--- Log opened Tue Jul 09 00:00:23 2013 | ||
shogun-buildbot | build #1342 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/1342 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>, Kevin <kevinhughes27@gmail.com> | 00:11 |
---|---|---|
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has quit [Quit: Leaving.] | 00:30 | |
@iglesiasg | ok guys, good night | 00:58 |
@iglesiasg | lisitsyn: have a nice flight! | 00:59 |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving] | 00:59 | |
-!- travis-ci [~travis-ci@ec2-54-235-27-28.compute-1.amazonaws.com] has joined #shogun | 01:21 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn'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/8863305 | 01:21 |
-!- travis-ci [~travis-ci@ec2-54-235-27-28.compute-1.amazonaws.com] has left #shogun [] | 01:21 | |
-!- nube [~rho@49.244.112.133] has quit [Ping timeout: 260 seconds] | 01:51 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 02:03 | |
-!- zxtx [~zv@rrcs-76-79-81-162.west.biz.rr.com] has quit [Ping timeout: 246 seconds] | 02:22 | |
-!- gsomix [~gsomix@109.188.125.190] has quit [Ping timeout: 256 seconds] | 02:44 | |
-!- nube [~rho@36.253.205.110] has joined #shogun | 03:56 | |
-!- nube1 [~rho@36.252.235.46] has joined #shogun | 04:03 | |
-!- nube [~rho@36.253.205.110] has quit [Ping timeout: 256 seconds] | 04:04 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 04:24 | |
-!- gsomix [~gsomix@109.188.127.50] has joined #shogun | 04:27 | |
shogun-buildbot | build #452 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/452 | 04:38 |
-!- nube1 [~rho@36.252.235.46] has quit [Read error: Connection reset by peer] | 04:58 | |
-!- hushell [~hushell@c-24-21-141-32.hsd1.or.comcast.net] has quit [Quit: WeeChat 0.3.7] | 05:42 | |
-!- nube [~rho@116.90.239.13] has joined #shogun | 06:25 | |
-!- foulwall [~user@2001:da8:215:6100:ad3e:1706:207e:6816] has joined #shogun | 06:44 | |
gsomix | sonney2k, sent PR. | 06:52 |
gsomix | need to go - CRAZY TRAIN! ALL ABOARD! | 06:54 |
-!- gsomix [~gsomix@109.188.127.50] has quit [Quit: Leaving] | 06:54 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 08:28 | |
shogun-notifier- | shogun: Evgeniy Andreev :develop * 9977edb / / (7 files): https://github.com/shogun-toolbox/shogun/commit/9977edbbdb3405ee9b1c5667f41b55f0d1ec540a | 08:28 |
shogun-notifier- | shogun: LineReader have configurable delimiters now; fixed some bugs | 08:28 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 7a5b65a / / (7 files): https://github.com/shogun-toolbox/shogun/commit/7a5b65adbe5044152d85f8d09422526d1b10397e | 08:28 |
shogun-notifier- | shogun: Merge pull request #1226 from gsomix/feature/linereader | 08:28 |
shogun-notifier- | shogun: | 08:28 |
shogun-notifier- | shogun: LineReader have configurable delimiters now | 08:28 |
shogun-notifier- | shogun: van51 :develop * 82715cb / / (3 files): https://github.com/shogun-toolbox/shogun/commit/82715cb78919b3a8998be3ae6c8350df6cf0df30 | 08:35 |
shogun-notifier- | shogun: Streaming Hashed Doc Features | 08:35 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * aef8ebc / / (3 files): https://github.com/shogun-toolbox/shogun/commit/aef8ebc05cfdbad2d60788195a1c050705a133ad | 08:35 |
shogun-notifier- | shogun: Merge pull request #1216 from van51/feature/hashing | 08:35 |
shogun-notifier- | shogun: | 08:35 |
shogun-notifier- | shogun: Streaming Hashed Doc Features | 08:35 |
shogun-buildbot | build #1221 of bsd1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1221 | 08:43 |
shogun-buildbot | build #1343 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/1343 blamelist: Soeren Sonnenburg <sonne@debian.org> | 09:12 |
-!- foulwall` [~user@2001:da8:215:6100:44cf:cdc9:76bb:c777] has joined #shogun | 09:34 | |
-!- foulwall [~user@2001:da8:215:6100:ad3e:1706:207e:6816] has quit [Ping timeout: 245 seconds] | 09:35 | |
shogun-buildbot | build #1344 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/1344 blamelist: Evgeniy Andreev <gsomix@gmail.com>, Soeren Sonnenburg <sonne@debian.org>, van51 <vangelis_51@hotmail.com> | 09:42 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 10:04 | |
shogun-notifier- | shogun-demo: foulwall :master * ea70146 / / (26 files): https://github.com/shogun-toolbox/shogun-demo/commit/ea70146277bde42117b8271a6257f5f24e4a5014 | 10:13 |
shogun-notifier- | shogun-demo: 1. add a data loader. 2. change the json structure, add a colomn for label value 3. fix the legend digit. 4. add auto-scale to the axis 5. remove the shogun-demo submodule | 10:13 |
shogun-notifier- | shogun-demo: Soeren Sonnenburg :master * 749255c / / (26 files): https://github.com/shogun-toolbox/shogun-demo/commit/749255cc0c32ff275f008f5b393ecce35823d0fe | 10:13 |
shogun-notifier- | shogun-demo: Merge pull request #12 from foulwall/perceptron | 10:13 |
shogun-notifier- | shogun-demo: | 10:13 |
shogun-notifier- | shogun-demo: data importer | 10:13 |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun | 10:28 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 10:28 | |
