| --- Log opened Sat Nov 24 00:00:15 2012 | ||
| -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 00:37 | |
| shogun-buildbot | build #183 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/183 | 04:19 |
|---|---|---|
| -!- ptizoom [~christian@79-71-89-182.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat] | 04:51 | |
| -!- ptizoom [~christian@79-71-89-182.dynamic.dsl.as9105.com] has joined #shogun | 04:52 | |
| wiking | anybody around? | 14:20 |
| wiking | blackburn: ping | 14:27 |
| blackburn | wiking: re | 15:26 |
| blackburn | wiking: what's up? | 15:35 |
| wiking | gmnp | 15:43 |
| wiking | why is it ported this way into shogun | 15:43 |
| wiking | now it's like aimed to be used for kernel machine | 15:44 |
| blackburn | wiking: uhm? | 15:45 |
| blackburn | can it be used for anything else? | 15:46 |
| wiking | yes | 15:47 |
| wiking | it's a general | 15:47 |
| wiking | qp soler | 15:47 |
| wiking | solver | 15:47 |
| blackburn | didn't know | 15:49 |
| blackburn | well | 15:50 |
| blackburn | may be it is not that cool for other problems | 15:50 |
| wiking | well it's just solves this problem | 16:05 |
| wiking | min 0.5*alpha'*H*alpha + c'*alpha | 16:05 |
| wiking | subject to sum(alpha) = 1, alpha(i) >= 0 | 16:05 |
| blackburn | wiking: so you mean it can be used for your problems? | 16:20 |
| blackburn | is it still state-of-the-art? | 16:20 |
| -!- zxtx [~zv@ool-45750cfe.dyn.optonline.net] has quit [Ping timeout: 244 seconds] | 16:38 | |
| @sonney2k | wiking, isn't it in libqp nowadyas anyways? | 19:13 |
| blackburn | sonney2k: ha, JL works on C api for VW | 19:16 |
| blackburn | even better | 19:16 |
| blackburn | we don't have to do anything just wait :D | 19:16 |
| wiking | sonney2k: ? | 19:30 |
| wiking | sonney2k: i don't know i mean i've just seen at vojtech's page that he used to release | 19:30 |
| wiking | different gmnp versions | 19:30 |
| wiking | i mean with different optimizers | 19:30 |
| wiking | but maybe it has been the past | 19:30 |
| wiking | and now he has a good reason why not to push forward gmnp releases | 19:31 |
| wiking | and that's why they decided to use libqp as well for the bmrm stuff with uricamic | 19:31 |
| wiking | anyhow | 19:33 |
| wiking | i'll just fucking push this stuff because i'm already ready with this | 19:33 |
| wiking | althoguth the implementation could be cleaner... | 19:33 |
| wiking | anybody up for reviewing | 19:33 |
| wiking | ? | 19:33 |
| @sonney2k | blackburn, I know - there was hope shashwat could do this but he is busy with real life | 19:38 |
| wiking | great sonney2k here u go then | 19:38 |
| wiking | do a review | 19:38 |
| blackburn | review what? | 19:41 |
| wiking | the new code | 19:41 |
| blackburn | wherE? | 19:42 |
| wiking | just a sec | 19:42 |
| blackburn | haha | 19:42 |
| wiking | booom | 19:46 |
| -shogungit:#shogun- [shogun] vigsterkr opened pull request #833: Introduction of a new non-convex bundle method risk minimization algorithm (master...so-dev) https://github.com/shogun-toolbox/shogun/pull/833 | 19:46 | |
| blackburn | do you take vodka? what makes you productive? | 19:46 |
| wiking | oh shit | 19:46 |
| wiking | waith | 19:46 |
| wiking | i think this wont compile | 19:47 |
| wiking | i'll have to push an extra commit | 19:47 |
| blackburn | #if 0 | 19:47 |
| blackburn | wtf? | 19:47 |
| wiking | where is it stuck? | 19:47 |
| wiking | blackburn: sorry i had to make some debugging | 19:47 |
| blackburn | well I mean | 19:47 |
