--- Log opened Thu Nov 15 00:00:17 2012 | ||
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 00:29 | |
-!- blackburn [~blackburn@37.61.179.13] has quit [Quit: Leaving.] | 00:29 | |
shogun-buildbot | build #171 of nightly_default is complete: Failure [failed tutorial] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/171 | 01:37 |
---|---|---|
wiking | nooooooooooooooooooooooobuildbot | 01:46 |
wiking | dont die | 01:46 |
shogun-buildbot | build #172 of nightly_default is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/172 | 03:40 |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 09:12 | |
blackburn | wiking: what can we do with your oppan gangnam style solver? | 09:32 |
wiking | :> | 09:40 |
wiking | solveit :) | 09:41 |
blackburn | heeey sexy solver | 09:41 |
blackburn | oop oop | 09:41 |
blackburn | oppan | 09:41 |
wiking | :P | 09:42 |
wiking | blackburn: new job? | 10:17 |
blackburn | wiking: yes | 10:17 |
wiking | blackburn: where to? | 10:17 |
wiking | and congratz | 10:18 |
blackburn | http://smartsolutions-123.ru/en/ | 10:18 |
blackburn | thanks | 10:18 |
wiking | is this the java company? | 10:20 |
blackburn | wiking: yeah they do java and C# | 10:24 |
blackburn | wiking: well they fucking write code | 10:24 |
blackburn | not fixing legacy bugs | 10:24 |
blackburn | here I fix legacy bugs | 10:24 |
blackburn | I've been fscking fixing bugs for an year | 10:24 |
wiking | :>> | 10:25 |
blackburn | I just want to http://stfuawsc.com/ | 10:26 |
wiking | ahhaha | 10:27 |
wiking | ohman | 10:31 |
wiking | this ascii serialization format is not the nicest parsable format | 10:31 |
blackburn | wiking: ah btw could you please take a look at https://github.com/lisitsyn/tapkee when you have time | 10:32 |
wiking | ah yeah a friend asked | 10:32 |
blackburn | may be you could have some thoughts what is wrong | 10:32 |
wiking | do you know a very efficient method to diagonilze small matrices | 10:32 |
wiking | but a lot of it | 10:32 |
blackburn | dense? | 10:32 |
wiking | small matrix | 10:33 |
wiking | 6x6 | 10:33 |
wiking | so yeah dense matrix | 10:33 |
wiking | ok what could be wrong with this? | 10:33 |
wiking | i mean it's just like design wise i should look | 10:33 |
wiking | or what exactly you would like to talk about? | 10:33 |
blackburn | wiking: yes, design wise | 10:33 |
blackburn | sure I don't ask you to find a lowlevel mistake somewhere :) | 10:33 |
wiking | :> | 10:34 |
wiking | ok i'll check it | 10:34 |
blackburn | okay actually what I know about diagonalization | 10:34 |
blackburn | eigen has bad and mkl has good | 10:34 |
blackburn | :D | 10:34 |
wiking | mmm | 10:37 |
wiking | why we cannot have a serializer like | 10:37 |
wiking | numpy has | 10:37 |
wiking | savetxt | 10:37 |
blackburn | wiking: what's the difference? | 10:39 |
wiking | well the format? :D | 10:43 |
wiking | have u seen the output of SerializableAsciiFile | 10:43 |
wiking | ? | 10:43 |
wiking | :> | 10:43 |
sonne|work | wiking: we have simple load / save routines | 10:46 |
sonne|work | serializable* stuff works for much more complex data | 10:47 |
sonne|work | like saving whole svm incl. kernel & features etc | 10:47 |
wiking | sonne|work: for sgvector? | 10:50 |
sonne|work | wiking: for *features | 10:50 |
wiking | :D | 10:50 |
sonne|work | but yes this should be moved into sgvector and friends | 10:50 |
wiking | i just want to save a simple sgvector | 10:51 |
sonne|work | but we didn't have that at the time | 10:51 |
sonne|work | basically features were what SGVector etc were before | 10:51 |
wiking | ok | 10:51 |
wiking | i'm chechking the code | 10:51 |
sonne|work | wiking: btw I asked again | 10:51 |
sonne|work | (not sure if you have seen) | 10:52 |
sonne|work | answer 'tonight' | 10:52 |
sonne|work | but that was yesterday | 10:52 |
wiking | sonne|work: yeah i've seen it | 10:52 |
sonne|work | no idea | 10:52 |
wiking | i've even answered | 10:52 |
sonne|work | heh I went to bed | 10:52 |