lisitsyn | flight time! | 11:02 |
lisitsyn | in 4 hours but I am packing notebook ;) | 11:02 |
-!- lisitsyn [~lisitsyn@31-28-52-76.clients.tlt.100megabit.ru] has left #shogun [] | 11:02 | |
-!- HeikoS [~heiko@nat-177-161.internal.eduroam.ucl.ac.uk] has joined #shogun | 11:08 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 11:08 | |
-!- Netsplit *.net <-> *.split quits: shogun-notifier-, sonne|work, zxtx | 11:12 | |
-!- Netsplit over, joins: shogun-notifier-, zxtx, sonne|work | 11:15 | |
@iglesiasg | sonne|work: I should really write a few unit tests for knn | 11:38 |
sonne|work | yeah refactoring is dangerous without | 11:39 |
@iglesiasg | sonne|work, HeikoS : are there already unit tests reading input data from files? | 11:40 |
@iglesiasg | to avoid using too trivial data | 11:40 |
sonne|work | don't understand the question | 11:40 |
sonne|work | iglesiasg: I think unit tests should rather work without loading data but just using some data generators | 11:41 |
@iglesiasg | sonne|work: difficult to test then functions like nearest neighbors where I want to see if the indices of the nearest neighbors are correct | 11:42 |
@iglesiasg | sonney2k: I read data generators as random data generators | 11:42 |
-!- van51 [~van51@79.131.114.98] has joined #shogun | 11:43 | |
sonne|work | iglesiasg: or you create the toy data your own | 11:43 |
@iglesiasg | sonne|work: then we fall back to the too trivial data problem I mentioned above :) | 11:44 |
sonne|work | iglesiasg: well knn still has integration tests | 11:44 |
@iglesiasg | sonney2k: are they testing for correctness of the results? | 11:45 |
@iglesiasg | oops bad name | 11:47 |
@iglesiasg | sonne|work: ^ | 11:47 |
sonne|work | iglesiasg: well they make sure that the same result comes out | 11:48 |
sonne|work | so if the previous was correct | 11:48 |
@iglesiasg | got it, it seems enough then | 11:49 |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-bxwsbyychdryheqd] has quit [Quit: Connection closed for inactivity] | 12:02 | |
@HeikoS | iglesiasg: hey | 12:16 |
@HeikoS | iglesiasg: usually, the unit tests are more for testing basic functionality, so fixed, very simple data | 12:16 |
@HeikoS | low dimensions etc | 12:17 |
@HeikoS | just to ensure things work | 12:17 |
@HeikoS | the big problems are tested in integration tests | 12:17 |
@HeikoS | votjakovr: hi! | 12:28 |
@iglesiasg | HeikoS: thanks HeikoS, it makes sense | 12:29 |
@HeikoS | iglesiasg: what I always do when I have reference implementations: set up a very small example that covers all cases and then copy/paste the result numbers to unit-test assertions | 12:29 |
votjakovr | HeikoS: hi | 12:30 |
@HeikoS | iglesiasg: have a look at votjakovr unit tests for gps, thats a good way to go | 12:30 |
@HeikoS | votjakovr: hey, how are things? | 12:30 |
votjakovr | HeikoS: not so bad, i'm working on stuff for numerical integration | 12:31 |
@HeikoS | votjakovr: nice, any particular problems with that? challenges? | 12:32 |
votjakovr | HeikoS: no problems, but source of alglib is a real challenge :) | 12:34 |
@HeikoS | votjakovr: yes thats true, how do you proceed in integrating it? :) | 12:34 |
votjakovr | HeikoS: i created mathematics/Integration with static method for integration with Gauss-Kronrod quadrature and mathematics/IntegrableFunction | 12:37 |
@HeikoS | votjakovr: good! | 12:38 |
@HeikoS | votjakovr: ok, also I wanted to let you know that I will go to Berlin today. I will work from there, but probably will not be around all day. If you cannot catch me in IRC, send a mail. I will also prepare the workshop slides a bit more and probably use your probit classifier :) | 12:39 |
votjakovr | HeikoS: ok, have a nice travel ;) | 12:40 |
@HeikoS | votjakovr: thanks :) | 12:40 |
-!- lambday [67157f37@gateway/web/freenode/ip.103.21.127.55] has joined #shogun | 12:43 | |
lambday | HeikoS: hi | 12:43 |
@HeikoS | lambday: hi! | 12:43 |
@HeikoS | how are things? | 12:43 |
lambday | HeikoS: I added SGSparseMatrix... copy and equals seems to work, unit-tested | 12:43 |
lambday | HeikoS: sending the PR, please have a look | 12:44 |
lambday | HeikoS: also, added "add" wrappers | 12:44 |
@HeikoS | lambday: added it where? | 12:44 |
@HeikoS | ah wow | 12:44 |
@HeikoS | in the parameter framework? | 12:44 |
@HeikoS | niiiiice :D | 12:44 |
lambday | HeikoS: yes | 12:44 |
@HeikoS | send it ill have alook | 12:44 |
lambday | HeikoS: :D first have a look and check if its okay | 12:45 |
lambday | HeikoS: sent | 12:46 |
-!- foulwall` [~user@2001:da8:215:6100:44cf:cdc9:76bb:c777] has quit [Ping timeout: 264 seconds] | 12:46 | |
@HeikoS | lambday: checking | 12:46 |
@HeikoS | lambday: wow nice with the unit tests | 12:49 |
@HeikoS | makes it quite easy to see that things work | 12:49 |
@HeikoS | did you check them for leaks? | 12:50 |
lambday | HeikoS: yes :) | 12:50 |
lambday | yes | 12:50 |
@HeikoS | lambday: can you quickly explain me the lines | 12:50 |