| blackburn | does it work to define 0? | 19:47 |
| blackburn | :D | 19:47 |
| wiking | it works to comment out something | 19:47 |
| wiking | ;) | 19:47 |
| blackburn | hmm | 19:48 |
| wiking | ah that shit has been removed already | 19:48 |
| wiking | so don't worry | 19:48 |
| blackburn | okayface | 19:48 |
| wiking | but still wait | 19:49 |
| wiking | because there's one compiling bug | 19:49 |
| wiking | but yeah it's not only about ncbm | 19:50 |
| wiking | it's as well some optimization of libbm* | 19:50 |
| wiking | it's a very weird matlab -> c code | 19:50 |
| wiking | so it still has to be cleaned up imho | 19:50 |
| blackburn | wiking: can you cleanup my code too? ;) | 19:52 |
| wiking | heheh | 19:52 |
| wiking | think about where should we put all these optimizers | 19:52 |
| wiking | like libbmrm etc | 19:52 |
| blackburn | just put somewhere | 19:52 |
| blackburn | :D | 19:52 |
| blackburn | nobody uses our shit anyway lol | 19:53 |
| wiking | :D | 19:53 |
| wiking | ok | 19:53 |
| blackburn | wiking: https://github.com/shogun-toolbox/shogun/tree/master/src/shogun/optimization | 19:53 |
| blackburn | why not here? | 19:53 |
| wiking | well we could | 19:54 |
| wiking | but currently the stuff there | 19:54 |
| wiking | are all external stuff | 19:54 |
| blackburn | feel free to move | 19:54 |
| blackburn | I wanted | 19:54 |
| wiking | ok i've pushed the fixing commit | 19:54 |
| blackburn | then I get back to my alco addiction and sank in vodka | 19:55 |
| wiking | :>>> | 19:55 |
| wiking | ok if anybody has time for this | 19:55 |
| wiking | do a review and then apply the patch | 19:55 |
| blackburn | I have | 19:55 |
| blackburn | wiking: #if 0 is still here | 19:56 |
| wiking | where? | 19:56 |
| wiking | mmm there's a problem now with gcc | 19:57 |
| wiking | in travis | 19:57 |
| wiking | so wait | 19:57 |
| blackburn | wiking: did you get a moment to check tapkee before? | 19:58 |
| wiking | none :(* | 19:59 |
| wiking | sorry | 19:59 |
| wiking | was working on this | 19:59 |
| blackburn | no problem just meant it is still actual :) | 19:59 |
| wiking | yeah will do | 20:00 |
| wiking | how's the stuff with | 20:00 |
| wiking | tsnr? | 20:00 |
| blackburn | I have doubts with some design things | 20:00 |
| blackburn | t-sne | 20:00 |
| wiking | yeah that shit | 20:00 |
| blackburn | hmm | 20:00 |
| wiking | implemented? | 20:00 |
| blackburn | well not yet | 20:00 |
| wiking | i'd be keen to try that | 20:00 |
| blackburn | I keep this in mind | 20:00 |
| blackburn | wiking: how great is your openmp experience? | 20:01 |
| wiking | not sooo great | 20:01 |
| wiking | tried it like | 20:01 |
| blackburn | I have some hard times with it | 20:01 |
| wiking | in 2004-5 | 20:01 |
| wiking | no wait | 20:01 |
| wiking | 2007-8 | 20:01 |
| wiking | what's the problem | 20:01 |
| wiking | ? | 20:01 |
| blackburn | https://github.com/lisitsyn/tapkee/blob/master/tapkee/routines/multidimensional_scaling.hpp wiking for example loop in lines 74-87 is slower when parallel | 20:03 |
| wiking | ok | 20:03 |
| wiking | gcc fails | 20:03 |
| wiking | whyyyyyyyy | 20:04 |
| wiking | why you nooo want to see | 20:04 |
| wiking | /home/travis/builds/shogun-toolbox/shogun/src/shogun/structure/libppbm.cpp:596: undefined reference to `shogun::clean_icp(shogun::ICP_stats*, shogun::bmrm_return_value_T&, shogun::bmrm_ll**, shogun::bmrm_ll**, double*&, double*&, double*&, bool*&, unsigned int, double*&, unsigned int*&, unsigned int)' | 20:04 |
| wiking | ?! | 20:05 |
