_wiking: okay joking - shashwat code is ok16:12
blackburn_baozeng is not :D16:13
wikingwell yeah16:13
wikingthat's another story16:13
blackburn_okay time to get hooome16:14
-!- blackburn_ [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:14
wikingsonne|work: i've downloaded tar.bz216:14
wikingsonne|work: should i build it inorder to have that src/lib/versionstring.h?16:15
sonne|workit is there already16:16
wikingi've just downloaded the tar.bz2 from the shogun site16:17
wikingand there's no such thing there16:17
wikingthe revision16:18
sonne|workgit show --pretty='format:%h'|head -116:18
wikinghow do u convert that into a git commit number?16:18
sonne|workthat's how it is created16:18
sonne|workjust do git show 9c8012f16:19
wikingfatal: ambiguous argument '9c8012f': unknown revision or path not in the working tree.16:19
sonne|worksomething went wrong.16:20
sonne|workwiking: does it work in your current checkout?16:20
wikinggit show --pretty='format:%h'|head -1 ?16:20
sonne|workI mean the id from your current build16:21
sonne|workthat you find in your git-master src/shogun/lib/versionstring.h16:21
wikingthat works16:21
sonne|worklooking at the logs I think we released 15ed29239e2f83e7cd8c5097d9180999cffeab5616:23
wikinglet's tag that then16:27
wikingand make a branch of it16:27
sonne|workwe should compare it to the stuff in the tar.bz2 before we do it16:34
sonne|workI have no idea what could have happened ...16:34
sonne|workwell release script needed a rewrite to work with new layout / git and we didn't release that often since then16:35
sonne|workso it probably still has issues16:35
wikingwe should test it before the next release then16:36
wikingcan we agree to throw this ctor out?16:36
wikingas it doesn't do anything16:37
wikingsonne|work: ^16:37
wikingsame for
wikingsonne|work: check it out16:38
wiking :)16:38
sonne|workthat oen used to do sth16:38
sonne|workneeds more inspection - maybe some code got removed16:38
wikingok i'm not going to go into that then now16:38
sonne|workit was used to read in files that have a label and features16:38
wikingsonne|work: yes but now that is moved16:38
wikingsonne|work: to StreamingSparseFeatures16:39
wikingand DenseFeatures16:39
sonne|workahh ok16:40
sonne|workthen it certainly makes sense yes16:40
wikingsonne|work: ok then i'll prepare now a fix16:41
-shogungit:#shogun- [shogun] vigsterkr opened pull request #816: Fix StreamingFeatures (master...latent)
wikingthere you go fixes17:06
wikingwe have a nice bug in CMulticlassLibSVM17:21
wikingor at least there's a problem with ./classifier_multiclasslibsvm17:21
wikingbtw anyone tested shogun with python3.0?17:25
-!- blackburn [~blackburn@] has joined #shogun17:48
wikingblackburn: readlogs :)17:48
blackburnwiking: a lot of things, what in particular?17:49
wiking7:20 < wiking> ooooh17:50
wiking17:21 < wiking> we have a nice bug in CMulticlassLibSVM17:50
wiking17:21 < wiking> or at least there's a problem with ./classifier_multiclasslibsvm17:50
wiking17:25 < wiking> btw anyone tested shogun with python3.0?17:50
wiking17:25 < wiking> :)17:50
blackburnwiking: what is da bug?17:51
wikingblackburn: try running ./classifier_multiclasslibsvm with HEAD17:51
wikingit fails here17:51
wiking190REQUIRE(set_svm(idx, svm),"SVM set failed");17:51
wikingthat's line 19017:51
wikingidx value there is -217:51
blackburnwoopie doopie17:51
blackburn2 classes17:54
blackburnwiking: here you go17:56
wikingwaiting :)17:57
wikingwhere r u shogungit?! :)17:57
wikingblackburn: and merge that PR ;)17:57
-shogungit:#shogun- [shogun] lisitsyn pushed 1 new commit to master:
-shogungit:#shogun- shogun/master ce7eafe Sergey Lisitsyn: Patched multiclass libsvm to support 2 classes17:57
blackburnyes master17:57
-shogungit:#shogun- [shogun] lisitsyn pushed 2 new commits to master:
-shogungit:#shogun- shogun/master a4bc1c9 Viktor Gal: Fix StreamingFeatures...17:57
-shogungit:#shogun- shogun/master 9e87cd5 Sergey Lisitsyn: Merge pull request #816 from vigsterkr/latent...17:57
wikingbug hunters we are :)18:00
-shogungit:#shogun- [shogun] cwidmer opened pull request #817: added serialization example (master...master)
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]18:32
-!- blackburn [~blackburn@] has joined #shogun18:33
-shogungit:#shogun- [shogun] lisitsyn pushed 3 new commits to master:
-shogungit:#shogun- shogun/master 5195b7f Christian Widmer: polished simple serialization example18:34
-shogungit:#shogun- shogun/master cfe0958 Christian Widmer: added serialization example for combined kernel consisting of spectrum and wdk kernels18:34
-shogungit:#shogun- shogun/master ad7a0e2 Sergey Lisitsyn: Merge pull request #817 from cwidmer/master...18:34
shogun-buildbotbuild #628 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>, Christian Widmer <>19:17
-shogungit:#shogun- [shogun] cwidmer opened pull request #818: fixed typo (master...master)
-!- zxtx_ [] has quit [Read error: No route to host]19:59
-!- zxtx [] has joined #shogun20:03
-!- cwidmer [8295587b@gateway/web/freenode/ip.] has joined #shogun20:11
