| --- Log opened Mon Jul 16 00:00:17 2012 | ||
| -!- heiko1 [~heiko@host86-181-80-163.range86-181.btcentralplus.com] has joined #shogun | 00:10 | |
| -!- heiko1 [~heiko@host86-181-80-163.range86-181.btcentralplus.com] has left #shogun [] | 00:11 | |
| -!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 01:53 | |
| -!- audy- is now known as audy | 02:10 | |
| -!- zxtx [~zv@ool-43559feb.dyn.optonline.net] has quit [Ping timeout: 246 seconds] | 06:29 | |
| gsomix | good morning | 06:41 |
|---|---|---|
| gsomix | sonney2k, inplace arithmetic operators works, I will send PR at evening. | 06:42 |
| -!- gsomix [~gsomix@188.168.5.232] has quit [Ping timeout: 265 seconds] | 07:33 | |
| -!- gsomix [~gsomix@r206-10.smr.ru] has joined #shogun | 08:39 | |
| -!- uricamic [~uricamic@2001:718:2:1634:f833:c09a:45c8:bbd3] has joined #shogun | 08:50 | |
| -!- uricamic [~uricamic@2001:718:2:1634:f833:c09a:45c8:bbd3] has quit [Ping timeout: 245 seconds] | 09:29 | |
| -!- uricamic [~uricamic@cmpgw-27.felk.cvut.cz] has joined #shogun | 09:30 | |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 09:54 | |
| yoo | hi | 09:55 |
| -!- heiko [~heiko@host86-181-80-163.range86-181.btcentralplus.com] has joined #shogun | 10:29 | |
| @sonney2k | gsomix, nice! | 11:17 |
| -!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 11:17 | |
| gsomix | aha, workin on surface triangulation now | 11:19 |
| @sonney2k | gsomix, booring | 11:28 |
| @sonney2k | gsomix, looking forward to your PR this evening | 11:28 |
| @sonney2k | heiko, can you have a look at puffins PR? | 11:28 |
| @sonney2k | heiko, ahh and one question - do we have serialization support for SGSparseMatrix ? | 11:29 |
| heiko | sonney2k, yes I will soon | 11:29 |
| @sonney2k | heiko, thanks | 11:29 |
| heiko | sonney2k, I think no | 11:29 |
| @sonney2k | heiko, and for SGStringList? | 11:29 |
| heiko | sonney2, I also just talked to this guy who found that x-val for multiclass machines doesnt work | 11:29 |
| heiko | sonney2k, not that I know of, I never touched it | 11:29 |
| @sonney2k | ok - so it is probably still the same code that we had with double* stuff right? | 11:30 |
| heiko | sorry, didnt get that | 11:30 |
| heiko | what do you mean now? | 11:30 |
| @sonney2k | heiko, before we had SGVector / SGMatrix | 11:31 |
| @sonney2k | we had double* int | 11:31 |
| heiko | ah I see | 11:31 |
| heiko | yes | 11:31 |
| heiko | same code | 11:31 |
| @sonney2k | k | 11:31 |
| heiko | actually, these vectors only register the array | 11:31 |
| heiko | the rest (the SGVector structure) stays with the class | 11:31 |
| heiko | so same thing as before SGVector for serialization | 11:32 |
| @sonney2k | so I guess the fix would be to do the same thing for SGSparseMatrix | 11:32 |
| @sonney2k | heiko, ohh one thing - what happens to the reference count of SGVector? | 11:32 |
| heiko | so far, nothing, its on my todo list :) | 11:32 |
| heiko | should just be svaed | 11:33 |
| heiko | saved | 11:33 |
| @sonney2k | that sounds like a crasher | 11:33 |
| @sonney2k | yeah | 11:33 |
| heiko | I think that should do it | 11:33 |
| @sonney2k | k | 11:33 |
| @sonney2k | heiko, I am asking because I am working on converting sparsefeatures -> SGSparseMatrix | 11:33 |
| heiko | I see | 11:34 |
| heiko | maybe this can be serialized | 11:34 |
| heiko | I just never used | 11:34 |
| heiko | was it possible before? | 11:34 |
| @sonney2k | I *think* so :D | 11:34 |
| heiko | hehe :D | 11:35 |
| @sonney2k | heiko, ok thanks sorry for the interruption | 11:36 |
| heiko | sonney2k, no worries | 11:36 |
| heiko | sonney2k, btw as I already said | 11:36 |
| heiko | there is this guy who found out that x-val doesnt work for anything but libsvm | 11:37 |
| heiko | is anybody aware of that? | 11:37 |
| heiko | I wasnt | 11:37 |
| @sonney2k | I never tried... | 11:40 |
| @sonney2k | heiko, ohh btw did you ask wrt. shogun toolbox foundation stuff? | 11:40 |
| @sonney2k | otherwise we might loose quite a bit of money. | 11:41 |
| heiko | oh, yes, I forgot that | 11:41 |
| heiko | I talked to a guy who does that for a plane-friend-organization | 11:41 |
| heiko | He told me the following: | 11:41 |
| heiko | its possible, however | 11:41 |
| heiko | its much work (we would have to do reports a few times a year, 4 or so) | 11:42 |
