--- Log opened Tue Mar 19 00:00:14 2013 | ||
-!- zxtx [zv@conference/pycon/x-rnauhrcjrxankcvr] has quit [Ping timeout: 260 seconds] | 00:07 | |
-!- zxtx [zv@conference/pycon/x-xhgwzhjusmxtgavv] has joined #shogun | 00:08 | |
shogun-buildbot_ | build #366 of rpm1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/366 blamelist: Soeren Sonnenburg <sonne@debian.org> | 00:09 |
---|---|---|
-!- zxtx [zv@conference/pycon/x-xhgwzhjusmxtgavv] has quit [Ping timeout: 255 seconds] | 00:27 | |
-!- zxtx [zv@conference/pycon/x-ujcvrjcchgjuzfif] has joined #shogun | 00:46 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 01:05 | |
-!- zxtx [zv@conference/pycon/x-ujcvrjcchgjuzfif] has quit [Ping timeout: 240 seconds] | 01:19 | |
-!- zxtx [zv@conference/pycon/x-hlwmyxstkbabctqs] has joined #shogun | 01:25 | |
-!- zxtx [zv@conference/pycon/x-hlwmyxstkbabctqs] has quit [Ping timeout: 252 seconds] | 02:01 | |
-!- sumit_ [73f91219@gateway/web/freenode/ip.115.249.18.25] has joined #shogun | 04:53 | |
-!- sumit_ [73f91219@gateway/web/freenode/ip.115.249.18.25] has quit [Ping timeout: 245 seconds] | 05:16 | |
@sonney2k | shogun-buildbot_, force build 'rpm1 - libshogun' | 05:48 |
shogun-buildbot_ | build forced [ETA 5m02s] | 05:48 |
shogun-buildbot_ | I'll give a shout when the build finishes | 05:48 |
shogun-buildbot_ | build #367 of rpm1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/367 | 05:53 |
-!- hoijui [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has joined #shogun | 07:41 | |
-!- zxtx [~zv@75-59-238-10.lightspeed.sntcca.sbcglobal.net] has joined #shogun | 08:05 | |
-!- blackburn [~blackburn@185.6.239.95] has quit [Quit: Leaving.] | 08:47 | |
-!- blackburn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:14 | |
sonne|work | blackburn: look http://www.linux-magazin.de/NEWS/Shogun-2.1.0-bringt-hunderte-Verbesserungen | 09:44 |
sonne|work | we've got some press coverage | 09:45 |
-!- blackburn [~lisitsin@mxs.kg.ru] has quit [Ping timeout: 264 seconds] | 09:45 | |
-!- blackburn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:46 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 09:53 | |
sonne|work | blackburn: ok again | 10:09 |
sonne|work | blackburn, n4nd0 look we've got some press coverage http://www.linux-magazin.de/NEWS/Shogun-2.1.0-bringt-hunderte-Verbesserungen | 10:09 |
n4nd0 | sonne|work: awesome | 10:09 |
blackburn | sonne|work: heh cool | 10:10 |
n4nd0 | sonne|work: you are famous | 10:10 |
sonne|work | now if we had fancy applications on the website (d3js ...) then these would be on... | 10:10 |
n4nd0 | sonne|work: blackburn and I are getting something with d3js in tapkee | 10:13 |
n4nd0 | we could put it to in shogun if you like it | 10:13 |
sonne|work | btw they also announced our workshop... http://www.linux-magazin.de/NEWS/Machine-Learning-Workshop-in-Berlin | 10:13 |
blackburn | sonne|work: n4nd0 is going to be a d3 expert soon as he prepares d3.js demo for tapkee :) | 10:13 |
blackburn | sonne|work: preparing DR idea description | 10:13 |
n4nd0 | cool | 10:13 |
n4nd0 | I didn't get anything from Christoph yet :S | 10:13 |
blackburn | who is Christoph? | 10:14 |
n4nd0 | I will write him again soon telling him about what I think it could be done in a project and ask him if he would be interested | 10:14 |
n4nd0 | blackburn: Christoph Lampert | 10:14 |
sonne|work | n4nd0: please do it today | 10:14 |
sonne|work | time is running... | 10:14 |
n4nd0 | I contacted him and asked him if he would like to mentor a project in Structured Learning | 10:14 |
n4nd0 | sonne|work: ok | 10:15 |
blackburn | n4nd0: ah I see | 10:15 |
n4nd0 | blackburn: he has done nice things in computer vision with structured learning | 10:15 |
sonne|work | n4nd0: maybe it helps if you CC me... | 10:15 |
n4nd0 | sonne|work: good idea | 10:15 |
n4nd0 | blackburn: with CRFs and PGMs in general | 10:15 |
n4nd0 | blackburn: http://www.adlibris.com/se/product.aspx?isbn=1601984561 | 10:16 |
n4nd0 | this is a very nice tutorial | 10:16 |
sonne|work | blackburn: btw I am extremely happy that we managed to get a new release out before gsoc :) | 10:18 |
sonne|work | this and that change in development model / tests should make the whole process go much smoother | 10:19 |
blackburn | sonne|work: yes that's cool | 10:23 |
sonne|work | blackburn: I've had a phone chat with alex yesterday | 10:24 |