@HeikoS | /* for ST_SPARSE allocate only for a vector of m_length_y */+ if (m_datatype.m_stype==ST_SPARSE)+ num_bytes=*m_datatype.m_length_y * m_datatype.sizeof_stype(); | 12:50 |
@HeikoS | lambday: ah no | 12:50 |
@HeikoS | I see | 12:51 |
lambday | lol.. I typed a long sentence and deleted | 12:51 |
lambday | :D | 12:51 |
lambday | HeikoS: this looks reasonable, no? | 12:51 |
lambday | also had to change the point where it starts reading the next vector | 12:52 |
lambday | cause iteration is over SGSparseVector<>* | 12:52 |
lambday | HeikoS: since copy works, so clone should also work once I add it as a param in a class, right? | 12:53 |
-!- foulwall [~foulwall@117.136.0.204] has joined #shogun | 12:53 | |
lambday | I can test that | 12:53 |
@HeikoS | lambday: yep looks all reasonable | 12:56 |
lambday | phew! :D | 12:56 |
@HeikoS | lambday: yes that would be good to add a unit test for clone for a class with a sgsparse matrix as parameter | 12:56 |
@HeikoS | lambday: but dont commit, just test it | 12:56 |
lambday | HeikoS: alright | 12:57 |
@HeikoS | wiking: will you finish the unit tests some time? Otherwise I will just hack something myself | 12:57 |
@HeikoS | lambday: we will soon have clone/equals test for all objects | 12:57 |
@HeikoS | so if you add a class which has a sgpsarsematrix member | 12:57 |
@HeikoS | it will automagically be tested | 12:57 |
lambday | HeikoS: that's great! | 12:57 |
van51 | sonne|work: hey | 12:58 |
@HeikoS | lambday: btw I am going to Berlin today. I will be working from there a bit, but not be around all day. so if you cannot catch me in irc, send me an email | 12:58 |
van51 | sonne|work: I was thinking of start a vw parser today, as it was in Olivier's milestones | 12:58 |
lambday | HeikoS: okay.. for the workshop? | 12:58 |
@HeikoS | lambday: exactly | 12:58 |
van51 | sonne|work: but I see there is already a vw parser and a class for vw examples | 12:58 |
shogun-notifier- | shogun: lambday :develop * d78cca8 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/d78cca84b554271f81b171d551db93dc55115846 | 12:58 |
shogun-notifier- | shogun: added sparse matrix under parameter framework | 12:58 |
shogun-notifier- | shogun: Heiko Strathmann :develop * fd2ba03 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/fd2ba0370749c27bc3d34006a159bd9ca5a58678 | 12:59 |
shogun-notifier- | shogun: Merge pull request #1227 from lambday/feature/log_determinant | 12:59 |
shogun-notifier- | shogun: | 12:59 |
shogun-notifier- | shogun: added sparse matrix under parameter framework | 12:59 |
van51 | sonne|work: should I just make use of those and create another dot features class? | 12:59 |
lambday | HeikoS: awesome! I'll get to see some video I guess | 12:59 |
@HeikoS | lambday: YES :) | 12:59 |
@HeikoS | we might have a video stream | 12:59 |
lambday | :D | 13:00 |
@HeikoS | lambday: ok nice, will leave now -- ill keep my phone so I can merge any PRs, see you later! | 13:12 |
lambday | HeikoS: alright :) | 13:12 |
-!- lambday [67157f37@gateway/web/freenode/ip.103.21.127.55] has quit [Ping timeout: 250 seconds] | 13:17 | |
-!- HeikoS [~heiko@nat-177-161.internal.eduroam.ucl.ac.uk] has left #shogun [] | 13:17 | |
-!- foulwall` [~user@2001:da8:215:6100:4466:5e8c:5b16:50d8] has joined #shogun | 13:31 | |
-!- foulwall [~foulwall@117.136.0.204] has quit [Quit: AndroIRC - Android IRC Client ( http://www.androirc.com )] | 13:32 | |
shogun-buildbot | build #1345 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/1345 blamelist: lambday <heavensdevil6909@gmail.com> | 14:01 |
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.] | 14:14 | |
shogun-buildbot | build #1346 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/1346 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 14:17 |
-!- foulwall` [~user@2001:da8:215:6100:4466:5e8c:5b16:50d8] has left #shogun ["ERC Version 5.3 (IRC client for Emacs)"] | 14:39 | |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 240 seconds] | 14:45 | |
-!- van51 [~van51@79.131.114.98] has quit [Read error: Connection reset by peer] | 14:53 | |
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun | 14:54 | |
-!- nube [~rho@49.244.94.201] has joined #shogun | 15:06 | |
-!- foulwall` [~user@2001:da8:215:6100:4466:5e8c:5b16:50d8] has joined #shogun | 15:48 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 15:59 | |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun | 16:02 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 16:02 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["ERC Version 5.3 (IRC client for Emacs)"] | 16:15 | |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 268 seconds] | 16:27 | |
-!- foulwall` [~user@2001:da8:215:6100:4466:5e8c:5b16:50d8] has quit [Ping timeout: 264 seconds] | 16:36 | |
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun | 16:37 | |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun | 16:42 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 16:42 | |
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has left #shogun ["QUIT :Leaving."] | 17:01 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 17:19 | |