| heiko | the leaders are repsonsible for finanzial stuff and also liable in case of a crash or so | 11:42 |
| heiko | they had this example where they bought some ground to do stuff and the leaders took the wrong decision and then a few hundred thousand euros were gone | 11:43 |
| heiko | but thats probably not the case for us as we dont deal with any "risk" | 11:43 |
| @sonney2k | no money no risk :D | 11:43 |
| heiko | except for ML risks ;) | 11:43 |
| @sonney2k | heh | 11:43 |
| heiko | yeah, I mean its quite some formal organizational stuff that we would have to do. Everybody would have to get together to vote about the Satzung etc. | 11:44 |
| heiko | and the guy told me that usually these Amt people are not well aware of internet based stuff | 11:44 |
| heiko | but that might have changed | 11:45 |
| @sonney2k | heiko, if you can easily find out what we have to do then please do otherwise it is what it is | 11:45 |
| @sonney2k | heiko, well I would just do it via internet | 11:45 |
| @sonney2k | I don't see any other possible way | 11:45 |
| heiko | yeah, that was more of a general discussion about it with that friend | 11:45 |
| heiko | yes Internet is the way. | 11:46 |
| heiko | I will try to call somebody this week and ask more questions, kind of forgot about that in the last weeks | 11:46 |
| heiko | when is your tax thing due? | 11:46 |
| @sonney2k | heiko, btw blackburn is digging into the modsel stuff so he might know more what works ok what not :) | 11:46 |
| heiko | sonney2k, good to hear | 11:47 |
| @sonney2k | heiko, I don't know - I am usually slow with this stuff | 11:47 |
| @sonney2k | heiko, and I actually would like gsomix to start with the new modsel typemaps rather soon (next week latest) so we should find out too :) | 11:48 |
| shogun-buildbot | build #151 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/151 | 11:48 |
| heiko | yes, I will investigate this multiclass problem later today | 11:48 |
| blackburn | multiclass is not modsel issue | 11:48 |
| blackburn | it is just multiple train call issue | 11:49 |
| @sonney2k | heiko, you heard the master :D | 11:49 |
| heiko | blackburn, yes, did you already check that out? | 11:49 |
| heiko | sonney2k, I am feeling spied now :D | 11:49 |
| blackburn | well what I know larank fails to work with modsel | 11:49 |
| blackburn | not because of modsel | 11:49 |
| heiko | blackburn, GMNPSVM fails with cross-validation evaluation already | 11:50 |
| blackburn | I am pretty sure the same reason | 11:50 |
| heiko | ok | 11:50 |
| heiko | what exactly is the reason? multiple training? whats that? shouldnt be bad | 11:51 |
| blackburn | heiko: train() train() yeah | 11:53 |
| heiko | why is that bad? | 11:53 |
| heiko | should be slow but not give false results? | 11:53 |
| blackburn | larank doesn't clean up some stuff | 11:54 |
| heiko | a ok | 11:54 |
| blackburn | so kaboooms | 11:54 |
| blackburn | heiko: did you like my modseloutput stuff? | 11:54 |
| heiko | blackburn, where? | 11:54 |
| heiko | the debug messages? | 11:54 |
| blackburn | not really - I added a class | 11:54 |
| heiko | havent seen yet | 11:55 |
| heiko | what does it do? | 11:55 |
| blackburn | modelselection/ModelSelectionOutput.h | 11:55 |
| blackburn | well just outputs test/train indices/results, evaluation result | 11:55 |
| heiko | ah cool | 11:55 |
| heiko | yes I will probably like it | 11:55 |
| blackburn | ah and for linear machine it outputs W | 11:55 |
| blackburn | next step would be to set up different output ways | 11:56 |
| blackburn | now it just prints | 11:56 |
| heiko | cool | 11:56 |
| heiko | good idea | 11:56 |
| blackburn | heiko: have you seen some eigen3 code is in? | 12:00 |
| heiko | I noticed it | 12:00 |
| heiko | whats it for? | 12:00 |
| blackburn | I bet you would like it | 12:01 |
| blackburn | well blas without 75878 parameters :) | 12:01 |
| blackburn | and faster | 12:01 |
| heiko | ah | 12:02 |
| heiko | nice :) | 12:02 |
| heiko | thats something I like :) | 12:02 |
| blackburn | heiko: trace(A*B') is (A*B.transpose()).trace() | 12:03 |
| heiko | ehm? yes? | 12:03 |
| heiko | so thats how you use it or what? | 12:03 |
| heiko | our SGMatrices got these operators now? | 12:04 |
| blackburn | no | 12:04 |
| blackburn | heiko: well that wouldn't work with sgmatrices anyway | 12:04 |