wiking | so yeah i know that around 1pm CET | 10:52 |
wiking | or somewhere there we should get some mail | 10:52 |
sonne|work | or not | 10:54 |
sonne|work | whatever strange handling | 10:54 |
wiking | :> | 10:54 |
wiking | sonne|work: why do you set/reset locale? | 11:00 |
sonne|work | 0,1 <-> 0.1 | 11:03 |
sonne|work | in germany it is 0,1 | 11:03 |
sonne|work | in us 0.1` | 11:03 |
sonne|work | so better use C always | 11:03 |
wiking | ah got it | 11:03 |
wiking | lol | 11:06 |
wiking | ../shogun/io/File.h:33:22: error: expected class name | 11:06 |
wiking | class CFile : public CSGObject | 11:06 |
wiking | wtf | 11:06 |
wiking | ok this is fucking hilarious | 11:09 |
wiking | http://pastebin.com/963ts5Rv | 11:09 |
wiking | hahah i cannot include File.h in SGVector | 11:10 |
wiking | it's like a snake biting his own tail | 11:14 |
wiking | somewhere we need a class CSGObject; ;) | 11:14 |
wiking | but it's hilarious | 11:15 |
wiking | whatever i start in the last 3 days nothing works | 11:15 |
wiking | so i think i shouldnt do anything | 11:15 |
sonne|work | ahhh | 11:21 |
sonne|work | that all needs to be in a .cpp file | 11:21 |
sonne|work | then you can use forward declarations | 11:21 |
sonne|work | I hate this dependency hell in C++ | 11:22 |
wiking | what all needs to be in a cpp file | 11:31 |
wiking | i mean yes sure but in order to be able to declare the load/save function for SGVEctor i have to include the header | 11:32 |
wiking | and now it's extremely fucked up | 11:32 |
wiking | because no matter where i put the class CSGObject; declaration | 11:32 |
wiking | i mean in which header | 11:32 |
wiking | the includes are first | 11:32 |
wiking | so i have to put it before any include | 11:33 |
wiking | ok solved it | 11:33 |
-shogungit:#shogun- [shogun] vigsterkr opened pull request #825: Add load/save functionality for SGVector (master...latent) https://github.com/shogun-toolbox/shogun/pull/825 | 12:31 | |
wiking | so here it goes... a bit of a fixing as well of precisions... | 12:31 |
wiking | oh sorry some shit got left in sgvector... | 12:32 |
sonne|work | what is the +GET_VECTOR(get_vector, get_int8_matrix, int8_t) | 13:04 |
sonne|work | 68 69 | 13:04 |
sonne|work | GET_VECTOR(get_vector, get_matrix, uint8_t) ? | 13:04 |
wiking | ? | 13:05 |
wiking | line 68/69 was there | 13:06 |
wiking | i've just added GET_VECTOR(get_vector, get_int8_matrix, int8_t) | 13:06 |
wiking | to support int8_t | 13:06 |
sonne|work | ahh I see. another transition needed :/ | 13:39 |
sonne|work | we decided to use overloading this GSoC | 13:39 |
wiking | ? | 13:41 |
blackburn | what? | 13:42 |
sonne|work | what?? | 13:43 |
wiking | whaaaat? | 13:43 |
blackburn | WAT | 13:44 |
blackburn | sonne|work: in germany you also have comma instead of point? | 14:04 |
sonne|work | yes | 14:05 |
blackburn | I thought it is used only in ussr :) | 14:06 |
sonne|work | you have it too :)( | 14:07 |
blackburn | I don't like that pretty much | 14:07 |
blackburn | and I don't like cyrillic letters actually | 14:08 |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] | 14:09 | |
sonne|work | that's the reason russians drink lot of vodka | 14:35 |
-!- n4nd0 [~nando@n150-p8.kthopen.kth.se] has joined #shogun | 16:59 | |
-!- n4nd0 [~nando@n150-p8.kthopen.kth.se] has quit [Quit: leaving] | 17:35 | |
wiking | sonne|work: i guess there's still no news :D | 17:53 |
-!- koen2 [9232343c@gateway/web/freenode/ip.146.50.52.60] has joined #shogun | 17:58 | |
-!- blackburn [~blackburn@109.226.125.245] has joined #shogun | 18:02 | |
-!- koen2 [9232343c@gateway/web/freenode/ip.146.50.52.60] has quit [Ping timeout: 245 seconds] | 18:02 | |
wiking | blackburn: yey | 18:16 |
wiking | although i haven't been looking for it | 18:16 |
wiking | but i found a solution for solving nonconvex function with bundiling methods :) | 18:17 |
wiking | yo anybody | 18:25 |
wiking | (Fenchel Dual) is only true for convex functions? | 18:26 |