shogun-notifier- | shogun-demo: foulwall :master * a3a4c81 / demo/kernel_matrix.py,templates/gp/index.html,templates/kernel_matrix/index.html: https://github.com/shogun-toolbox/shogun-demo/commit/a3a4c81a6277f84e4371b7cb6c073ad21a84f5bb | 17:19 |
shogun-notifier- | shogun-demo: 1. fix kernel_matrix in new style. 2. fix the curve animation in gp demo. | 17:19 |
shogun-notifier- | shogun-demo: Soeren Sonnenburg :master * f75d935 / demo/kernel_matrix.py,templates/gp/index.html,templates/kernel_matrix/index.html: https://github.com/shogun-toolbox/shogun-demo/commit/f75d9357122e9d13618da9f40261b505856de525 | 17:19 |
shogun-notifier- | shogun-demo: Merge pull request #13 from foulwall/kernel_matrix | 17:19 |
shogun-notifier- | shogun-demo: | 17:19 |
shogun-notifier- | shogun-demo: kernel_matrix | 17:19 |
@iglesiasg | sonne|work: just in case you wanted to check ;) | 17:21 |
sonne|work | iglesiasg: I hope you did :) | 17:22 |
shogun-notifier- | shogun: Fernando Iglesias :develop * 2a59ace / src/shogun/multiclass/KNN.cpp,src/shogun/multiclass/KNN.h: https://github.com/shogun-toolbox/shogun/commit/2a59acebc3036476dd00e9cbf4b16d7102a980a3 | 17:23 |
shogun-notifier- | shogun: Add method in KNN to obtain nearest neighbors. | 17:23 |
shogun-notifier- | shogun: Internally, the apply now calls the new method to obtain the nearest | 17:23 |
shogun-notifier- | shogun: neighbors first and then computes the labels. | 17:23 |
shogun-notifier- | shogun: Fernando Iglesias :develop * 28e5183 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/28e518379a79a097692a21fa50d2adf7b2621b1c | 17:23 |
shogun-notifier- | shogun: libshogun simple kNN example with random data | 17:23 |
shogun-notifier- | shogun: Fernando Iglesias :develop * fb0c239 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/fb0c2399bbf4c33c59b1a4c8194c93836cbcbc17 | 17:23 |
shogun-notifier- | shogun: Use nearest neighbors method in classify for multiple k. | 17:23 |
shogun-notifier- | shogun: Also, avoid code duplication by introducing choose for multiple k | 17:23 |
shogun-notifier- | shogun: method, used in classify for multiple k w/ and w/o cover tree. | 17:23 |
shogun-notifier- | shogun: Fernando Iglesias :develop * 3c8ea00 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/3c8ea00915ba0462fb86b2d099ed396d33857878 | 17:23 |
shogun-notifier- | shogun: Merge pull request #1225 from iglesias/feature/knn_public_neighbors | 17:23 |
shogun-notifier- | shogun: | 17:23 |
shogun-notifier- | shogun: Feature/knn public neighbors | 17:23 |
-!- vgorbati [~vgorbati@91.216.173.29] has joined #shogun | 17:25 | |
-!- foulwall` [~user@2001:da8:215:c252:24ad:63e:d49e:8091] has joined #shogun | 17:26 | |
shogun-buildbot | build #1226 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1226 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 17:36 |
shogun-buildbot | build #1227 of bsd1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1227 | 17:50 |
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving] | 18:11 | |
shogun-buildbot | build #1347 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/1347 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 18:16 |
-!- foulwall` [~user@2001:da8:215:c252:24ad:63e:d49e:8091] has quit [Ping timeout: 264 seconds] | 18:32 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 206a880 / src/shogun/ (6 files): https://github.com/shogun-toolbox/shogun/commit/206a880f5af85c4a6cf6c8f7387a6dd262b25a0e | 18:38 |
shogun-notifier- | shogun: fix get_name in MKL classes fixing serialization / parameter framework | 18:38 |
shogun-buildbot | build #1348 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/1348 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 18:43 |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has joined #shogun | 18:50 | |
sonne|work | pickle27: hey | 18:51 |
pickle27 | hey | 18:51 |
sonne|work | pickle27: please also do the audio application | 18:51 |
pickle27 | oh yeah Im planning to do both! | 18:51 |
sonne|work | pickle27: I find that more fancy than some bioinformatics stuff | 18:51 |
sonne|work | very good | 18:51 |
pickle27 | the audio example itself shouldn't be too tough | 18:52 |
pickle27 | its very similar to the graphical example I made for the modular python | 18:52 |
pickle27 | the tough part will actually be the user interface which I have less experience with | 18:53 |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving] | 19:16 | |
sonne|work | pickle27: I think foulwall can help here | 19:19 |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has quit [Quit: Leaving] | 19:20 | |
sonne|work | pickle27: you should ask him | 19:20 |
-!- vgorbati [~vgorbati@91.216.173.29] has quit [Quit: vgorbati] | 19:27 | |
shogun-buildbot | build #1349 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/1349 blamelist: Soeren Sonnenburg <sonne@debian.org> | 19:30 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 19:38 | |