| heiko | what was that then? | 12:05 |
| blackburn | heiko: they have MatrixXd for that | 12:05 |
| heiko | ah ok | 12:05 |
| blackburn | however | 12:05 |
| blackburn | if you want to use some preallocated memory | 12:05 |
| blackburn | you may use Map<MatrixXd> stuff | 12:05 |
| blackburn | this way you can modify sgmatrix's data | 12:06 |
| heiko | I am a bit lost now | 12:06 |
| heiko | what can one do nwo? | 12:07 |
| heiko | now | 12:07 |
| blackburn | heiko: well you may use eigen3 both for sgmatrices and some internal MatrixXd matrices | 12:07 |
| heiko | ah now I see | 12:07 |
| blackburn | heiko: http://latex.codecogs.com/gif.latex?\min_\beta%20\frac{1}{2}%20\beta^T%20K%20\beta%20-%20\kappa%20\beta%20\\%20\beta_i%20\in%20[0,B],%20~~~%20\left|\sum_i%20\beta_i%20-%20n_s\right|%20\leqslant%20n_s%20\varepsilon | 12:23 |
| blackburn | any idea how to approach this problem? | 12:23 |
| blackburn | uricamic: may be you? :) | 12:25 |
| heiko | blackburn, I cannot read that | 12:30 |
| blackburn | heiko: everything except beta is given | 12:30 |
| heiko | no the font | 12:30 |
| blackburn | ehm | 12:30 |
| blackburn | again? :D | 12:30 |
| heiko | yeah | 12:30 |
| heiko | black on black | 12:30 |
| blackburn | ok will try different latex generator | 12:30 |
| blackburn | heiko: how about that? http://www.numberempire.com/equation.render?\min_\beta%20\frac{1}{2}%20\beta^T%20K%20\beta%20-%20\kappa%20\beta\\\beta_i%20\in%20[0,B],%20~~~%20\left|\sum_i%20\beta_i%20-%20n_s\right|%20\leqslant%20n_s%20\varepsilon | 12:31 |
| heiko | same thing | 12:31 |
| heiko | ust send the tex | 12:31 |
| heiko | Ill insert it myself | 12:31 |
| heiko | ah i have it | 12:32 |
| blackburn | heiko: \min_\beta \frac{1}{2} \beta^T K \beta - \kappa \beta | 12:32 |
| blackburn | \\ | 12:32 |
| blackburn | \beta_i \in [0,B], ~~~ \left|\sum_i \beta_i - n_s\right| \leqslant n_s \varepsilon | 12:32 |
| heiko | without this abs value thing it would be easy right? | 12:32 |
| blackburn | yeah | 12:32 |
| heiko | this looks a bit like a L1 norm regularizer, isnt there a lasso method that does a similar thing? | 12:33 |
| heiko | blackburn, otherwise. no idea, never saw one like this :) | 12:33 |
| blackburn | heiko: I don't know actually.. | 12:34 |
| blackburn | it is a KMM | 12:34 |
| blackburn | kernel mean matching | 12:34 |
| heiko | mmh | 12:34 |
| blackburn | some stuff by your supervisor btw :) | 12:34 |
| heiko | really? | 12:35 |
| heiko | we compute the distance between the means | 12:35 |
| heiko | we dont match them :) | 12:35 |
| blackburn | http://listengine.tuxfamily.org/lists.tuxfamily.org/eigen/2012/07/msg00101.html | 12:36 |
| blackburn | hehe | 12:36 |
| blackburn | heiko: http://www.kyb.mpg.de/fileadmin/user_upload/files/publications/attachments/shift-book-for-LeEtAl-webversion_5376%5b0%5d.pdf | 12:37 |
| heiko | oh well | 12:38 |
| heiko | at least not me :) | 12:38 |
| blackburn | v-trick | 12:42 |
| blackburn | hmm what is v-trick | 12:42 |
| blackburn | or rather \nu-trick | 12:43 |
| -!- Netsplit *.net <-> *.split quits: wiking_ | 13:43 | |
| -!- Netsplit over, joins: wiking_ | 13:44 | |
| blackburn | [DEBUG] 201 iteration passed, objective = 301058401603326972979179723582841701075978963558336672155439500039329112745427114261305300261979153027907701835932192738286119401824164274262496646950403644063744.000000 | 14:18 |
| blackburn | ROFL | 14:18 |
| heiko | hehe :) | 14:19 |
| CIA-18 | shogun: Sergey Lisitsyn master * r7b9ac25 / (8 files in 4 dirs): Added Multitask Trace Logistic Regression - http://git.io/r8n9Ww | 14:24 |
| blackburn | argh | 14:24 |
| CIA-18 | shogun: Sergey Lisitsyn master * r4f42e08 / src/shogun/mathematics/Math.h : Guarded eigen3 include - http://git.io/XRlOLA | 14:25 |
| shogun-buildbot | build #141 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/141 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 14:28 |
| shogun-buildbot | build #142 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/142 | 14:30 |
| -!- os252 [~stegle@as-net.tuebingen.mpg.de] has joined #shogun | 14:31 | |
| -!- os252 [~stegle@as-net.tuebingen.mpg.de] has quit [Client Quit] | 14:32 | |
| -!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 14:35 | |
| -!- gsomix [~gsomix@r206-10.smr.ru] has quit [Quit: Ex-Chat] | 15:05 | |
