--- Log opened Mon Aug 26 00:00:31 2013 | ||
shogun-buildbot | build #1310 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1310 blamelist: Soeren Sonnenburg <sonne@debian.org>, Kevin <kevinhughes27@gmail.com> | 00:02 |
---|---|---|
shogun-buildbot | build #1521 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1521 blamelist: Kevin <kevinhughes27@gmail.com> | 00:03 |
shogun-buildbot | build #49 of osx1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/osx1%20-%20libshogun/builds/49 blamelist: Kevin <kevinhughes27@gmail.com> | 00:09 |
shogun-buildbot | build #1491 of deb2 - static_interfaces is complete: Failure [failed compile cmdline_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1491 blamelist: Kevin <kevinhughes27@gmail.com> | 00:18 |
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has quit [Quit: Ex-Chat] | 00:21 | |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has joined #shogun | 00:22 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10604003 | 00:22 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has left #shogun [] | 00:22 | |
shogun-buildbot | build #987 of rpm1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/987 blamelist: Kevin <kevinhughes27@gmail.com> | 00:26 |
shogun-buildbot | build #1492 of deb2 - static_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1492 blamelist: Soeren Sonnenburg <sonne@debian.org> | 00:42 |
shogun-buildbot | build #1606 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1606 blamelist: Soeren Sonnenburg <sonne@debian.org> | 00:43 |
shogun-buildbot | build #1311 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1311 blamelist: Kevin <kevinhughes27@gmail.com> | 00:46 |
shogun-buildbot | build #1522 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1522 blamelist: Soeren Sonnenburg <sonne@debian.org> | 00:49 |
shogun-buildbot | build #50 of osx1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/osx1%20-%20libshogun/builds/50 blamelist: Soeren Sonnenburg <sonne@debian.org> | 00:50 |
shogun-buildbot | build #988 of rpm1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/988 blamelist: Soeren Sonnenburg <sonne@debian.org> | 01:23 |
shogun-buildbot | build #1312 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1312 blamelist: Soeren Sonnenburg <sonne@debian.org> | 01:29 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 02:15 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["WeeChat 0.4.0"] | 02:32 | |
-!- pickle27 [~kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun | 03:06 | |
-!- pickle27 [~kevin@d67-193-243-174.home3.cgocable.net] has left #shogun [] | 03:08 | |
-!- sonne|osx_ [~sonne@g225080152.adsl.alicedsl.de] has joined #shogun | 03:53 | |
-!- sonne|osx [~sonne@f053044032.adsl.alicedsl.de] has quit [Ping timeout: 264 seconds] | 03:55 | |
-!- sonne|osx_ is now known as sonne|osx | 03:55 | |
-!- lisitsyn [~lisitsyn@109-226-90-47.clients.tlt.100megabit.ru] has left #shogun [] | 04:59 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 07:21 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 4d0b496 / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/4d0b496966cc1738613b19d125cd5f66c89d5a8b | 07:21 |
shogun-notifier- | shogun: use octave 3.6 ppa | 07:21 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * d3541f5 / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/d3541f565be1caad7f9d41ffdebfe2adb546c212 | 07:45 |
shogun-notifier- | shogun: add ppa to defaults | 07:45 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * a2f7d25 / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/a2f7d256385d71753f1fe99a02d7874b26df633f | 07:54 |
shogun-notifier- | shogun: answer "yes" | 07:54 |
-!- sonne|osx [~sonne@g225080152.adsl.alicedsl.de] has quit [Quit: sonne|osx] | 08:01 | |
shogun-buildbot | build #1523 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1523 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:06 |
-!- travis-ci [~travis-ci@ec2-54-234-140-21.compute-1.amazonaws.com] has joined #shogun | 08:07 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10611905 | 08:07 |
-!- travis-ci [~travis-ci@ec2-54-234-140-21.compute-1.amazonaws.com] has left #shogun [] | 08:07 | |
shogun-buildbot | build #1493 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/1493 | 08:10 |
shogun-buildbot | build #1524 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1524 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:11 |
shogun-buildbot | build #1494 of deb2 - static_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1494 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:12 |
shogun-buildbot | build #1495 of deb2 - static_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1495 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:12 |
shogun-buildbot | build #1608 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1608 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:13 |
shogun-buildbot | build #989 of rpm1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/989 | 08:13 |
shogun-buildbot | build #1525 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1525 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:17 |
-!- sonne|osx [~sonne@89.204.139.45] has joined #shogun | 08:17 | |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has joined #shogun | 08:34 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10611905 | 08:34 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has left #shogun [] | 08:34 | |
shogun-buildbot | build #1313 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1313 blamelist: Soeren Sonnenburg <sonne@debian.org> | 08:44 |