@sonney2k | votjakovr, did you have a chance to look at the regression_gaussian_process_modular.py setting 1/2 failure? | 19:41 |
votjakovr | sonney2k: hi, i found, that this commit cd27d057b1c1a7cd2e5a22741c9b7be1e1d8e896 breaks it, and now i'm trying to figure out why | 19:53 |
@sonney2k | votjakovr, nice thanks. | 19:55 |
-!- vgorbati [~vgorbati@91.216.173.29] has joined #shogun | 19:59 | |
votjakovr | sonney2k: i think, i have find what gives different results :) | 20:16 |
votjakovr | sonney2k: i've just sent a PR | 20:32 |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun | 20:34 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 20:34 | |
naywhayare | sonney2k: is there a k-means integration test or something so I can test if my patch works, or do I have to take a look at the results and compare with output of something else? | 20:45 |
naywhayare | it's a two-character patch :) | 20:45 |
@sonney2k | votjakovr, cool! | 20:46 |
@sonney2k | votjakovr, so it is not really a bug right? | 20:50 |
votjakovr | sonney2k: yep | 20:51 |
votjakovr | sonney2k: but now regression_gaussian_process_modular.py is OK on my machine | 20:52 |
@sonney2k | votjakovr, yeah numerics :( it is good that you've found it | 20:54 |
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun | 20:55 | |
votjakovr | sonney2k: yep :) | 20:56 |
shogun-notifier- | shogun: Roman Votyakov :develop * 55adddc / src/shogun/machine/gp/ExactInferenceMethod.cpp: https://github.com/shogun-toolbox/shogun/commit/55adddc9506cdc54573897564acb2a398359fe8a | 21:00 |
shogun-notifier- | shogun: fix bug which caused different results in integration test | 21:00 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 044d4ed / src/shogun/machine/gp/ExactInferenceMethod.cpp: https://github.com/shogun-toolbox/shogun/commit/044d4ed31b7cb471b34ff6eb01a660be3ef20f50 | 21:00 |
shogun-notifier- | shogun: Merge pull request #1228 from votjakovr/develop | 21:00 |
shogun-notifier- | shogun: | 21:00 |
shogun-notifier- | shogun: potentially fix integration test | 21:00 |
@sonney2k | naywhayare, yes there is | 21:01 |
@sonney2k | naywhayare, just do a pull request the buildbot will then build your code :) | 21:02 |
naywhayare | ok. I live in the past, so let me learn a bit about git :) | 21:02 |
@sonney2k | naywhayare, actually *I* haven't done any PR but AFAIK it is just creating a branch derived from develop and then you can press a pull request knob on github | 21:04 |
@sonney2k | van51, how is it going? | 21:05 |
naywhayare | haha. I figure it's time for me to learn about this stuff anyway... git isn't exactly "new technology" anymore :) | 21:05 |
van51 | sonney2k: hey | 21:06 |
@sonney2k | naywhayare, well github is making things easy https://help.github.com/articles/using-pull-requests | 21:06 |
van51 | sonney2k: I 've only started a bit the nwordtokenizer | 21:07 |
van51 | sonney2k: I was planning on working thourgh the night, it's less hot :) | 21:07 |
@sonney2k | van51, what is the temperature down there? | 21:07 |
van51 | sonney2k: should I proceed with what I told you before? I mean a DotFeatures class based on the current vw_example and vw_parser classes | 21:07 |
van51 | sonney2k: it's around 30 degrees now | 21:08 |
@sonney2k | woah | 21:08 |
@sonney2k | yeah berlin is a chilly place | 21:08 |
@sonney2k | 23 C now | 21:08 |
van51 | I'm jealous | 21:08 |
van51 | although you get used to it | 21:08 |
@sonney2k | van51, I actaully don't know what this vw* stuff is for | 21:09 |
van51 | during the day is the worst | 21:09 |
van51 | sonney2k: there is a whole bunch of classes | 21:09 |
van51 | sonney2k: for vw stuff | 21:09 |
van51 | sonney2k: I think it was an integration of vw | 21:09 |
@sonney2k | van51, yeah we tried that 2 gsoc's ago but nowadays I would rather turn vw into a library and have some shallow shogun wrappers around it | 21:10 |
@sonney2k | anyways I don't really know why we have that | 21:10 |
@sonney2k | I mean vw is also only using an example == (features, plus one label) | 21:11 |
@sonney2k | and it can use shogun's streamingfeatures | 21:11 |
van51 | sonney2k: Olivier's and Benoit's suggestion was to have datasets in vw format and then build DotFeatures object with them that hash the namespace | 21:12 |
van51 | sonney2k: like vw does | 21:12 |
@sonney2k | van51, I don't know what vw does here and what that means - could you elaborate? | 21:13 |
van51 | sonney2k: from what I understand, each feature is accompanied by a name which is hashed and that is how they find where to store its value | 21:14 |
* sonney2k wonders if lisitsyn made it to berlin | 21:15 | |
@sonney2k | van51, so it is hashed *together* with the feature? | 21:15 |
van51 | sonney2k: something like that, I;m trying to find the page where I had read it | 21:16 |
@sonney2k | van51, if that is the case we should add features supporting this. it is kind of a (sparse) feature matrix plus some category label | 21:19 |
van51 | sonney2k: I can ask Benoit for more details on our hangout | 21:20 |