| -!- heiko [~heiko@host86-181-80-163.range86-181.btcentralplus.com] has quit [Quit: Leaving.] | 15:18 | |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has left #shogun [] | 15:22 | |
| -!- wiking_ is now known as wiking | 15:22 | |
| -!- wiking [~wiking@info2k1.hu] has quit [Changing host] | 15:23 | |
| -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 15:23 | |
| -!- zxtx [~zv@ool-457e7550.dyn.optonline.net] has joined #shogun | 15:58 | |
| -!- zxtx [~zv@ool-457e7550.dyn.optonline.net] has quit [Ping timeout: 246 seconds] | 16:17 | |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 17:03 | |
| -!- zxtx [~zv@ool-457e7550.dyn.optonline.net] has joined #shogun | 17:03 | |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed] | 17:35 | |
| -!- zxtx [~zv@ool-457e7550.dyn.optonline.net] has quit [Ping timeout: 246 seconds] | 17:59 | |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 18:00 | |
| -!- gsomix [~gsomix@188.168.5.70] has joined #shogun | 18:23 | |
| gsomix | good evening | 18:24 |
| -!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed] | 18:25 | |
| -!- uricamic [~uricamic@cmpgw-27.felk.cvut.cz] has quit [Quit: Leaving.] | 18:33 | |
| @sonney2k | gsomix, hey there? doing shogun stuff now? | 18:53 |
| -!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 19:18 | |
| gsomix | sonney2k, yes, of course | 19:25 |
| -!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 19:30 | |
| -!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 19:31 | |
| -!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 19:37 | |
| -!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 19:38 | |
| gsomix | sonney2k, ta-dam | 19:44 |
| -!- heiko [~heiko@host86-185-15-105.range86-185.btcentralplus.com] has joined #shogun | 19:55 | |
| @sonney2k | gsomix, ta-dam? ka-boom? crashing? | 19:56 |
| gsomix | there is "da bum tss" for crashing. | 19:56 |
| @sonney2k | gsomix, adn ta-dam? | 19:56 |
| gsomix | for success | 19:57 |
| @sonney2k | heh | 19:57 |
| gsomix | ok, need some defines. | 19:57 |
| gsomix | and PR. | 19:57 |
| heiko | blackburn, around? | 20:04 |
| blackburn | yeah | 20:04 |
| heiko | I wanted to ask about that multiple train() bug again | 20:04 |
| blackburn | aham | 20:04 |
| heiko | we need to train multiple times while cv | 20:04 |
| blackburn | exactly | 20:04 |
| heiko | so how did you resolve the other one? | 20:04 |
| blackburn | I didn't | 20:04 |
| blackburn | it is not fixed yet | 20:05 |
| heiko | ah ok | 20:05 |
| heiko | I will check whether its really the resion | 20:05 |
| heiko | reason | 20:05 |
| blackburn | heiko: just add double train call in | 20:05 |
| blackburn | classifier_larank_modular | 20:05 |
| heiko | blackburn, just did :) | 20:05 |
| blackburn | gmnp works though | 20:06 |
| heiko | mmh | 20:07 |
| heiko | yes | 20:07 |
| heiko | but I train on same features | 20:07 |
| heiko | Ill train on different and then on original again | 20:07 |
| heiko | no its fine | 20:08 |
| heiko | so thats not the problem for gmnp | 20:08 |
| heiko | blackburn, any other suspicions? | 20:09 |
| blackburn | heiko: no, not really | 20:09 |
| blackburn | we need to check what fails | 20:09 |
| blackburn | and what works | 20:09 |
| -!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun | 20:09 | |
| KMcQuisten | Hello all | 20:09 |
| blackburn | it's been a long, KMcQuisten | 20:10 |
| blackburn | how are you? | 20:10 |
| blackburn | heiko: I will take care of larank | 20:10 |
| KMcQuisten | I'm doing ok, but I'm having some troubles. Can't even get a simple 2kernel MKL to train. | 20:10 |
| heiko | blackburn, ok, wait a bit | 20:10 |
| heiko | perhaps the fault is in the CV | 20:10 |
| blackburn | do you think so? | 20:10 |
| heiko | blackburn, you mean me? | 20:10 |
| blackburn | heiko: what is wrong in CV? | 20:11 |
| heiko | if so: I think the recent multiclass-change might have caused problems | 20:11 |
| heiko | I remember fixing something when I did my multi-class bioinfo coursework on christmas | 20:11 |
| blackburn | hmm | 20:11 |
| heiko | but I only did this for libsvm | 20:11 |
| heiko | remember? | 20:11 |
| blackburn | not really | 20:11 |
| blackburn | KMcQuisten: what is the problem? | 20:11 |
| heiko | labels were not set correctly to all submachine or something | 20:11 |