sonne|work | he is working out the 2 days post the shogun workshop | 10:24 |
sonne|work | it is likely to work out well.. so we can probably announce things end of the week | 10:25 |
blackburn | sonne|work: alright, cool! | 10:25 |
n4nd0 | SWIG question | 10:34 |
n4nd0 | I am making an instance of RealMatrixFeatures in python | 10:34 |
n4nd0 | and giving it to a StructuredModel | 10:34 |
n4nd0 | the member of the StructuredModel is of class Features, parent of RealMatrixFeatures | 10:35 |
n4nd0 | I am trying to call get_features_vector() but it doesn't find it in Features - it doesn't exist there | 10:35 |
n4nd0 | but it does exist in RealMatrixFeatures | 10:35 |
n4nd0 | I am guessing that it either requires some SWIG code or an obtain_from_generic method to cast in python Features to RealMatrixFeatures | 10:36 |
blackburn | sonne|work: I shared a doc with ideas - please send my idea around when you have time | 10:37 |
sonne|work | n4nd0: exactly | 10:38 |
sonne|work | blackburn: just do this directly on the website... | 10:38 |
blackburn | sonne|work: lets keep that in doc yet and publish in a batch mode | 10:38 |
n4nd0 | sonne|work: what is prefereable, obtain_from_generic or swig code? | 10:39 |
blackburn | the thing that worries me is that I'd like to get my idea done by myself LOL | 10:39 |
n4nd0 | blackburn: with the info that Georg wrote we can probably write in there the description | 10:39 |
blackburn | n4nd0: yes I'll put it in | 10:39 |
n4nd0 | blackburn: haha :) it is a nice description | 10:40 |
sonne|work | blackburn: that should not worry you - to the contrary. that will help massively - you might even join the student in coding stuff... | 10:41 |
blackburn | sonne|work: prerequisites: C++, metagenomics | 10:43 |
blackburn | :D | 10:43 |
blackburn | sonne|work: alright 2 ideas ready | 10:54 |
sonne|work | blackburn: put them live and then we can send around another call! | 10:55 |
blackburn | sonne|work: can I put some CI stuff for tapkee on one of our buildbots? | 10:58 |
blackburn | sonne|work: btw I learned how to use gcov so will put it for shogun soon too | 10:59 |
blackburn | sonne|work: tapkee offers me a possibility to play as it compiles just in a few minutes, not hour :D | 11:00 |
sonne|work | blackburn: what is wrong with travis for tapkee then? | 11:00 |
blackburn | sonne|work: I can't put anything out of travis | 11:00 |
sonne|work | what not? only issue why we don't put everything on travis is that shogun's stuff takes too long | 11:01 |
blackburn | sonne|work: how can I push generated lcov output somewhere? | 11:02 |
sonne|work | ? | 11:03 |
blackburn | sonne|work: I generate stuff (valgrind output, etc), how can I get it back from travis? | 11:04 |
sonne|work | no idea but how would you get it back from shogun's bb's? | 11:05 |
blackburn | sonne|work: I can store some pass there may be | 11:07 |
blackburn | okay I'll try to do that with travis then | 11:07 |
sonne|work | blackburn: you mean ssh key and we rsync stuff? or put it in some folder that can be seen from the web? | 11:07 |
blackburn | sonne|work: yeah may be ssh | 11:07 |
blackburn | sonne|work: other way could be to update gh_pages folder | 11:08 |
sonne|work | gh_pages? | 11:08 |
blackburn | sonne|work: webpage on github | 11:09 |
blackburn | sonne|work: we could have shogun-toolbox.github.com/shogun-toolbox page if we had gh_pages branch | 11:09 |
blackburn | or shogun-toolbox.github.com | 11:10 |
blackburn | sonne|work: just like that http://lisitsyn.github.com/tapkee | 11:10 |
blackburn | django editor drives me mad | 11:11 |
blackburn | fscking wysiwyg | 11:12 |
sonne|work | blackburn: it is not django editor | 11:13 |
blackburn | http://shogun-toolbox.org/page/Events/gsoc2013 sonne|work I put them live | 11:13 |
sonne|work | it is tinymce | 11:13 |
sonne|work | bern4rd added it for convenience | 11:13 |
blackburn | sonne|work: I see | 11:14 |
blackburn | sonne|work: anyway it works like windows | 11:14 |
blackburn | in the worst mean of these words | 11:14 |
-!- hoijui [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has quit [Ping timeout: 255 seconds] | 11:19 | |
n4nd0 | hehe | 11:20 |
n4nd0 | blackburn: it fails pretty bad lot of times | 11:20 |
n4nd0 | if happens to me sometimes that I click on "save and continue" and errors start popping up | 11:20 |