@sonney2k | van51, yeah ask him | 21:24 |
@sonney2k | iglesiasg, is there no way to avoid the eigen3 code in the .h? | 21:25 |
@iglesiasg | sonney2k: I am not sure of it if I want to use it in the arguments of the private methods | 21:26 |
@iglesiasg | sonney2k: can that become a problem? | 21:26 |
@iglesiasg | sonney2k: I can add mathematics/eigen3 instead and use typedef for MatrixXd | 21:26 |
@sonney2k | iglesiasg, it is all about compile speed with swig | 21:28 |
@sonney2k | iglesiasg, do you know how sergey is hiding tapkee's eigen3 use in shogun? | 21:29 |
@iglesiasg | sonney2k: the methods are implemented in hpp methods that are called from the cpp | 21:30 |
@iglesiasg | there are also some code in between tapkee and Shogun | 21:30 |
@sonney2k | swig is only wrapping the specified .h files | 21:31 |
@sonney2k | so none of this is visible directly to swig but will still be needed at compile time | 21:31 |
@sonney2k | so an option would be to have some class in your code that does the eigen3 std:: etc stuff but that is not %included from swig | 21:32 |
@sonney2k | and you only need a member of that class and all access to the class is in the .cpp file | 21:33 |
@sonney2k | van51, ohh and please add the benchmark code to benchmarks/ | 21:33 |
@sonney2k | van51, we should really add more benchmarks also for all the other dot features | 21:34 |
van51 | sonney2k: yea the representation is like that. you have a sparse feature matrix and the name of each feature is hashed to find its index in the matrix | 21:34 |
@sonney2k | van51, sry parse error | 21:35 |
van51 | sonney2k: I 've jotted it down to push the benchmark | 21:35 |
@sonney2k | kk | 21:35 |
@sonney2k | so you have a sparse matrix ok | 21:36 |
van51 | sonney2k: I just have to add some fixed strings as it was reading a file | 21:36 |
@sonney2k | but what about the string? | 21:36 |
@sonney2k | you have one for each example or each feature? | 21:36 |
@sonney2k | and why a string not a number | 21:36 |
van51 | sonney2k: each feature has a string name | 21:37 |
van51 | sonney2k: which is hashed pretty much like we were doing for the tokens | 21:37 |
@iglesiasg | sonney2k: mm I think it is weird to create a new class to implement private methods | 21:37 |
@sonney2k | van51, ok but it would make sense to have just a nubmer instead of a string | 21:38 |
van51 | sonney2k: https://github.com/JohnLangford/vowpal_wabbit/wiki/Tutorial#a-first-dataset | 21:39 |
@sonney2k | iglesiasg, yeah well but it would solve it | 21:39 |
* sonney2k afk | 21:39 | |
@iglesiasg | sonney2k: what is the part exactly that would make compilation slower? | 21:40 |
@iglesiasg | sonney2k: include set and vector? include Eigen/Dense in the header? | 21:40 |
@iglesiasg | sonney2k: if it is about Eigen, we already have a header that includes them, mathematics/eigen3 so I could just include that one instead | 21:41 |
@sonney2k | iglesiasg, but mathematics/eigen3 is not included from swig | 21:44 |
@iglesiasg | sonney2k: does that mean I cannot include it from the header? | 21:45 |
@sonney2k | iglesiasg, well your class would be the only one doing it | 21:47 |
@sonney2k | so try to avoid it | 21:47 |
@sonney2k | van51, pretty complex format | 21:50 |
@sonney2k | van51, got it now though | 21:50 |
@sonney2k | van51, so an extra category vector would do it | 21:50 |
van51 | sonney2k: what for? | 21:52 |
@iglesiasg | sonney2k: http://stackoverflow.com/questions/5665471/avoiding-declaring-private-functions-in-class-header-files-c | 21:53 |
@iglesiasg | sonney2k: see the first answer. Something like this would be good | 21:53 |
@sonney2k | van51, instead of having a name for each feature you could use a number | 21:53 |
van51 | sonney2k: the name is used in the data file that is in that format | 21:55 |
@sonney2k | iglesiasg, yes exactly that is what I had in mind | 21:56 |
van51 | sonney2k: once read an example would consist of pairs (index,value) | 21:56 |
@iglesiasg | sonney2k: all right then, it feels better to see that it has name and it seems something people already use :) http://en.wikipedia.org/wiki/Opaque_pointer#C.2B.2B | 21:56 |
@sonney2k | van51, ahh OK single pass over the data so one could as well do expensive ops on the strings | 21:57 |
@sonney2k | van51, btw does StreamingVwFeatures.h already do this? | 21:57 |
van51 | sonney2k: yea I think | 21:58 |
@sonney2k | iglesiasg, hmmmhh maybe it can even help us to reduce swig wrapper size | 21:58 |
van51 | sonney2k: I had only seend the VwParser | 21:58 |
@sonney2k | iglesiasg, we cannot use nested classes though due to limitations of the interfaces swig is wrapping to | 21:59 |
@iglesiasg | sonney2k: but this does not need to be a nested class, right? | 22:00 |
@sonney2k | van51, you might follow a different approach | 22:00 |
@iglesiasg | with nested class I understand class Foo { class bar {}; }; | 22:00 |
@sonney2k | van51, consider you have for each feature some category number to be more concrete: | 22:01 |
@iglesiasg | this seems to be possible just having one pointer to Bar in Foo | 22:01 |