| KMcQuisten | blackburn: Same kind of thing as before. Using multiple string kenrels in MKLregression, setup up the proper features and kernels, then when I train, sometimes it will train, and sometimes I get: | 20:13 |
| KMcQuisten | SystemError: [ERROR] Assertion R >= 0 failed! | 20:13 |
| KMcQuisten | And it seems random | 20:13 |
| KMcQuisten | i think we dealt with this once before, but it's become unfixed, as it were | 20:14 |
| KMcQuisten | I have a test script I can email you if you want to reproduce | 20:14 |
| blackburn | KMcQuisten: sure | 20:15 |
| KMcQuisten | Sent | 20:19 |
| KMcQuisten | Did that make it through to you? | 20:25 |
| blackburn | KMcQuisten: yeah fails here too | 20:25 |
| blackburn | checking code | 20:25 |
| KMcQuisten | Righto | 20:25 |
| blackburn | KMcQuisten: for now you may disable linadd again | 20:26 |
| KMcQuisten | Ah, yes | 20:26 |
| blackburn | slower but seems to work | 20:26 |
| KMcQuisten | Sounds like a plan for now | 20:27 |
| -!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Ping timeout: 245 seconds] | 20:31 | |
| -!- audy [~audy@heyaudy.com] has quit [Changing host] | 20:41 | |
| -!- audy [~audy@unaffiliated/audy] has joined #shogun | 20:41 | |
| -!- os253 [~stegle@as-net.tuebingen.mpg.de] has joined #shogun | 20:43 | |
| blackburn | hey os253 | 20:44 |
| blackburn | heiko: any thoughts? | 20:44 |
| os253 | hey Blackburn | 20:44 |
| blackburn | os253: do you know kernel mean matching method? | 20:45 |
| blackburn | I ask everyone today :D | 20:45 |
| wiking | boooyaaa | 20:46 |
| blackburn | awakening of wiking | 20:46 |
| wiking | heheh yeps | 20:46 |
| os253 | I have compared to this at some point: it's very simple and works ok. | 20:46 |
| os253 | http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.165.8929 | 20:46 |
| wiking | blackburn: new pr | 20:46 |
| wiking | ! | 20:46 |
| wiking | so take it | 20:46 |
| wiking | oh no just a sec to pull it | 20:47 |
| wiking | blackburn: btw have u had time to check on larank and cv? | 20:47 |
| blackburn | os253: I'd like to implement it but stucked with formulation | 20:47 |
| blackburn | os253: can you help me there? | 20:47 |
| @sonney2k | blackburn, heiko is the expert for that isn't he? | 20:47 |
| wiking | blackburn: and do you have a hog implementation at hand? | 20:47 |
| blackburn | wiking: I have python hog implementation only actually | 20:47 |
| wiking | blackburn: what are the dependencies? | 20:47 |
| wiking | apart from python ;) | 20:47 |
| os253 | blackburn: with the exact identical method? | 20:48 |
| os253 | I.e. you need an implementation of this? | 20:48 |
| blackburn | os253: well I am confused with |\sum \beta_i - n_tr | < e n_tr | 20:48 |
| wiking | sonney2k: i think u could run libshogun test on freebsd bot | 20:48 |
| blackburn | os253: how one can handle this? | 20:48 |
| blackburn | wiking: I didn't get to larank yet | 20:48 |
| wiking | sonney2k: i've installed some extra dependencies on that machine for shogun | 20:48 |
| wiking | blackburn: link for hog? | 20:49 |
| blackburn | wiking: I'll send you a link in a sec | 20:49 |
| wiking | blackburn: cool thnx | 20:49 |
| blackburn | wiking: I used this one http://code.google.com/p/python-cvgreyc/source/browse/trunk/cvgreyc/features/HOG.py?r=83 | 20:49 |
| os253 | To be honest I looked at this years ago and compared to it only. The implementation was from elephant (an ML toolbox), I think. Maybe I also coded it up, I can take a look. | 20:49 |
| blackburn | os253: formulation is | 20:50 |
| blackburn | http://latex.codecogs.com/gif.latex?\min_\beta%20\frac{1}{2}%20\beta^T%20K%20\beta%20-%20\kappa%20\beta%20\\%20\beta_i%20\in%20[0,B],%20~~~%20\left|\sum_i%20\beta_i%20-%20n_s\right|%20\leqslant%20n_s%20\varepsilon | 20:50 |
| wiking | blackburn: thnx i'll give it a go | 20:50 |
| blackburn | sonney2k: may be you can help too? | 20:50 |
| blackburn | :) | 20:50 |
| os253 | get's scrambled in this IRC client. Give me a bit, need to finish 1,2 things and then will take a look. | 20:51 |
| blackburn | ok sure | 20:51 |
| gsomix | sonney2k, PR | 21:04 |
| blackburn | wiking: I can't see any PR :) | 21:07 |
| * wiking mumblesbadshit | 21:09 | |
| -!- yoo [575b08cb@gateway/web/freenode/ip.87.91.8.203] has joined #shogun | 21:15 | |