-!- hoijui_ [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has joined #shogun | 11:20 | |
sonne|work | n4nd0: which errors? | 11:23 |
n4nd0 | sonne|work: it couldn't be save, try again later | 11:24 |
n4nd0 | I ended up not using save and continue | 11:24 |
sonne|work | I mean I don't mind to return to using just a raw text form | 11:24 |
sonne|work | n4nd0: when did that occur and what do you have to do for that? | 11:24 |
n4nd0 | sonne|work: it has occurred when I have added some info in the webpage, for the workshop program for example that I remember now | 11:25 |
sonne|work | n4nd0: speeking of which please add the date to the schedule :) | 11:26 |
n4nd0 | sonne|work: hehe sure | 11:27 |
sonne|work | n4nd0: and if the problem occurs again tell it to us | 11:29 |
n4nd0 | sonne|work: error occurred :) | 11:30 |
sonne|work | n4nd0: so what do you do exactly? | 11:30 |
n4nd0 | I edited on workshop2013_program "in July" to "on July 12th" | 11:30 |
n4nd0 | save and continue | 11:30 |
n4nd0 | that worked fine | 11:30 |
sonne|work | and then? | 11:30 |
n4nd0 | although I was directed to the same page but with poor formatting | 11:31 |
n4nd0 | not the normal Django admin style page | 11:31 |
n4nd0 | then I clicked save | 11:31 |
n4nd0 | and | 11:31 |
n4nd0 | An internal server error occurred. Please try again later. | 11:31 |
sonne|work | but that is not the editor then | 11:32 |
sonne|work | I wonder why we didn't receive an email with the exception | 11:32 |
n4nd0 | sonne|work: this is what I mean with page with poor formatting http://dl.dropbox.com/u/11020840/2013-03-19-113315_1600x900_scrot.png | 11:32 |
n4nd0 | sonne|work: anyway the modification is fine | 11:33 |
sonne|work | n4nd0: I am not so worried about the formatting (IIRC reloading would fix this) but more about the internal server error... | 11:36 |
blackburn | sonne|work: how to say 'thanks, just in time' in german? :D | 11:57 |
blackburn | I do not trust google translate | 11:57 |
sonne|work | gerade noch rechtzeitig | 11:57 |
sonne|work | but this sounds a bit negative | 11:58 |
blackburn | sonne|work: what about positive? | 11:58 |
blackburn | I am sorry to distract you - got curious to answer andreas in german | 11:58 |
sonne|work | no idea | 11:58 |
blackburn | alright ;) | 11:59 |
n4nd0 | blackburn: andreas from scikits? | 12:05 |
blackburn | n4nd0: no, Ziehe | 12:06 |
blackburn | not mueller :) | 12:06 |
sonne|work | blackburn: ha pushing helps :) | 12:06 |
blackburn | sonne|work: pushing? | 12:07 |
sonne|work | blackburn: I've email andreas just an hour ago or so | 12:07 |
blackburn | ahh :) | 12:07 |
blackburn | they forgot I guess | 12:07 |
blackburn | sonne|work: let me kick some asses too :D | 12:07 |
blackburn | sonne|work: 3 ideas done | 12:18 |
-!- heiko [~heiko@nat-187-2.internal.eduroam.ucl.ac.uk] has joined #shogun | 12:21 | |
blackburn | heiko: could you please fill in your gsoc idea? | 12:22 |
blackburn | I shared a doc | 12:22 |
heiko | blackburn: yes I saw it | 12:24 |
heiko | I will do it soon | 12:25 |
heiko | blackburn: and I will also help on the application | 12:25 |
heiko | blackburn: can you update me on the deadlines? | 12:25 |
blackburn | heiko: 29 march | 12:25 |
blackburn | heiko: so 10 days, pretty enough | 12:25 |
heiko | blackburn: ok, I will add some stuff tonight and continue to update this week | 12:26 |
heiko | blackburn: what do you think on a project on GP classification? | 12:26 |
heiko | (in case nobody wants to do the large sparse Gaussian stuff) | 12:27 |
heiko | could do: binary GP with laplace approx and EP, same for multiclass, unit-testing the existing stuff ....and thats already a nice little project | 12:28 |
blackburn | heiko: mentor? | 12:32 |
heiko | blackburn: me | 12:32 |
blackburn | sure | 12:32 |
blackburn | :D | 12:32 |
heiko | blackburn: I implemented most of this stuff end of last year, not all though, but its all covered by the GP book | 12:32 |
blackburn | heiko: I never heard about these crazy letters so that's just a mentor availability | 12:33 |
heiko | blackburn: also, there is no multiclass GP implementation out there that is easy to use | 12:33 |
heiko | the book matlab toolbox doesnt contain multiclass | 12:33 |
heiko | EP=expectation propagation | 12:33 |