wiking | well i'll assume this :DD | 18:32 |
wiking | and let's see where it takes me | 18:32 |
blackburn | wiking: hey | 18:47 |
blackburn | non-convex? | 18:47 |
wiking | aayep | 18:52 |
wiking | i'm just checking if i can apply the dual form | 18:52 |
wiking | as for that we have libqp as a solver | 18:52 |
wiking | or do we have a primal solver for min_w \lambda/2 ||w||^2+max_{j=1..t}{c_j(w)} | 18:54 |
wiking | i guess this is a no | 19:08 |
wiking | :) | 19:08 |
blackburn | re | 19:17 |
blackburn | wiking: wat ist da c_j? | 19:17 |
wiking | c_j(w) = <a_j, w> + b_j | 19:19 |
wiking | it's a cutting plane basically | 19:19 |
blackburn | wiking: can't libqp solve this? | 19:27 |
wiking | well the dual form of it yes | 19:29 |
wiking | i'm just not sure that the dual that i get with Fenchel Dual theorem stands if that function is non-convex | 19:30 |
blackburn | I feel stupid | 19:32 |
wiking | yeah me too | 19:33 |
wiking | :))) | 19:33 |
wiking | but i'm just bruteforcing it | 19:33 |
wiking | blackburn: what's gonna happen with that 2 PR? :) | 19:43 |
blackburn | wiking: no idea I didn't take a look | 19:45 |
blackburn | did soeren? | 19:45 |
blackburn | wiking: why to load sgvector from file? | 19:46 |
wiking | blackburn: why not | 19:46 |
wiking | i needed now | 19:47 |
wiking | :D | 19:47 |
blackburn | I see | 19:50 |
blackburn | how do you use it though? | 19:50 |
-shogungit:#shogun- [shogun] lisitsyn pushed 3 new commits to master: https://github.com/shogun-toolbox/shogun/compare/0cdc7eb45a44...2cd18bcf21d4 | 19:56 | |
-shogungit:#shogun- shogun/master df7368c Viktor Gal: Add load/save functionality for SGVector | 19:56 | |
-shogungit:#shogun- shogun/master 887e00d Viktor Gal: Remove commented code from SGVector | 19:56 | |
-shogungit:#shogun- shogun/master 2cd18bc Sergey Lisitsyn: Merge pull request #825 from vigsterkr/latent... | 19:56 | |
wiking | jej | 20:10 |
wiking | if ya wanna save a w of an svm | 20:10 |
wiking | there a new one | 20:12 |
-shogungit:#shogun- [shogun] vigsterkr opened pull request #826: Fix dual structure solvers to use CMath::sqrt (master...latent) https://github.com/shogun-toolbox/shogun/pull/826 | 20:12 | |
-shogungit:#shogun- [shogun] lisitsyn pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/2cd18bcf21d4...097f9c80f031 | 20:13 | |
-shogungit:#shogun- shogun/master 98b85f6 Viktor Gal: Fix dual structure solvers to use CMath::sqrt | 20:13 | |
-shogungit:#shogun- shogun/master 097f9c8 Sergey Lisitsyn: Merge pull request #826 from vigsterkr/latent... | 20:13 | |
wiking | blackburn: here? | 20:50 |
blackburn | yeh | 20:50 |
wiking | blackburn: why we only have gmnp_imdm ? | 20:51 |
blackburn | malozemov demyanov mitchell? | 20:51 |
wiking | Improved Mitchell-Demyanov-Malozemov | 20:51 |
blackburn | just random guess - it is the best | 20:51 |
blackburn | :D | 20:51 |
blackburn | okay actually I asked myself why | 20:51 |
blackburn | but I didn't ask vojtech | 20:52 |
blackburn | may be he was lazy | 20:52 |
blackburn | may be it is the best approach | 20:52 |
wiking | because he has other ones as well | 20:52 |
wiking | in his other | 20:52 |
wiking | lib | 20:52 |
wiking | mdm ... Mitchell-Demyanov-Malozemov | 20:52 |
wiking | % imdm ... Improved Mitchell-Demyanov-Malozemov (default). | 20:52 |
wiking | % iimdm ... Improved (version 2) Mitchell-Demyanov-Malozemov. | 20:52 |
wiking | % kozinec ... Kozinec algorithm. | 20:52 |
wiking | % keerthi ... Derived from NPA algorithm by Keerthi et al. | 20:52 |
wiking | % kowalczyk ... Based on Kowalczyk's maximal margin perceptron. | 20:52 |
blackburn | well | 20:52 |
blackburn | what do you expect from say kozinec? | 20:52 |
blackburn | I mean they are all nearly the same I believe | 20:53 |
wiking | oh nothing | 20:53 |
wiking | i was just wondering why | 20:53 |
blackburn | ahh | 20:53 |
blackburn | okay | 20:53 |
blackburn | I just mean | 20:53 |
blackburn | not really sure we can speed it up | 20:53 |
blackburn | or anything | 20:53 |