| yoo | good evening there | 21:22 |
| blackburn | evening | 21:22 |
| @sonney2k | gsomix, I don't get it | 21:24 |
| heiko | yoo, still looking at the problem | 21:25 |
| @sonney2k | gsomix, are these operators for += ? | 21:25 |
| heiko | quick fix didnt work ;) | 21:25 |
| @sonney2k | etc | 21:25 |
| yoo | heiko: brb | 21:25 |
| gsomix | sonney2k, +=, *=, -=. more operators are needed? | 21:26 |
| @sonney2k | gsomix, but isn't this doing x+=y where x are CDenseFeatures and y is a numpy matrix? | 21:26 |
| @sonney2k | gsomix, why don't you write instead of | 21:28 |
| @sonney2k | result_mat[num_feat*i + j] = lhs[num_feat*i + j] ## operator ## buf[num_feat*i + j] | 21:28 |
| @sonney2k | lhs[num_feat*i + j] ## operator ## buf[num_feat*i + j] | 21:28 |
| @sonney2k | and operator then is += etc? | 21:28 |
| @sonney2k | instead of creating a new one? | 21:29 |
| @sonney2k | gsomix, ^ | 21:29 |
| gsomix | sonney2k, what do you mean? it works for DenseFeature+(*, -)=DenseFeatures and for DenseFeatures+(*,-)=numpy.array | 21:30 |
| gsomix | sonney2k, oops, my fault | 21:30 |
| gsomix | one moment | 21:30 |
| blackburn | gsomix: but you treat it inplace | 21:30 |
| blackburn | :) | 21:30 |
| heiko | blackbrun, is there a test for the store_model_features of multi class kernel machine? | 21:30 |
| blackburn | heiko: nein | 21:30 |
| blackburn | it is not even implemented | 21:31 |
| blackburn | I am afraid | 21:31 |
| heiko | it is | 21:31 |
| heiko | I bet this is causing the errors | 21:31 |
| heiko | wrong sv indices | 21:31 |
| blackburn | heiko: ohh | 21:32 |
| blackburn | that can be | 21:32 |
| wiking | heiko: that has been implemented in a rush afaik | 21:32 |
| heiko | Chiyuan Zhang wrote it | 21:32 |
| blackburn | heiko: ok this describes gmnp failure | 21:33 |
| blackburn | but larank still to fix | 21:33 |
| heiko | The only thing is that why works it for libsvm multiclass | 21:33 |
| blackburn | interesting | 21:33 |
| wiking | heiko: when the new machines were introduced somebody realised that cv doesnt work for kernelmachines | 21:33 |
| wiking | i mean kernel multiclass machines | 21:33 |
| yoo | heiko: one thg I havnt checked is the sv indices. I m double checking. | 21:33 |
| wiking | and then it got implemented fast... | 21:33 |
| wiking | that was like 2-3 months ago | 21:33 |
| wiking | rather 3 | 21:33 |
| heiko | mmmh | 21:34 |
| wiking | heiko: you can check the correspoinding chat logs for the git commit date... | 21:34 |
| heiko | we need more tests ... | 21:34 |
| heiko | this should be captured automatically | 21:34 |
| heiko | but it works for libsvm so I will search for the differences | 21:35 |
| wiking | heiko: CMulticlassLibLinear? | 21:37 |
| heiko | wiking, have a look at MultiClassLibSVM | 21:37 |
| heiko | .cpp | 21:37 |
| heiko | line 197 | 21:37 |
| heiko | there I wrote a fix for this problem | 21:37 |
| heiko | which is rather unefficient, but works | 21:37 |
| heiko | I think this would be needed for the other multi class svms too | 21:37 |
| gsomix | compling... | 21:38 |
| heiko | its almost the only difference between working and not-working implementations | 21:38 |
| heiko | and whats this? | 21:39 |
| heiko | svm_svs()=SGVector<index_t>(m_kernel->get_num_vec_lhs()); | 21:39 |
| -!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 21:39 | |
| wiking | dunno | 21:40 |
| -!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 21:43 | |
| heiko | yoo, its a bit more complicated, Ill have to dive a bit into it so it may take some time ... | 21:46 |
| yoo | heiko: ok. guide me to help if needed. what type of fix did u do for MultiClassLibSVM ? | 21:48 |
| heiko | yoo, there was a simila problem and I just put in a hack where I just stored *all* SV | 21:49 |
| heiko | so the problem is probably related to wrongly set SVs | 21:49 |
| @sonney2k | wiking, can you test whether make tests works on your buildslave? | 21:49 |
| heiko | but I currently cannot oversee that all | 21:50 |
| heiko | has to be debugged step by step unfortunately | 21:50 |
| wiking | sonney2k: sec | 21:50 |
| @sonney2k | wiking, to do so just go into the build/src directory | 21:50 |
| heiko | which is a lot of work | 21:50 |
| @sonney2k | then run make install | 21:50 |
| @sonney2k | make tests | 21:50 |
| yoo | heiko: arf | 21:50 |