blackburn | GP, EP, laplace, go figure :D | 12:33 |
heiko | variational optimisation | 12:33 |
blackburn | heiko: jk | 12:33 |
heiko | haha :) | 12:33 |
heiko | GP=Grand prix | 12:34 |
heiko | multiclass GP = mens/woman Grand prix | 12:34 |
blackburn | heiko: ooh that's interesting! | 12:34 |
blackburn | heiko: it is out of law to mix up men and women here though! | 12:34 |
heiko | in Russia? :) | 12:35 |
blackburn | heiko: well not really out of law but pretty pressing | 12:35 |
blackburn | heiko: I'd kill myself if I was a gay in russia :D | 12:35 |
heiko | blackburn: yeah, you dont like gay guys | 12:36 |
n4nd0 | any idea why trying to asign an numpy array of float64 to an SGVector<float64> fails in python_modular?? | 12:36 |
heiko | n4nd0: it might be the form | 12:36 |
blackburn | heiko: that's because we are barbarians living in 1600 :) | 12:36 |
n4nd0 | the error in particular is *** TypeError: in method 'ResultSet_psi_pred_set', argument 2 of type 'shogun::SGVector< float64_t > *' | 12:36 |
heiko | n4nd0: you need [ [row 1] [row 2]] | 12:36 |
blackburn | n4nd0: assign? | 12:36 |
n4nd0 | blackburn: = | 12:36 |
blackburn | n4nd0: that won't work | 12:36 |
n4nd0 | heiko: it is a vector, just one row, or? | 12:36 |
heiko | blackburn: Germany were also barbarians | 12:36 |
heiko | n4nd0: ah sorry | 12:37 |
blackburn | heiko: yes originally :D | 12:37 |
n4nd0 | blackburn: mmm why not? typemaps do not handle that? | 12:37 |
n4nd0 | heiko: no problem :) | 12:37 |
heiko | n4nd0: I had this problem quite often | 12:37 |
blackburn | n4nd0: no, they handle only methods | 12:37 |
blackburn | AFAIK | 12:37 |
n4nd0 | blackburn: aham I see | 12:37 |
heiko | n4nd0: sometimes the type has to be set explicitly with dtype="float64" | 12:37 |
heiko | I made this experience | 12:37 |
n4nd0 | heiko: yep, I tried doing that | 12:37 |
heiko | strange | 12:37 |
blackburn | heiko: I am afraid assigning won't work with types | 12:37 |
heiko | I am doing that all the type | 12:38 |
blackburn | heiko: I think you misunderstood something! :) | 12:38 |
blackburn | heiko: if x is sgvector | 12:38 |
blackburn | x = numpy.array([3,4.0]) | 12:38 |
blackburn | that won't work AFAIK | 12:38 |
heiko | ah I see | 12:38 |
blackburn | I mean it would work but x now is an array | 12:38 |
heiko | blackburn: I usually understand things wrong :D | 12:39 |
blackburn | heiko: set_pizza(numpy.array([3.0,4.0])) | 12:39 |
n4nd0 | blackburn: I probably need to put a setter or the like to make it work then | 12:39 |
blackburn | heiko: you meant that? | 12:39 |
heiko | blackburn: what I meant is I pass numpy arrays to methods that require an sgvector all the time | 12:39 |
n4nd0 | heiko: yeah | 12:39 |
blackburn | n4nd0: yes | 12:39 |
n4nd0 | I thought also that meant you could do = | 12:39 |
n4nd0 | but it seems not :) | 12:39 |
blackburn | n4nd0: you just assign it to new thing | 12:40 |
heiko | n4nd0: ah I see, so you have a sgvector and want to replace its content so that somebody else implicitly does something different | 12:40 |
n4nd0 | blackburn: the thing is since this is a field or an struct | 12:40 |
n4nd0 | it doesn't let me do it directly | 12:40 |
n4nd0 | field of* an struct | 12:41 |
blackburn | n4nd0: could you provide an example? | 12:41 |
heiko | blackburn: whats the desired field in the gsoc? | 12:42 |
n4nd0 | struct CResultSet { SGVector<float64_t> psi; } | 12:42 |
n4nd0 | from python I am doing | 12:42 |
blackburn | heiko: field? | 12:42 |
blackburn | heiko: ahh | 12:42 |
blackburn | heiko: I meant desired skills or so | 12:42 |
n4nd0 | result_set.psi = method_that_gives_numpy_array() | 12:42 |
blackburn | heiko: feel free to update it with something better | 12:42 |
blackburn | n4nd0: no, it needs a setter | 12:42 |
heiko | and the difference to prerequisites? | 12:43 |
n4nd0 | blackburn: all right, thanks! | 12:43 |
blackburn | heiko: prereqs are must, desired is would be better - I meant that | 12:43 |
n4nd0 | blackburn: hey | 13:38 |
blackburn | n4nd0: hey | 13:38 |
n4nd0 | what about a director class with a method returning a numpy array | 13:38 |
n4nd0 | where the return value of that method in C++ is defined to be an SGVector | 13:38 |
n4nd0 | it fails for some reason | 13:38 |
n4nd0 | TypeError: SWIG director type mismatch in output value of type 'shogun::SGVector< float64_t >' | 13:39 |