blackburn | but yes would be nice to have all of them | 20:53 |
blackburn | I wanted to implement that even | 20:53 |
blackburn | copy and paste code I mean | 20:53 |
wiking | it's just porting basicaly | 20:54 |
wiking | at first step | 20:54 |
blackburn | I am afraid it is vojtech's phd related | 20:54 |
blackburn | if you know what I mean | 20:54 |
blackburn | :D | 20:54 |
blackburn | it looks more powerful when you have 6 methods | 20:54 |
blackburn | and his phd is all about multiclass | 20:55 |
wiking | :> | 20:56 |
wiking | ok so now the last question | 20:58 |
wiking | which is better to use | 20:58 |
wiking | gnmp or libqp? | 20:58 |
wiking | as i can see | 20:58 |
wiking | gnmplib solves almost the same (actually exactly what i need) | 20:59 |
wiking | libqp is more general | 20:59 |
wiking | ah ok | 20:59 |
wiking | no i dont get it | 20:59 |
blackburn | yeah libqp is more general | 20:59 |
wiking | :) | 21:00 |
blackburn | do you think gmnplib solves the same problem? | 21:00 |
wiking | well not the same problem | 21:00 |
wiking | but the non-convex thing | 21:00 |
blackburn | is MC problem non-convex? | 21:00 |
wiking | well | 21:00 |
wiking | i have this problem | 21:00 |
wiking | 1/2 ||w||^2 + R(w) | 21:01 |
wiking | usual bundle problem | 21:01 |
wiking | and the dual form of this | 21:01 |
wiking | is actually exactly what gnmplib solves | 21:01 |
blackburn | hmm I see | 21:01 |
blackburn | then use it :) | 21:01 |
wiking | u mean gnmplib? | 21:01 |
blackburn | but I didn't like gmnp when I use it | 21:01 |
wiking | i wonder why uricamic used libqp and not gnmplib | 21:01 |
blackburn | converges for hours | 21:02 |
wiking | i guess vojtec had a good reason for telling him to use libqp instead of gnmp | 21:02 |
wiking | ;P | 21:02 |
wiking | or? :) | 21:02 |
blackburn | yeah he probably knows better | 21:02 |
blackburn | the prob is 1.0 | 21:02 |
wiking | well actually i'll write both | 21:02 |
wiking | and one will be able to choose which QP solver to use | 21:03 |
blackburn | what are you developing? | 21:03 |
wiking | and we can even compare | 21:03 |
blackburn | is that latent? | 21:03 |
wiking | this nonconvex bundle method | 21:03 |
wiking | well | 21:03 |
blackburn | oppan gangnam style? | 21:03 |
wiking | yeah | 21:03 |
wiking | latent model's formulation is usually nonconvex | 21:03 |
blackburn | I see | 21:03 |
wiking | for solving this | 21:03 |
wiking | usually one uses somekind of an approx | 21:03 |
wiking | like CCCP | 21:03 |
blackburn | haha | 21:04 |
wiking | but this solver would fucking solve it in your face :) | 21:04 |
wiking | without approxx | 21:04 |
wiking | :>> | 21:04 |
wiking | gangnam style shit bro | 21:04 |
blackburn | how can they name method like that | 21:04 |
blackburn | CCCP | 21:04 |
blackburn | damn | 21:04 |
wiking | blackburn: well yeah | 21:04 |
blackburn | I have seen that before | 21:04 |
blackburn | but always laugh | 21:04 |
wiking | indeed | 21:04 |
wiking | me too | 21:04 |
wiking | but that's how it is | 21:04 |
blackburn | I will call my next method ????? | 21:05 |
blackburn | or ?????? | 21:05 |
wiking | stalin is cool | 21:05 |
wiking | ;) | 21:05 |
wiking | ctalnh | 21:05 |
wiking | almost fonetic of the cyrilic | 21:05 |
wiking | so that u dont have to use utf8 | 21:05 |
wiking | ;) | 21:05 |
blackburn | CTALNH | 21:06 |
blackburn | hah | 21:06 |
blackburn | lol | 21:06 |
wiking | yeah indeed | 21:07 |
wiking | mmm shitfuck | 21:07 |
wiking | i think i had an enlightment moment now :) | 21:07 |
blackburn | hallelujah | 21:07 |
blackburn | :) | 21:07 |
wiking | huh | 21:09 |
wiking | where the fuck was this paper i was looking for now | 21:09 |
wiking | :)))) | 21:09 |
blackburn | in journal | 21:09 |
blackburn | hope that helps! | 21:10 |
blackburn | :D | 21:10 |
-!- blackburn [~blackburn@109.226.125.245] has quit [Quit: Leaving.] | 23:36 | |
--- Log closed Fri Nov 16 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!