| wiking | and only gcc fails with it | 20:05 |
| blackburn | heh | 20:05 |
| wiking | maybe because it's and inline function? | 20:05 |
| blackburn | yeah can be | 20:06 |
| blackburn | I feel it makes no sense to inline function defined in .cpp | 20:06 |
| blackburn | am I wrong? | 20:06 |
| wiking | why not? | 20:07 |
| wiking | ok here's an attempt | 20:07 |
| wiking | let's see what travis does | 20:08 |
| blackburn | okay how can one inline a function | 20:08 |
| blackburn | if it have no implementation available | 20:08 |
| blackburn | has* argh | 20:09 |
| wiking | inline man | 20:09 |
| wiking | it should act as if a macro | 20:09 |
| wiking | almost | 20:10 |
| blackburn | yes | 20:10 |
| blackburn | but how if you have no impl around? | 20:10 |
| wiking | but the syntax is 'nicer' | 20:10 |
| blackburn | you have only pointer to that function in lib | 20:10 |
| blackburn | can't get code back | 20:10 |
| wiking | clang has no problem with it | 20:12 |
| wiking | btw | 20:12 |
| wiking | ;) | 20:12 |
| blackburn | clang is smart | 20:13 |
| @sonney2k | wiking, whcih line has the problem? | 20:14 |
| wiking | evaluation_cross_validation_multiclass_mkl | 20:14 |
| wiking | there's a problem in this one | 20:14 |
| wiking | are we aware of this? | 20:15 |
| @sonney2k | wiking, yes | 20:15 |
| @sonney2k | there are IIRC 3 valgrind errors in examples | 20:15 |
| wiking | ok i'm just saying | 20:16 |
| @sonney2k | wiking, I meant in your code though :) | 20:16 |
| wiking | ah well in general | 20:16 |
| @sonney2k | the compile error? | 20:16 |
| wiking | no i hope that's going to be fixed now | 20:16 |
| wiking | but like the stuff i did with uricamic's implementation for example | 20:16 |
| @sonney2k | wiking, what was it? | 20:16 |
| @sonney2k | did you have an inline functions that was not defined in a .h? | 20:17 |
| wiking | it was defined | 20:17 |
| wiking | but not implemented | 20:17 |
| @sonney2k | yeah then your bad | 20:17 |
| wiking | yeah it seems gcc doesn't like it | 20:17 |
| wiking | sonney2k: noup | 20:17 |
| wiking | it's gcc's bad | 20:17 |
| @sonney2k | yeah it's not standard | 20:17 |
| wiking | clang had no problem with it | 20:17 |
| wiking | anyhow it's fixed | 20:17 |
| @sonney2k | believe me clang has problems with other code | 20:17 |
| wiking | i wouldn't say it's non standard ... | 20:17 |
| @sonney2k | it is | 20:18 |
| @sonney2k | inline (by standard) has to be in .h | 20:18 |
| wiking | mmm | 20:18 |
| @sonney2k | definition and implementation | 20:18 |
| wiking | i doubt this very much | 20:18 |
| wiking | afaik c++ standard has no such clause | 20:18 |
| wiking | it's really a matter of compiler implementation | 20:18 |
| @sonney2k | http://stackoverflow.com/questions/3992980/c-inline-member-function-in-cpp-file | 20:19 |
| wiking | again | 20:19 |
| wiking | this is not in c++ standard | 20:19 |
| wiking | it's the matter of compiler | 20:20 |
| wiking | afaik vc++ would act very different as well | 20:20 |
| wiking | anyhow it's fixed now and works with both gcc and clang | 20:21 |
| wiking | hence i saved some code lines :) | 20:21 |
| wiking | but i have a feeling that all this stuff should be taken out from structure | 20:23 |
| wiking | and create an abstract class for this stuff | 20:23 |
| wiking | that would look nicely | 20:24 |
| wiking | this way people will not realise that they can have these solvers by themselves | 20:24 |
| wiking | as it's too much tied to structure .... | 20:24 |
| blackburn | to SO you mean? | 20:28 |