sonne|work | n4nd0: code signature? | 14:00 |
n4nd0 | virtual SGVector< float64_t > get_joint_feature_vector(int32_t feat_idx, CStructuredData* y); | 14:02 |
n4nd0 | sonne|work: ^ | 14:02 |
n4nd0 | I don't really get why the typemap doesn't handle this | 14:03 |
sonne|work | n4nd0: so you get back a SGVector? | 14:03 |
sonne|work | in python I mean? | 14:03 |
n4nd0 | sonne|work: yes, SGVector is the return value | 14:03 |
n4nd0 | sonne|work: yeah, I am doing a StructuredModel in python using directors | 14:03 |
n4nd0 | sonne|work: and I am writing this get_joint_feature_vector method in python | 14:04 |
sonne|work | n4nd0: well then use vec.get_vector() | 14:04 |
n4nd0 | sonne|work: in python? vec would be the numpy array? | 14:04 |
n4nd0 | mmm no, I guess no | 14:05 |
sonne|work | n4nd0: no vec would be SGVector | 14:05 |
sonne|work | and then vec.get() | 14:05 |
sonne|work | it will give you a numpy translated SGVector | 14:05 |
sonne|work | n4nd0: but note that you can do things like vec +=1 IIRC | 14:05 |
n4nd0 | sonne|work: but how can I get and SGVector from python?? | 14:06 |
n4nd0 | sonne|work: I mean, is it possible to create a new SGVector from python interface? | 14:09 |
sonne|work | sure | 14:10 |
sonne|work | RealVector() | 14:10 |
sonne|work | not typically used but I've seen heiko abusing it :) | 14:10 |
heiko | sonne|work: yep :) but not necessary actually | 14:11 |
n4nd0 | heiko, sonne|work : and how would you create the SGVector from the numpy array? | 14:21 |
n4nd0 | I'd rather avoid setting element by element | 14:22 |
n4nd0 | but I don't manage to get anything to work using the constructor | 14:22 |
sonne|work | heiko: btw http://www.linux-magazin.de/NEWS/Shogun-2.1.0-bringt-hunderte-Verbesserungen :) | 14:22 |
heiko | n4nd0: I dont understand, you can pase numpy arrays to functions that accept SGVectors right? | 14:24 |
sonne|work | n4nd0: looks like we don't have a convenience constructor for that yet - you could create a vector of appropriate size, zero it and += a numpy array to it | 14:24 |
heiko | sonne|work: haha | 14:25 |
heiko | great :) | 14:25 |
n4nd0 | heiko: for some reason the SGVector constructor that accepts another SGVector constructor doesn't work for this | 14:25 |
sonne|work | n4nd0: well it needs no const whatever crap | 14:26 |
n4nd0 | sonne|work: ok - out of curiosity, why the SGVector constructor that gets another SGVector doesn't work in this case? | 14:26 |
heiko | n4nd0: weird | 14:26 |
n4nd0 | sonne|work: aaah because of the const | 14:26 |
heiko | sonne|work: they also mentioned the c-base workshop! | 14:26 |
sonne|work | n4nd0: you need SGVector(SGVector vec); | 14:26 |
sonne|work | then it would work | 14:26 |
sonne|work | heiko: yeah cool isn't it? | 14:26 |
heiko | sonne|work: indeed cool | 14:27 |
heiko | sonne|work: did they mention us before? | 14:28 |
sonne|work | heiko: no idea - I accidentally stumbled upon this | 14:28 |
heiko | sonne|work: blackburn, I added my two project suggestions, feel free to feedback (will also continue to improve) | 14:29 |
heiko | time for lunch! | 14:29 |
-!- FSCV [~FSCV@65.19.131.247] has joined #shogun | 14:34 | |
n4nd0 | sonne|work: isn't weird that some_SGVector = some_numpy_array | 14:34 |
sonne|work | ? | 14:34 |
n4nd0 | doesn't work while it does work with += | 14:34 |
blackburn | heiko: cool! thanks | 14:34 |
sonne|work | n4nd0: just write the operator! | 14:35 |
n4nd0 | sonne|work: yeah yeah, I am doing it like that, I am just wondering :D | 14:35 |
blackburn | heiko: haha what are you waiting for, an einstein? | 14:35 |
sonne|work | n4nd0: and the constructor too please :) | 14:36 |
blackburn | n4nd0: no he meant add operator= and it could work | 14:36 |
blackburn | (maybe) | 14:36 |
n4nd0 | getting lost here :S | 14:36 |
blackburn | I am curious where can we find such students haha | 14:37 |
sonne|work | yes exactly n4nd0 please write the constructor and the operator= | 14:38 |
blackburn | Prerequisites: C++, (sparse) numerical linear algebra (positive definiteness, trace, Cholesky decomposition, Eigenvalues,iterative methods for linear systems (conjugate gradients), preconditioning), Basic probability theory: ( Gaussian distributions and manipulating them, expected value and variance), Graph colourings | 14:38 |