@sonney2k | consider you use DenseFeatures | 22:01 |
@sonney2k | you could then create CategorialDenseFeatures out of it by adding a vector of size #dims - which you then do hash together with its index | 22:02 |
shogun-buildbot | build #1350 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/1350 | 22:03 |
@sonney2k | same with sparse features | 22:03 |
@sonney2k | votjakovr, success!! | 22:03 |
@sonney2k | iglesiasg, yes | 22:04 |
@sonney2k | iglesiasg, I think using this idiom would really help us to improve shogun's compile speed and readability! | 22:05 |
van51 | sonney2k: you mean an extra vector? with eg random indices for each dimension? | 22:06 |
votjakovr | sonney2k: good :) | 22:07 |
van51 | sonney2k: and then you would hash value with index? for each feature | 22:07 |
@sonney2k | votjakovr, I think the bot will be all green tonight! | 22:07 |
@sonney2k | van51, yes an extra vector so one value per dimension with the 'true' dimension being computed based on one value of the extra vector (corresponding to one dimension) and the dimension index | 22:08 |
@sonney2k | van51, based on this a sparse feature matrix can be created from any categorial one | 22:12 |
van51 | sonney2k: I think the only problem would be if you a large number of dimensions | 22:12 |
van51 | have* | 22:12 |
@sonney2k | van51, for sparse you mean? | 22:13 |
@sonney2k | van51, but note that the learned w vector is dense anyway | 22:13 |
van51 | sonney2k: for that extra vector | 22:13 |
@sonney2k | so it is not that much of a problem | 22:13 |
van51 | sonney2k: can't the dimension index be hashed directly? what good is the extra vector for? | 22:16 |
naywhayare | okay, I think I did this pull request right: https://github.com/shogun-toolbox/shogun/pull/1230 | 22:17 |
naywhayare | sonney2k: so you say this will build automatically and run integration tests? | 22:17 |
@sonney2k | van51, sure you could create a sparse matrix immediately | 22:20 |
@sonney2k | naywhayare, yeah look at https://github.com/shogun-toolbox/shogun/pull/1230 | 22:20 |
@sonney2k | and click details on Determining merge status — The Travis CI build is in progress (Details) | 22:20 |
@sonney2k | then you will see the tests being run | 22:20 |
naywhayare | spiffy! I like it | 22:20 |
@sonney2k | yeah it is cool stuff | 22:21 |
van51 | sonney2k: cool I'll get started with that then | 22:21 |
van51 | sonney2k: and try to arrange a hangout with Benoit for tomorrow | 22:21 |
@sonney2k | van51, btw gsomix' line reader is in | 22:21 |
van51 | sonney2k: yea I saw it :0 | 22:21 |
van51 | :) * | 22:21 |
@sonney2k | together with support for your delimiter features | 22:21 |
van51 | he also made some imporvements! | 22:21 |
@sonney2k | tokenizer I mean | 22:22 |
* sonney2k tired | 22:22 | |
van51 | yes yes :D | 22:22 |
@sonney2k | so parsing ascii should be much much easier now | 22:22 |
@sonney2k | if it is all correct | 22:22 |
van51 | sonney2k: idd | 22:22 |
@sonney2k | of course | 22:22 |
van51 | sonney2k: but not in streaming features | 22:22 |
van51 | sonney2k: they use the streamingAsciiFile class | 22:22 |
@sonney2k | van51, well that has to be converted! | 22:23 |
van51 | sonney2k: hehe I was thinking you'd say that :) | 22:23 |
@sonney2k | van51, we only want one well tested line reader | 22:23 |
@sonney2k | and in shogun we have currently 10 or so ;) | 22:23 |
@sonney2k | van51, heh so then what are you waiting for :P | 22:24 |
van51 | sonney2k: hehe those streaming I/O classes look a bit messy | 22:24 |
@sonney2k | van51, yeah they were written pre-existince of SGVector & friends | 22:24 |
van51 | sonney2k: all in due time :D | 22:24 |
van51 | sonney2k: I don't mind starting on that as well | 22:25 |
@sonney2k | van51, the more you do the more you will realize that there is even more to do :) | 22:26 |
van51 | sonney2k: so you're saying I should be doing less :p | 22:27 |
van51 | less is more | 22:27 |
@sonney2k | van51, to the contrary give up real live that is all we want :-)\ | 22:27 |
van51 | they have a reason for saying that :p | 22:27 |
van51 | and here I was thinking you were over-demanding :) | 22:28 |
@sonney2k | no way | 22:28 |
van51 | but to be honest not much else to do now | 22:28 |
van51 | most friends are sitting exams | 22:28 |
van51 | ! | 22:28 |
@sonney2k | see | 22:28 |
@sonney2k | you have no excuse | 22:28 |
@sonney2k | van51, why am I still talking to you then :P | 22:29 |
van51 | sonney2k: dunno! I am starting :) | 22:29 |
@sonney2k | naywhayare, btw the fix seems indeed a surgical operation | 22:43 |
naywhayare | :) | 22:44 |
naywhayare | it took a long time to understand the basic ideas of what was going on | 22:44 |
@sonney2k | all that condensed into 2 ascii chars | 22:44 |
@sonney2k | and then forgotten to be rediscovered by other poor individuals | 22:45 |
naywhayare | so this is all the same benchmarking program I used with k-means for the mlpack paper | 22:45 |