| @sonney2k | wiking, one step at a time | 20:28 |
| wiking | yep | 20:29 |
| @sonney2k | wiking, we should get your patch in first and then define such RiskMinimizer framework | 20:29 |
| wiking | to SO | 20:29 |
| wiking | as really these can be solved for anything in form of 1/2\lambda ||w||^2 + R(w), where now with NCBM we can solve non-convex risk as well | 20:29 |
| wiking | so like in my case | 20:30 |
| wiking | latent so svm is: 1/2 ||w||^2 + c/n \sum max_{y',h'} (w*psi(x,y',h') + loss(y,y',h')) - c/n \sum max_h (w*psi(x,y,h)) | 20:31 |
| wiking | and now the good thing with ncbm | 20:32 |
| wiking | that i dont care about the problem not being convex | 20:32 |
| @sonney2k | wiking, well feel free to start CRiskMinimizer in shogun/optimization! | 20:32 |
| wiking | but yeah as i said earlier | 20:32 |
| wiking | sonney2k: heheh i have to produce a paper in 5 days | 20:33 |
| wiking | so maybe next year ;) | 20:33 |
| wiking | but this can already be part of 2.1.0 release ;P | 20:33 |
| wiking | just to have some valueable contribution for that release | 20:34 |
| blackburn | wiking: you run macosx right? | 20:34 |
| blackburn | can you try to compile git://github.com/lisitsyn/tapkee.git? | 20:34 |
| wiking | letmecone | 20:35 |
| wiking | clone | 20:35 |
| wiking | error | 20:35 |
| blackburn | cool | 20:35 |
| wiking | ok wait | 20:35 |
| wiking | i think header paths are not ok | 20:35 |
| wiking | is there like a configure script? | 20:36 |
| blackburn | argh | 20:36 |
| blackburn | no | 20:36 |
| blackburn | you know it is immature | 20:36 |
| wiking | ok compiling | 20:37 |
| wiking | i dont have openmp | 20:37 |
| wiking | so let's see how it handles it | 20:37 |
| blackburn | it is ok I don't use it yet | 20:37 |
| wiking | oh shit | 20:37 |
| wiking | lapack | 20:37 |
| wiking | :D | 20:37 |
| wiking | yeah dude | 20:38 |
| wiking | u need a configure script | 20:38 |
| wiking | ;) | 20:38 |
| wiking | configure.ac and friends | 20:38 |
| blackburn | of course I need | 20:38 |
| blackburn | wiking: I templated everything to random access iterator - do you think it is reasonable? | 20:41 |
| blackburn | instead of forcing indexing 0,...,N-1 I mean | 20:41 |
| blackburn | I probably know what sonney2k thinks about that | 20:42 |
| blackburn | :D | 20:42 |
| wiking | hehehe | 20:44 |
| wiking | well yeah then there's a lot of R-B tree operation involved | 20:44 |
| wiking | so think twice if u need that really | 20:44 |
| wiking | if not then try to use other containers | 20:44 |
| blackburn | R-B? | 20:44 |
| blackburn | why? | 20:45 |
| wiking | red-black tree | 20:45 |
| blackburn | where? | 20:45 |
| wiking | that's behind std::vector for example | 20:45 |
| wiking | check c++ lib implementation | 20:45 |
| wiking | and u'll see | 20:45 |
| blackburn | for what? | 20:45 |
| wiking | ah no | 20:45 |
| wiking | i'm stupid | 20:46 |
| blackburn | red black is only in map | 20:46 |
| blackburn | and set | 20:46 |
| wiking | and set | 20:46 |
| blackburn | (as I thought) | 20:46 |
| blackburn | you confused me quite a lot :D | 20:46 |
| wiking | still | 20:47 |
| wiking | u stl smart | 20:47 |
| wiking | although i myself tend to throw things | 20:47 |
| wiking | at random | 20:47 |
| blackburn | it is not really stl related | 20:47 |
| blackburn | I mean do you think | 20:47 |
| wiking | and my first choice is really always std::vector :D | 20:47 |
| blackburn | having ability to change the type | 20:47 |
| blackburn | of data you operate on | 20:48 |