| heiko | yoo, yes annoying | 21:51 |
| wiking | sonney2k: ah usual stuff: #!/bin/bash | 21:52 |
| wiking | sonney2k: in check.sh | 21:52 |
| wiking | sonney2k: should i replace it or will you to #!/usr/bin/env bash | 21:53 |
| wiking | ? | 21:53 |
| yoo | heiko: I will then check the mailing list to be aware of any fix for this bug. | 21:53 |
| yoo | thx | 21:53 |
| heiko | yoo, I fear there currently is not | 21:54 |
| heiko | is has to be "really" fixed | 21:54 |
| @sonney2k | wiking, I am doing... but just replace it locally and try to get it to work... | 21:54 |
| @sonney2k | maybe there is more | 21:55 |
| wiking | ok trying | 21:55 |
| wiking | Error expanding embedded variable. | 22:01 |
| wiking | wht is this? | 22:01 |
| wiking | ah ok i know | 22:02 |
| wiking | so the problem is that ./check.sh calls make | 22:02 |
| wiking | but it should call gmake on freebsd | 22:02 |
| wiking | will replace it locally just to test it | 22:02 |
| wiking | ok that solved the problem | 22:04 |
| wiking | woah but there's a problem with compilation | 22:04 |
| wiking | //usr/home/wiking/shogun-buildbot/bsd1_-_libshogun/build/src/build/lib/libshogun.so: undefined reference to `std::__detail::_List_node_base::_M_hook(std::__detail::_List_node_base*)@GLIBCXX_3.4.15' | 22:04 |
| CIA-18 | shogun: Soeren Sonnenburg master * r2f6ce63 / (25 files in 24 dirs): replace #!/bin/bash with #!/usr/bin/env bash - http://git.io/lxgOMQ | 22:04 |
| wiking | //usr/home/wiking/shogun-buildbot/bsd1_-_libshogun/build/src/build/lib/libshogun.so: undefined reference to `std::ctype<char>::_M_widen_init() const@GLIBCXX_3.4.11' | 22:05 |
| yoo | heiko: need to leave. thks for the info. its too bad that the easy fix didnt work but the bug is identified now. hope some1 is going to work on this (?) cya. | 22:05 |
| heiko | yoo, I will work on it this week | 22:06 |
| heiko | hope to fix it in a few days | 22:06 |
| yoo | heiko: great I ll come back to support ^^ ++ | 22:06 |
| -!- yoo [575b08cb@gateway/web/freenode/ip.87.91.8.203] has quit [Quit: Page closed] | 22:06 | |
| @sonney2k | wiking, wait your gcc is 3.X sth? | 22:07 |
| gsomix | sonney2k, fixed | 22:08 |
| wiking | sonney2k: noup.. i think it's more that it tries to use the wrong lib | 22:08 |
| @sonney2k | gsomix, looking | 22:08 |
| @sonney2k | wiking, do you run make check-examples from the build/src dir? | 22:10 |
| wiking | sonney2k: gmake tests | 22:12 |
| wiking | i did this | 22:12 |
| CIA-18 | shogun: Evgeniy Andreev master * rb5127eb / (2 files in 2 dirs): inplace arithmetic operators for dense features - http://git.io/Av3-fQ | 22:13 |
| CIA-18 | shogun: Soeren Sonnenburg master * rdb85eb2 / (2 files in 2 dirs): Merge pull request #650 from gsomix/buffer_protocol - http://git.io/aTWcrA | 22:13 |
| @sonney2k | wiking, yeah that is the same | 22:13 |
| @sonney2k | gsomix, alright - could you then tomorrow do slicing support? | 22:14 |
| wiking | sonney2k: apparently not ;) | 22:14 |
| wiking | i mean make != gmake on bsd | 22:14 |
| gsomix | sonney2k, ok | 22:14 |
| gsomix | sonney2k, tnx for review | 22:15 |
| @sonney2k | gsomix, thanks for the code :) | 22:17 |
| @sonney2k | btw wiking, gsomix did you send your weekly report yet? | 22:17 |
| gsomix | sonney2k, I'm writing now. | 22:17 |
| wiking | sonney2k: so there should be a variable in the check.sh for make | 22:17 |
| wiking | or seomthing | 22:17 |
| @sonney2k | wiking, sure but you can patch this locally for now | 22:19 |
| wiking | sonney2k: i did it | 22:19 |
| CIA-18 | shogun: Sergey Lisitsyn master * r6b69c17 / src/shogun/multiclass/LaRank.cpp : Fixed larank multiple training issue - http://git.io/je1UZA | 22:19 |
| @sonney2k | wiking, I am rather interested in understanding why you get such weird erros | 22:20 |
| wiking | i just need to find outnow about those build errors | 22:20 |
| @sonney2k | wiking, grep M_widen_init /usr/lib/x86_64-linux-gnu/libs* -r | 22:21 |
| @sonney2k | Binary file /usr/lib/x86_64-linux-gnu/libstdc++.so.6 matches | 22:21 |
| -!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun | 22:33 | |
| CIA-18 | shogun: Soeren Sonnenburg master * r98f4788 / (2 files in 2 dirs): call $MAKE from configure... - http://git.io/cwgqMA | 22:46 |
| @sonney2k | wiking, that ^ could do it... please test | 22:47 |