n4nd0 | blackburn: lol yeah, maybe some of those are more things the student will work with than actual reqs | 14:39 |
n4nd0 | at least IMO | 14:39 |
sonne|work | blackburn: would you qualify? | 14:43 |
blackburn | sonne|work: I am afraid no | 14:43 |
sonne|work | scary isn't it? | 14:43 |
blackburn | sonne|work: yeah so heiko needs someone more qualified than me and probably you :D | 14:44 |
n4nd0 | indeed | 14:44 |
n4nd0 | the scary I meant | 14:44 |
blackburn | sonne|work: well preconditioners and conj gradient are most specific - most know what it is but I can't say I deeply understand that | 14:45 |
sonne|work | blackburn: yeah an each set of algorithms in shogun is so specialized that there likely is just 1 shogun dev who *really* knows about it | 14:45 |
sonne|work | s/an// | 14:45 |
blackburn | sonne|work: yeah, I don't know about string kernels at all and you probably don't know how these tapkee impls work | 14:46 |
blackburn | that's normal though I think | 14:46 |
sonne|work | yup | 14:46 |
sonne|work | and when ICA enters shogun then we have something only andreas and 1 student know about | 14:47 |
blackburn | sonne|work: I was scaried with his idea | 14:47 |
sonne|work | it is somehow crucial that we keep having experts for each ML domain | 14:47 |
blackburn | sonne|work: it is just some domain people avoid :D | 14:48 |
sonne|work | and for sure tests to ensure that this stuff does sth | 14:48 |
blackburn | sonne|work: not really a lot of references for that | 14:48 |
sonne|work | blackburn: ? | 14:48 |
blackburn | sonne|work: ICA | 14:48 |
sonne|work | andreas' abstract was perfect | 14:48 |
sonne|work | wikipedia then | 14:48 |
blackburn | sonne|work: no I mean in machine learning ICA is not well acknowledged | 14:48 |
blackburn | and BSS problem at all | 14:49 |
sonne|work | http://en.wikipedia.org/wiki/Independent_component_analysis | 14:49 |
sonne|work | but in signal processing | 14:49 |
sonne|work | and even ICA has its own conference | 14:49 |
blackburn | sonne|work: it is close to matrix factorization and etc | 14:51 |
blackburn | sonne|work: so igor's ideas are close to | 14:52 |
blackburn | too* | 14:52 |
blackburn | sonne|work: I don't like any separation - math is the same | 14:52 |
blackburn | so that's quite nice we can unify things here in shogun | 14:53 |
sonne|work | thats true | 14:54 |
sonne|work | after all that is what is shogun for | 14:55 |
n4nd0 | my SSVM learning for grid CRFs is crap :D | 14:55 |
sonne|work | ? | 14:56 |
n4nd0 | sonne|work: I am implementing this in python using the DirectorStructuredModel | 14:57 |
n4nd0 | but the results are bad - I have to debug it | 14:57 |
n4nd0 | sonne|work: what I mean graphically, http://dl.dropbox.com/u/11020840/2013-03-19-150607_1600x900_scrot.png | 15:05 |
n4nd0 | top row is ground truth and bottom prediction | 15:06 |
sonne|work | n4nd0: not totally random though :) | 15:06 |
n4nd0 | sonne|work: hehe, I have (at least) a bug for sure | 15:06 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 15:24 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 15:59 | |
shogun-notifier- | shogun: iglesias :master * 8f1cfdd / src/shogun/structure/ (2 files): https://github.com/shogun-toolbox/shogun/commit/8f1cfdd90e694167e95feac14582d760ce8ea8a2 | 15:59 |
shogun-notifier- | shogun: some fixes for DirectorStructuredModel to make it work with PrimalMosekSOSVM | 15:59 |
shogun-notifier- | shogun: remove general delta_loss and get_joint_feature_vector that are in StructuredModel | 15:59 |
shogun-notifier- | shogun: add init_opt and check_training_setup (used by PrimalMosekSOSVM) | 15:59 |
shogun-notifier- | shogun: iglesias :master * 06df817 / src/shogun/structure/PrimalMosekSOSVM.cpp: https://github.com/shogun-toolbox/shogun/commit/06df817fa4d2ffcf749921bbdebe3e50bfe0ff5a | 15:59 |
shogun-notifier- | shogun: add debug messages in PrimalMosekSOSVM | 15:59 |
shogun-notifier- | shogun: iglesias :master * 6072d89 / src/shogun/features/ (3 files): https://github.com/shogun-toolbox/shogun/commit/6072d8967884a4018960e3d61f5c5b980d38cbfc | 15:59 |
shogun-notifier- | shogun: add obtain_from_generic method in MatrixFeatures | 15:59 |
shogun-notifier- | shogun: iglesias :master * 2fddcb7 / src/shogun/structure/StructuredModel.cpp,src/shogun/structure/StructuredModel.h: https://github.com/shogun-toolbox/shogun/commit/2fddcb7d9b5b6f3e17f48193c5412c1aba9e23bd | 15:59 |