| blackburn | is a good thing | 20:48 |
| blackburn | i.e. writing a callback that operates on your structs not 0,...,N-1 | 20:48 |
| blackburn | see what I mean? | 20:48 |
| wiking | aaah | 20:49 |
| blackburn | I thought it is useful | 20:49 |
| wiking | well | 20:49 |
| wiking | depends | 20:49 |
| blackburn | but now I am unsure | 20:50 |
| wiking | if it doesn't give u too much of a headache to handle this | 20:50 |
| wiking | then it's usuful | 20:50 |
| wiking | as you give more freedom for the users of the library | 20:50 |
| blackburn | it is done already so the question is whether to get rid of it or not | 20:50 |
| blackburn | :D | 20:50 |
| wiking | i wouldn't | 20:50 |
| blackburn | one thing I will get instantly is easier parallelization | 20:52 |
| wiking | sonney2k: still here? | 21:05 |
| wiking | blackburn: what would you inherit CBMRM from? | 21:05 |
| blackburn | what is CBMRM? | 21:06 |
| blackburn | :D | 21:06 |
| wiking | bundle method risk minimization | 21:06 |
| wiking | or CRiskMinimizer as sonney2k suggested | 21:06 |
| blackburn | I have no special vision on that actually | 21:12 |
| wiking | mmm fuck i need to resolve some stuff with my stash | 21:14 |
| wiking | argh i hate this :( | 21:14 |
| blackburn | stash if useful to waste code | 21:14 |
| -!- zxtx [~zv@ool-45750cfe.dyn.optonline.net] has joined #shogun | 21:19 | |
| wiking | oh fuck | 21:23 |
| wiking | how can i reset a file? | 21:23 |
| blackburn | wiking: reset to? | 21:27 |
| wiking | i've realised | 21:29 |
| wiking | it's a bit complicated | 21:29 |
| wiking | so if you want to undo changes to a files by a patch | 21:29 |
| wiking | then you have to first | 21:29 |
| wiking | git reset file | 21:29 |
| wiking | and then | 21:29 |
| wiking | git checkout file | 21:29 |
| blackburn | yes | 21:30 |
| blackburn | wiking: my openmp impl is 6 times slower :D | 21:30 |
| wiking | heheh that cna happen | 21:31 |
| wiking | too much of a context change unfortunately | 21:31 |
| blackburn | I don't want to write pthreads code again | 21:32 |
| wiking | well that wouldn't help u | 21:32 |
| wiking | i mean what u can do with pthreads | 21:33 |
| wiking | you can do with openmp as well | 21:33 |
| blackburn | yeah but openmp is less explicit and takes time to get into it | 21:33 |
| blackburn | so I am in trouble hmm | 21:34 |
| wiking | another pr | 21:44 |
| wiking | boom | 21:44 |
| -shogungit:#shogun- [shogun] vigsterkr opened pull request #834: Template specialization for add_to_dense_vec in DenseFeatures (master...master) https://github.com/shogun-toolbox/shogun/pull/834 | 21:45 | |
| blackburn | okay | 21:45 |
| blackburn | that one is easy | 21:45 |
| -shogungit:#shogun- [shogun] lisitsyn pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/49682e5e5390...c250bb9beb21 | 21:45 | |
| -shogungit:#shogun- shogun/master bfdff65 Viktor Gal: Template specialization for add_to_dense_vec in DenseFeatures | 21:45 | |
| -shogungit:#shogun- shogun/master c250bb9 Sergey Lisitsyn: Merge pull request #834 from vigsterkr/master... | 21:45 | |
| shogun-buildbot | build #670 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/670 | 22:33 |
| wiking | yey the buildbot agrees with the patch | 22:43 |
| blackburn | wiking: have you seen boost.graph? | 22:43 |
| wiking | yeps | 22:44 |
| blackburn | wiking: http://www.boost.org/doc/libs/1_52_0/libs/graph/doc/dijkstra_shortest_paths_no_color_map.html I feel dizzy | 22:44 |
| blackburn | 10 template parameters oh my f | 22:44 |