naywhayare | the most confusing thing to me is why it worked before | 22:45 |
naywhayare | with shogun 1.0.0 | 22:45 |
-!- lisitsyn [~lisitsyn@p5099e2ec.dip0.t-ipconnect.de] has joined #shogun | 22:45 | |
lisitsyn | HA-HA! I am in germany | 22:46 |
naywhayare | I suspect that I compiled it without asserts (to be as fast as possible) and it just wrote over some random memory bits somewhere... | 22:46 |
@sonney2k | lisitsyn, yeaaah! | 22:46 |
lisitsyn | I don't understand what people say! | 22:46 |
@sonney2k | lisitsyn, you managed to catch your plane! | 22:47 |
lisitsyn | sonney2k: oh it was easy | 22:47 |
@sonney2k | lisitsyn, now you finally understand how I feel talking to you | 22:47 |
lisitsyn | sonney2k: bahh you make me sad with such words | 22:47 |
@sonney2k | lisitsyn, why? did the other one have a delay ? | 22:47 |
lisitsyn | sonney2k: no it was really fast | 22:47 |
lisitsyn | customs is just a walk through a corridor | 22:48 |
lisitsyn | in SXF it was the same btw | 22:48 |
@iglesiasg | lisitsyn: hello! | 22:49 |
lisitsyn | iglesiasg: hey | 22:49 |
@iglesiasg | lisitsyn: how is Germany then? | 22:49 |
lisitsyn | iglesiasg: I am scared! :D | 22:50 |
@iglesiasg | lisitsyn: haha why? | 22:51 |
lisitsyn | iglesiasg: I am afraid with communicating with people a bit! | 22:52 |
@iglesiasg | lisitsyn: hehe but German people should be good at English! | 22:54 |
lisitsyn | iglesiasg: it is me who is bad at english! | 22:54 |
@iglesiasg | lisitsyn: lol I don't think so! | 22:55 |
lisitsyn | iglesiasg: believe me! | 22:55 |
@iglesiasg | lisitsyn: probably you are just a bit shy to speak right away | 22:55 |
@sonney2k | lisitsyn, don't be afraid most germans don't speak english nor german :) | 22:59 |
@iglesiasg | nor German lol | 22:59 |
lisitsyn | sonney2k: taxi driver speaked russian :D | 23:02 |
@sonney2k | lisitsyn, hehe | 23:04 |
shogun-notifier- | shogun: ryan :develop * a4adc15 / src/shogun/clustering/KMeans.cpp: https://github.com/shogun-toolbox/shogun/commit/a4adc155f550d37aeb5c6f49d1f1372ef166705f | 23:04 |
shogun-notifier- | shogun: Only calculate distances to the correct number of centroids when using prespecified mus. | 23:04 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * b33d6b4 / src/shogun/clustering/KMeans.cpp: https://github.com/shogun-toolbox/shogun/commit/b33d6b4243b7b52273d25fc53b05bd1422fdacfe | 23:04 |
shogun-notifier- | shogun: Merge pull request #1230 from rcurtin/develop | 23:04 |
shogun-notifier- | shogun: | 23:04 |
shogun-notifier- | shogun: Only calculate distances to the correct number of centroids when using prespecified mus. | 23:04 |
@sonney2k | naywhayare, ^merged | 23:04 |
naywhayare | thanks! | 23:07 |
naywhayare | I am glad I could help :) | 23:07 |
@sonney2k | thanks a lot | 23:11 |
@sonney2k | naywhayare, so now the class overloading with the predefined centers will work? | 23:11 |
@sonney2k | iglesiasg, could you please twitter about our video recording of the workshop & final program? | 23:11 |
naywhayare | sonney2k: yeah, it will, but there's no good interface to it. you have to overload KMeans and rewrite train_machine() and then call clustknb(true, initial_mus) | 23:13 |
@iglesiasg | sonney2k: done | 23:15 |
@sonney2k | naywhayare, well it is not problem to add a setter for initial mus | 23:17 |
-!- vgorbati [~vgorbati@91.216.173.29] has quit [Ping timeout: 264 seconds] | 23:17 | |
naywhayare | sonney2k: it'll require a little more refactoring than that, but yeah, it won't be too incredibly difficult | 23:18 |
@iglesiasg | sonney2k, lisitsyn retweet please guys | 23:18 |
van51 | iglesiasg: maybe the date should be specified on the tweet? or at the site | 23:19 |
@iglesiasg | van51: yes | 23:19 |
@iglesiasg | van51: let me do something about that :) | 23:19 |
@sonney2k | lisitsyn, have you seen serialhex is still alive :) | 23:20 |
van51 | iglesiasg: cool :) | 23:20 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["Fallen asleep!"] | 23:20 | |
@sonney2k | naywhayare, yeah sure one (short) train ride ;) | 23:20 |
lisitsyn | sonney2k: yes sure | 23:21 |
lisitsyn | iglesiasg: in a minute | 23:21 |
lisitsyn | retweeted | 23:21 |
@iglesiasg | thanks | 23:21 |
@iglesiasg | van51: date in both places now | 23:23 |
van51 | iglesiasg: good job :) | 23:23 |
@iglesiasg | van51: thank you for pointing it out! | 23:23 |
van51 | iglesiasg: heh np | 23:24 |
lisitsyn | okay I'll take some rest now | 23:25 |
lisitsyn | see you | 23:25 |
-!- lisitsyn [~lisitsyn@p5099e2ec.dip0.t-ipconnect.de] has left #shogun [] | 23:26 | |
shogun-buildbot | build #1028 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1028 blamelist: ryan <ryan@shoeshine.cc.gt.atl.ga.us> | 23:27 |
@sonney2k | $%*($%(*CF cygwin! | 23:28 |
@sonney2k | if I only knew what it's problem is | 23:29 |
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has quit [Read error: No route to host] | 23:45 | |
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun | 23:47 | |
--- Log closed Wed Jul 10 00:00:25 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!