shogun-notifier- | shogun: move implementation of CResultSet methods from header to cpp | 15:59 |
shogun-notifier- | shogun: Soeren Sonnenburg :master * 323cf50 / src/shogun/ (8 files): https://github.com/shogun-toolbox/shogun/commit/323cf5034fc76693c4935d9602ddf09b57c3511c | 15:59 |
shogun-notifier- | shogun: Merge pull request #945 from iglesias/structure | 15:59 |
shogun-notifier- | shogun: | 15:59 |
shogun-notifier- | shogun: Minor fixes for structured learning | 15:59 |
-!- blackburn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 16:07 | |
-!- flxb [~flxb@master.ml.tu-berlin.de] has joined #shogun | 16:21 | |
shogun-buildbot_ | build #952 of deb3 - modular_interfaces is complete: Failure [failed git] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/952 blamelist: Soeren Sonnenburg <sonne@debian.org>, iglesias <fernando.iglesiasg@gmail.com> | 16:24 |
-!- hoijui_ [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has quit [Ping timeout: 264 seconds] | 16:49 | |
heiko | n4nd0. write tests ! :) | 16:57 |
sonne|work | heiko: that was just doc updates error message fixes :) | 17:04 |
heiko | sonne|work: I meant the bugs on the results | 17:05 |
sonne|work | heiko: and n4nd0 ran away :-) | 17:05 |
sonne|work | ahh heh | 17:05 |
heiko | sonne|work: hehe :) as blackburn said | 17:05 |
heiko | I walk around with a gun ;D | 17:05 |
sonne|work | heiko: US would be the place to go then :D | 17:08 |
heiko | sonne|work: uuh, maybe I better relax again in this case ;) | 17:09 |
-!- zxtx [~zv@75-59-238-10.lightspeed.sntcca.sbcglobal.net] has quit [Ping timeout: 245 seconds] | 17:21 | |
shogun-buildbot_ | build #953 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/953 | 17:22 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 17:37 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Client Quit] | 17:39 | |
-!- hoijui_ [~hoijui@141.23.76.85] has joined #shogun | 17:40 | |
-!- hoijui_ [~hoijui@141.23.76.85] has quit [Ping timeout: 260 seconds] | 18:21 | |
shogun-notifier- | shogun: Soeren Sonnenburg :master * 3630c69 / src/shogun/kernel/Kernel.cpp: https://github.com/shogun-toolbox/shogun/commit/3630c69d73d674f48c773db36a4b55d88af4e16a | 18:33 |
shogun-notifier- | shogun: handle missing case in kernel | 18:33 |
-!- zxtx [zv@conference/pycon/x-ngldgkqwgolzbklc] has joined #shogun | 19:18 | |
shogun-notifier- | shogun: Soeren Sonnenburg :master * 5e619ed / tests/integration/python_modular/ (2 files): https://github.com/shogun-toolbox/shogun/commit/5e619ed4d21a14fcb65166d1b26471dece4a8e98 | 19:19 |
shogun-notifier- | shogun: blacklist a few failing integration tests and exit with exit code 1 if a test fails | 19:19 |
-!- zxtx [zv@conference/pycon/x-ngldgkqwgolzbklc] has quit [Ping timeout: 258 seconds] | 19:44 | |
-!- zxtx [zv@conference/pycon/x-xjwwzbcndhdisryg] has joined #shogun | 19:45 | |
-!- blackburn [~blackburn@188.168.3.40] has joined #shogun | 19:52 | |
heiko | blackburn: I found out how to store matrix factor for later use | 19:54 |
blackburn | heiko: insight me! | 19:54 |
heiko | x = L.triangularView<Lower>().solve(b); | 19:54 |
heiko | x = L.triangularView<Lower>().adjoint().solve(x); | 19:54 |
heiko | where L is the factor | 19:54 |
blackburn | heiko: ohh nice to know! | 19:55 |
heiko | simply do this "double backslash" solve that you might know from matlab | 19:55 |
heiko | quite cool | 19:55 |
blackburn | heiko: no matlab is a foreign thing for me :) | 19:55 |
blackburn | heiko: that's quite cool! | 19:55 |
heiko | blackburn: you do often | 19:56 |
heiko | L=chol(A) | 19:56 |
heiko | inv(A)*b equals to L\L'\b | 19:56 |
heiko | but whatever | 19:56 |
heiko | same idea | 19:56 |
heiko | trianular solver | 19:56 |
heiko | cool stuff this library | 19:56 |
heiko | and also actively maintained :) | 19:57 |
blackburn | heiko: hah much more better than blas thing | 19:57 |
heiko | thats a good trick if one has to solve multiple systems with the same matrix | 19:57 |
heiko | one can simply store the cholesky | 19:57 |
heiko | and then its very fast | 19:57 |
heiko | blackburn: yes much better | 19:57 |
heiko | I will put it into the GP project: replace all lapack/blas calls by eigen :) | 19:58 |
@sonney2k | apologies guys | 19:58 |
blackburn | sonney2k: ? | 19:58 |