| wiking | yeah | 22:45 |
| wiking | usually a graph lib is like that | 22:45 |
| blackburn | ohh | 22:45 |
| wiking | once i've implemented some basic graph stuff | 22:45 |
| wiking | and then u realise that if u want to support funky algos | 22:45 |
| wiking | you'll end up fucked up | 22:45 |
| blackburn | sure | 22:45 |
| wiking | "weighted, directed or undirected graph" | 22:46 |
| wiking | yep | 22:46 |
| wiking | this part makes it really complicated already | 22:46 |
| wiking | to support this all | 22:46 |
| wiking | but boost is really cool | 22:46 |
| wiking | but sometimes of course overhyped | 22:47 |
| wiking | like for example | 22:47 |
| wiking | all the people in quantitative finance use boost | 22:47 |
| blackburn | why? | 22:47 |
| wiking | well that's the thing | 22:47 |
| blackburn | or rather what for? | 22:47 |
| wiking | HFT | 22:47 |
| wiking | high frequency trading | 22:47 |
| blackburn | well still what is the place for boost here? | 22:47 |
| wiking | a) portability of the library | 22:48 |
| wiking | b) has some really handy threadpooling | 22:48 |
| wiking | as well as memory managment | 22:48 |
| blackburn | ah that part | 22:49 |
| wiking | as well as shared pointers | 22:49 |
| wiking | so all these stuff that makes your coding easier | 22:49 |
| blackburn | that's quite reasonable | 22:49 |
| wiking | and they want c++ | 22:49 |
| wiking | as it's the fastest | 22:49 |
| wiking | i mean c/c++ | 22:49 |
| wiking | and as they do high frequency trading | 22:49 |
| wiking | speed is everything there | 22:49 |
| blackburn | why do you know that stuff? | 22:50 |
| blackburn | :) | 22:50 |
| wiking | http://www.boost.org/doc/libs/1_35_0/libs/statechart/doc/faq.html#HardRealtime | 22:50 |
| wiking | this is handy as well | 22:51 |
| wiking | if we could make some of our algos | 22:51 |
| wiking | possible to set time | 22:51 |
| wiking | i mean hard real-time requiremtn | 22:51 |
| wiking | i'm sure shogun would be used ;) | 22:51 |
| wiking | why do i know these stuff | 22:52 |
| wiking | because i'm thinking about getting job in the sector | 22:52 |
| blackburn | I wouldn't mind as well :D | 22:52 |
| wiking | it's hard though | 22:52 |
| blackburn | I bet this guys have enough money | 22:52 |
| blackburn | these* | 22:53 |
| wiking | blackburn: http://www.quantfinancejobs.com/jobs/job.aspx?JobID=13550 | 22:53 |
| wiking | all the rocket scientist of the future end up working for investment banks ;) | 22:53 |
| wiking | because of the cash | 22:53 |
| blackburn | ah you forgot to say that there is no way to get there | 22:53 |
| wiking | most sought people for these jobs are phds in physics | 22:54 |
| wiking | and there's a little bit of indian maffia involved in the thing | 22:54 |
| wiking | as the best quants out there are indians | 22:54 |
| blackburn | hehe | 22:55 |
| wiking | but yeah | 22:55 |
| wiking | it'd be great if we could do some funky stuff with our learners | 22:56 |
| wiking | and especially classification | 22:56 |
| blackburn | regressors you mean? | 22:56 |
| wiking | nono | 22:56 |
| wiking | so in general | 22:56 |
| wiking | one could set a real time constraint | 22:56 |
| wiking | on the classification | 22:56 |
| blackburn | you mean classify that stuff in 0.0001s? | 22:57 |
| wiking | let's say learning time you dont care about | 22:57 |
| wiking | but you want to be able to set to classify with the learnt model | 22:57 |
| wiking | within given hard deadline | 22:57 |
| wiking | so yeah | 22:57 |