wikingcwidmer: hey20:14
cwidmerhi there20:15
wikingcwidmer: do you happen to have any DNA/gene dataset where one has to find motifs?20:15
cwidmerwell, I can give you some splice site data... another one would be gene starts20:16
cwidmerwhat are you trying to do?#20:16
wikingi'm really not into this topic but20:17
cwidmeryou got interested ;)?20:18
wikingMotifs are repeated patterns in DNA sequences that are believed to have biological significance. Our dataset consists of ARSs (autonomously replicating se- quences) screened in two yeast species S. kluyveri and S. cerevisiae. Our task is to predict whether a par- ticular sequence is functional (i.e., whether they start the replication process) in S. cerevisiae and to find out the motif responsible. All the native ARSs in S. cerevisiae are labeled as pos20:18
wikingthis is in a paper20:18
wikingi was wondering whether there's a similar dataset available20:19
cwidmerwhich paper is that?20:19
cwidmerI see, and i guess you've checked if their data set is available (website, supplement, ask authors)?20:21
wikingcwidmer: yeps20:22
wikingunfortunately it's not20:22
wikingbut basically any similar problem for me would be good20:22
wikingcwidmer: i mean yeah i've contacted them but unfortunately this dataset in particular cannot be handed out... private etc..20:22
cwidmerhmm, too bad20:23
wikingi've just seen your pull request20:23
cwidmerso they do some labeled sequence learning on motifs?20:23
wikingthat's why i thought i give a shot to ask u something20:23
cwidmerheh, sure20:23
wikingwell as much one can find out from their paper/code20:24
wikingtheir data looks something like this: some_seq:1:1:ACGTACGT20:25
wikingand then they try to find the starting position of the motif as well as classify it20:26
cwidmerI have to read the data part in detail to see what's going on20:27
cwidmerfrom glancing over it, it seems like a binary classification task20:27
cwidmerwith latent variables20:27
cwidmerwhere label is, if functional or not20:27
cwidmerand latent variable is motiv position20:28
cwidmerI was expecting SO20:28
wikingyeah they didn't go that deep :)20:28
cwidmerhmm, ok20:29
cwidmeralso, are they only looking for one particular motif?20:29
cwidmerand the sequences can be of variable length?20:29
wikingif u have anything at hand that u could share for testing shogun's machine20:31
wikinglet me know :)20:31
cwidmerso you're saying all those 'autonomously replicating sequences' are of identical length?20:31
wikingcwidmer: well as far as i can see from the code20:31
cwidmerwell, there is something that a colleague of mine, sebi, worked on: KIRMES20:31
cwidmerhe is also looking at motifs in promoter sequences20:31
cwidmerbut several20:31
cwidmerand he encodes their relative distances etc as features20:32
cwidmerand learns on that20:32
cwidmerpossibly, you can take a look at that20:32
cwidmerbut I suspect the problem is much harder than the latent svm paper one20:33
wikingok i'll check that one out20:34
wikingget at least an idea :)20:35
cwidmerjust curious, what method do you want to test on this data?20:35
wikingsimple latent svm and latent so svm20:36
cwidmeris the code in shogun ported from thorstens's tool or do we have a from scratch implementation?20:37
cwidmerit is based on the same paper at least, right?20:37
wikingcwidmer: we have both20:41
wikingi mean i have it for so20:41
cwidmerbut what's in shogun is based on the "Learning Structural SVMs with Latent Variables" paper or something else?20:42
wikingcwidmer: both20:42
wikingcwidmer: i've tested on one dataset and the one i have from scratch implemented performs better than the algo in ""Learning Structural SVMs with Latent Variables"20:43
cwidmerah nice20:44
wikingbut i need more tests20:44
cwidmerso what you've implemented from scratch is in libocas?20:45
wikingthe simple latent svm is based on libocas20:45
wikingthe latent so is based on the BMRM in shogun20:46
wikingand i've made it possible to work with the primal mosek solver as well20:46
cwidmeri'll check it out some time, seems interesting20:47
wikingcwidmer: the latent so is not commited yet20:47
wikingit's in my git stash for a long time now20:49
-shogungit:#shogun- [shogun] lisitsyn pushed 1 new commit to master:
-shogungit:#shogun- shogun/master a325ca7 Sergey Lisitsyn: Attempted to fix strings serialization process21:05
-shogungit:#shogun- [shogun] lisitsyn pushed 2 new commits to master:
-shogungit:#shogun- shogun/master 67681ed Christian Widmer: fixed typo21:05
-shogungit:#shogun- shogun/master 754d81b Sergey Lisitsyn: Merge pull request #818 from cwidmer/master...21:05
-!- n4nd0 [] has joined #shogun21:34
shogun-buildbotbuild #629 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #630 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Christian Widmer <>21:57
-!- blackburn [~blackburn@] has left #shogun []22:44
-!- blackburn [~blackburn@] has joined #shogun22:44
-!- cwidmer [8295587b@gateway/web/freenode/ip.] has quit [Quit: Page closed]23:30
--- Log closed Fri Nov 09 00:00:17 2012