-!- zxtx [zv@conference/pycon/x-xjwwzbcndhdisryg] has quit [Ping timeout: 264 seconds] | 19:58 | |
blackburn | heiko: eigen3 is well tested also - you may like it | 19:59 |
heiko | blackburn: I know, it passes the blas testsuite :) | 19:59 |
heiko | blackburn: thats crucial for these things | 19:59 |
heiko | imagine your solver returns nonsense :) | 20:00 |
heiko | will never find that bug | 20:00 |
heiko | blackburn: | 20:00 |
blackburn | friends presented me a toy helicopter I am playing around | 20:00 |
heiko | http://forum.kde.org/viewforum.php?f=74 | 20:00 |
heiko | thats a good ressource | 20:00 |
blackburn | heiko: yes, though sometimes I think they lack some doc | 20:00 |
heiko | indeed, I found the thing in the forum | 20:01 |
blackburn | heiko: btw I think we should use stackoverflow as that kind of forum | 20:01 |
heiko | how can we do that? | 20:02 |
blackburn | heiko: just subscribe to shogun tag at stackoverflow | 20:03 |
wiking | heiko: yo | 20:03 |
heiko | blackburn: good idea, we could also mention on the website | 20:03 |
heiko | wiking: hi | 20:04 |
heiko | was just about to leave :) | 20:04 |
wiking | ah ok | 20:04 |
wiking | talk to u then anoher time | 20:04 |
wiking | :> | 20:04 |
heiko | wiking: I will be here tonight or if not tomorrow | 20:04 |
heiko | but its already 7 and still at work, should go home :) | 20:04 |
heiko | see you later all | 20:05 |
wiking | laterz | 20:05 |
-!- heiko [~heiko@nat-187-2.internal.eduroam.ucl.ac.uk] has left #shogun [] | 20:05 | |
shogun-notifier- | shogun: Soeren Sonnenburg :master * d3b6183 / / (14 files): https://github.com/shogun-toolbox/shogun/commit/d3b61835423b86b7aa4c06dc12f75e6ab2b7f28e | 20:07 |
shogun-notifier- | shogun: rename Bmrm_return_value_T -> BmrmStatistics | 20:07 |
shogun-notifier- | shogun: Soeren Sonnenburg :master * 81a8ffb / src/Makefile,src/Makefile.template,src/configure: https://github.com/shogun-toolbox/shogun/commit/81a8ffb8803de92ea3f4b905e3dbb0a70df87026 | 20:07 |
shogun-notifier- | shogun: make failed examples actually fail things | 20:07 |
@sonney2k | blackburn, now you know | 20:07 |
@sonney2k | lots of fixing needed to get this stuff operational again | 20:08 |
shogun-buildbot_ | build #955 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/955 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:10 |
shogun-buildbot_ | build #661 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/661 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:14 |
-!- travis-ci [~travis-ci@ec2-184-73-6-22.compute-1.amazonaws.com] has joined #shogun | 20:17 | |
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/5634960 | 20:17 |
-!- travis-ci [~travis-ci@ec2-184-73-6-22.compute-1.amazonaws.com] has left #shogun [] | 20:17 | |
shogun-buildbot_ | build #374 of rpm1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/374 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:25 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 20:26 | |
shogun-buildbot_ | build #662 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/662 | 20:36 |
shogun-buildbot_ | build #956 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/956 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:55 |
shogun-buildbot_ | build #957 of deb3 - modular_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/957 blamelist: Soeren Sonnenburg <sonne@debian.org> | 21:10 |
-!- zxtx [zv@conference/pycon/x-jnkmsxrsyydeitia] has joined #shogun | 21:20 | |
-!- hoijui_ [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has joined #shogun | 21:39 | |
-!- zxtx [zv@conference/pycon/x-jnkmsxrsyydeitia] has quit [Ping timeout: 252 seconds] | 22:19 | |
-!- zxtx [zv@conference/pycon/x-gjvzfaqdorndenme] has joined #shogun | 22:24 | |
-!- zxtx [zv@conference/pycon/x-gjvzfaqdorndenme] has quit [Ping timeout: 258 seconds] | 22:53 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 245 seconds] | 23:05 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 23:05 | |
-!- hoijui_ [~hoijui@dslb-088-074-124-160.pools.arcor-ip.net] has quit [Quit: Leaving] | 23:07 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 23:07 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 23:21 | |
-!- zxtx [zv@conference/pycon/x-mkrdomcndvqaktzk] has joined #shogun | 23:27 | |
--- Log closed Wed Mar 20 00:00:15 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!