| wiking | if u want it you could set a soft/hard deadline | 22:58 |
| wiking | for that given function call | 22:58 |
| wiking | i think that'd be interesting for a lot of parties | 22:58 |
| blackburn | I see that out of shogun | 22:58 |
| wiking | i mean shogun has the advantage | 22:59 |
| blackburn | oh my these rates are crazy | 22:59 |
| wiking | that thanks to sonney2k it's really trying to as less reseource wasting as much possible | 22:59 |
| wiking | so it's a good basis to start with | 22:59 |
| wiking | but needs some work as far as i can see | 23:00 |
| wiking | to be able to support such things | 23:00 |
| wiking | i dont even know how we could start this | 23:01 |
| wiking | so for example that we have a simple linearMachine | 23:01 |
| wiking | and have it's classifier be able to support it | 23:02 |
| blackburn | wiking: if you are interested we could play with it somewhere in a separate lib | 23:02 |
| wiking | yeah we could make a fork | 23:02 |
| blackburn | just sharing ideas from/with shogun | 23:02 |
| wiking | but how would u do it? | 23:03 |
| blackburn | do what? | 23:03 |
| wiking | well start it | 23:03 |
| wiking | i mean essentially i think if we could do this right | 23:03 |
| wiking | we could merge it in shogun for sure | 23:03 |
| blackburn | I think best way is to use shared ideas | 23:03 |
| blackburn | not code | 23:03 |
| wiking | mm | 23:39 |
| wiking | what was that procject | 23:40 |
| wiking | project | 23:40 |
| wiking | the opencl one | 23:40 |
| blackburn | ehm | 23:40 |
| blackburn | that one french guy is from? | 23:40 |
| wiking | viennacl | 23:40 |
| blackburn | ahh | 23:40 |
| blackburn | yes | 23:40 |
| wiking | mmm | 23:50 |
| wiking | it supports eigen as well | 23:50 |
| blackburn | wiking: yes I am curious now too | 23:54 |
| wiking | we could add SGVector | 23:54 |
| wiking | to support this | 23:55 |
| wiking | btw | 23:55 |
| wiking | why is this | 23:55 |
| wiking | static inline float64_t dot( | 23:55 |
| wiking | const uint64_t* v1, const uint64_t* v2, int32_t n) | 23:55 |
| wiking | { | 23:55 |
| wiking | i mean why a dot product of two uint64 is a flot? | 23:55 |
| wiking | float? | 23:55 |
| wiking | indeed | 23:56 |
| wiking | cant we just have | 23:56 |
| wiking | a non-static | 23:56 |
| blackburn | in shogun everything is float64 :D | 23:56 |
| wiking | T dot(const SGVector<T>& v) const; | 23:56 |
| wiking | ? | 23:56 |
| wiking | i mean even more | 23:56 |
| wiking | T operator*(const SGVector<T>& v) | 23:57 |
| wiking | ? | 23:57 |
| blackburn | no that's not cool | 23:57 |
| wiking | why? | 23:57 |
| blackburn | int8 * int8? | 23:57 |
| wiking | ah u are staring to worry about overflow? | 23:57 |
| blackburn | wiking: yes a little | 23:58 |
| wiking | because then it's the same with float64 | 23:58 |
| wiking | it should return at least floatmax | 23:58 |
| wiking | ;P | 23:58 |
| wiking | but ok let's say | 23:58 |
| blackburn | well but with int8 it will happen for sure | 23:58 |
| wiking | something dot(SGVector<T>& asd) const | 23:58 |
| wiking | so that i dont have to call on two | 23:58 |
| wiking | so that one can call directly a dot product | 23:58 |
| wiking | because now a dot product is really stupid | 23:59 |
| wiking | even if both the vectors are in SGVector | 23:59 |
| blackburn | I if i could say it is not | 23:59 |
| blackburn | :D | 23:59 |
| wiking | ? | 23:59 |
| --- Log closed Sun Nov 25 00:00:02 2012 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!