| @sonney2k | wiking, and again please don't forget about your weekly report | 22:47 |
| wiking | sonney2k: for both: yep just doing them! | 22:51 |
| @sonney2k | wiking, thanks... | 22:51 |
| wiking | sonney2k: just had to organize soemthing about my wedding... | 22:51 |
| wiking | crazy times | 22:51 |
| blackburn | wedding? | 22:52 |
| blackburn | :) | 22:52 |
| wiking | blackburn: yes :) | 22:52 |
| blackburn | my brother is going to have wedding on 27th Jul | 22:52 |
| wiking | congrtz | 22:52 |
| gsomix | good night guys | 22:54 |
| heiko | blackburn, my sister maried a few weeks ago :) | 22:56 |
| blackburn | chain | 22:56 |
| wiking | sonney2k: here? | 23:09 |
| shogun-buildbot | build #154 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/154 blamelist: Evgeniy Andreev <gsomix@gmail.com> | 23:10 |
| wiking | sonney2k: the problem about the compiling: the c++ has not been set in the right way... | 23:10 |
| wiking | sonney2k: the configure script doesn't set it | 23:10 |
| wiking | sonney2k: in examples/undocumented/libshogun/Makefile the CC variable should be set by ./configure script | 23:11 |
| wiking | as now it's always CC=c++ which in this box will fail as the whole shogun has been built with c++46 | 23:11 |
| wiking | which is set by the configure script | 23:12 |
| wiking | sonney2k: if i replace CC=c++46 in that Makefile then it compiles | 23:12 |
| -!- heiko1 [~heiko@host86-181-81-54.range86-181.btcentralplus.com] has joined #shogun | 23:16 | |
| wiking | sonney2k: ok almost... there's still some weird errors... just trying with a clean build... | 23:17 |
| -!- heiko [~heiko@host86-185-15-105.range86-185.btcentralplus.com] has quit [Ping timeout: 265 seconds] | 23:17 | |
| wiking | ok this part i dont get yet: | 23:22 |
| wiking | running basic_minimal .. ERROR | 23:22 |
| wiking | /usr/lib/libstdc++.so.6: version GLIBCXX_3.4.11 required by //usr/home/wiking/shogun-buildbot/bsd1_-_libshogun/build/src/build/lib/libshogun.so.12 not found | 23:22 |
| wiking | ok i think there's a problem with linking | 23:25 |
| wiking | since it's actually according to ldd libshogun.so.12 is requiring /usr/lib/libstdc++.so.6 but it should use /usr/local/lib/gcc46/libstdc++.so.6 instead since that's the one for c++46 on my system... | 23:26 |
| shogun-buildbot | build #155 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/155 blamelist: Soeren Sonnenburg <sonne@debian.org>, Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:33 |
| shogun-buildbot | build #136 of deb2 - static_interfaces is complete: Failure [failed test octave_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/136 blamelist: Evgeniy Andreev <gsomix@gmail.com>, Soeren Sonnenburg <sonne@debian.org>, Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:36 |
| -!- rieck [~rieck@paul.ml.tu-berlin.de] has joined #shogun | 23:37 | |
| rieck | moin | 23:37 |
| rieck | anyone home | 23:37 |
| blackburn | hey | 23:37 |
| rieck | sonney2k: still alive? | 23:38 |
| rieck | that's a pitty, got an IM 40 mins ago | 23:39 |
| shogun-buildbot | build #137 of deb2 - static_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/137 | 23:43 |
| rieck | blackburn: still working on shogun extensions? | 23:44 |
| blackburn | rieck: yeah breaking shogun still ;) | 23:44 |
| rieck | blackburn: did gsoc work out last year? did you have fun and get some money? | 23:45 |
| blackburn | rieck: yes, I am participating this year too | 23:45 |
| blackburn | rieck: this year we've got 8 students | 23:46 |
| wiking | bzzzz | 23:48 |
| * wiking does not understand freebsd's moves in this linking... | 23:48 | |
| rieck | cool. | 23:49 |
| rieck | support from google is clearly a boost for shogun | 23:49 |
| rieck | it's also good to see people working again in gsoc | 23:49 |
| blackburn | rieck: yeah things have changed here | 23:50 |
| rieck | i am tired. i have to get some sleep, tomorrow is a written exam | 23:51 |
| blackburn | not for you I guess :D | 23:51 |
| rieck | no. its my students and my exam. still, i need to be somewhat prepared and awake, otherwise it would not be fair | 23:52 |
| rieck | nite | 23:52 |
| blackburn | heh good luck to you students then | 23:52 |
| blackburn | night | 23:52 |
| --- Log closed Tue Jul 17 00:00:17 2012 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!