-!- sonne|osx [~sonne@89.204.139.45] has quit [Quit: sonne|osx] | 08:45 | |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has joined #shogun | 09:06 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10612388 | 09:06 |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has left #shogun [] | 09:06 | |
-!- sonne|work [~sonnenbu@91-64-72-127-dynip.superkabel.de] has joined #shogun | 09:08 | |
shogun-buildbot | build #1607 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/1607 | 09:09 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:19 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 09:19 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 5dc2eb7 / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/5dc2eb773f5d30ddc20dc2cfd9d216c14b79a5ed | 09:21 |
shogun-notifier- | shogun: add another ppa for octave | 09:21 |
sonne|work | ohh man | 09:24 |
sonne|work | anyone on ubuntu 12.04 LTS here? | 09:24 |
@lisitsyn | not anymore | 09:25 |
sonne|work | I wish travis was not using ubuntu LTS | 09:25 |
sonne|work | stuff is too old | 09:25 |
@lisitsyn | sonne|work: I can install it to VM | 09:26 |
@lisitsyn | sonne|work: should I? | 09:27 |
sonne|work | lisitsyn: I am just trying to install octave 3.6 w/ shogun now: http://blogs.bu.edu/mhirsch/2012/08/octave-3-6-on-ubuntu-12-04/ | 09:27 |
sonne|work | and it fails... | 09:27 |
@lisitsyn | sonne|work: what fails? | 09:27 |
sonne|work | lisitsyn: well travis | 09:27 |
@lisitsyn | okay checking | 09:27 |
sonne|work | lisitsyn: no idea why it doesn't work | 09:28 |
@lisitsyn | ValueError: No JSON object could be decoded | 09:28 |
@lisitsyn | this one? | 09:28 |
shogun-buildbot | build #1314 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1314 blamelist: Soeren Sonnenburg <sonne@debian.org> | 09:28 |
shogun-buildbot | build #1609 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1609 blamelist: Soeren Sonnenburg <sonne@debian.org> | 09:28 |
@lisitsyn | sonne|work: I think ppa is not working | 09:29 |
@lisitsyn | sonne|work: https://launchpad.net/~mvanderkolff/+archive/octave-3.6 what about that ppa? | 09:29 |
shogun-buildbot | build #1496 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/1496 | 09:30 |
sonne|work | lisitsyn: well no idea - it is trial and error | 09:31 |
shogun-buildbot | build #1526 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1526 blamelist: Soeren Sonnenburg <sonne@debian.org> | 09:33 |
-!- thoralf [~thoralf@enki.zib.de] has joined #shogun | 09:41 | |
thoralf | Hey. | 09:42 |
sonne|work | thoralf: you ubuntu 12.04 LTS? | 09:44 |
thoralf | Only on my old laptop, which isn't available before tomorrow. | 09:45 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has joined #shogun | 10:01 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10614057 | 10:01 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has left #shogun [] | 10:01 | |
-!- iglesiasg [~iglesias@2001:6b0:1:1da0:19a:206d:eb6b:cdad] has joined #shogun | 10:03 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 10:04 | |
@lisitsyn | sonne|work: tapkee is in jmlr vol. 14 pages 2287-2291 :) | 10:09 |
@lisitsyn | should be ready soon we made final corrections | 10:10 |
shogun-buildbot | build #1315 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1315 blamelist: Soeren Sonnenburg <sonne@debian.org> | 10:11 |
@lisitsyn | HeikoS: https://github.com/shogun-toolbox/shogun/issues/1265 | 10:41 |
sonne|work | lisitsyn: congrats | 10:43 |
@lisitsyn | sonne|work: thanks | 10:43 |
@lisitsyn | sonne|work: please take a look https://github.com/shogun-toolbox/shogun/issues/1265 | 10:44 |
@lisitsyn | sonne|work: I do not suggest to use it yet as it looks messy but it is a way | 10:44 |
sonne|work | lisitsyn: I think we are in trouble with swig - we would need to create %template for every getter | 10:44 |
@lisitsyn | sonne|work: well we will have like n_of_types getters | 10:45 |
@lisitsyn | lets say we have 10 types | 10:46 |
@lisitsyn | so we have 10 getters and 10 setters for all possible parameters | 10:46 |
@lisitsyn | oops | 10:49 |
@lisitsyn | :D | 10:49 |
@iglesiasg | lisitsyn, *minima* concrete example :P? | 10:49 |
@iglesiasg | minimal* | 10:49 |
@lisitsyn | iglesiasg: yes it is like what is the minimal music style to music | 10:50 |
@lisitsyn | I mean too minimal | 10:50 |
@lisitsyn | :D | 10:50 |
@iglesiasg | hehe | 10:50 |
@lisitsyn | HeikoS: iglesiasg: sonne|work: please see the latest comment https://github.com/shogun-toolbox/shogun/issues/1265 | 10:54 |
@lisitsyn | please ask me if you didn't get the idea or see so major drawback I missed | 10:56 |
@lisitsyn | s/so/a/ | 10:57 |
@iglesiasg | lisitsyn, let me see | 10:57 |
-!- van51 [~van51@ppp-94-66-12-14.home.otenet.gr] has joined #shogun | 10:57 | |
sonne|work | lisitsyn: the get/set you implement in SGObject right? | 10:57 |
@lisitsyn | sonne|work: yes | 10:57 |
sonne|work | van51: hey there! How is it goind? | 10:57 |
@lisitsyn | get(Keyword<int>), get(Keyword<float>), get(Keyword<float-moat>) | 10:58 |
van51 | sonne|work: hello! | 10:58 |
@lisitsyn | whatever types | 10:58 |
van51 | sonne|work: first of all I 've got some results on the lang_detection | 10:58 |
@iglesiasg | lisitsyn, what if you try to do some_object.get/set(keywords.parameter, ...) | 10:58 |
sonne|work | lisitsyn: what do we do when we have say a combined kernel and want kernels as input? | 10:58 |
van51 | sonne|work: turns out with a larger hash size, the quadratics perform better | 10:58 |
@iglesiasg | lisitsyn, and that some_object doesn't have that member | 10:58 |
sonne|work | van51: excellent! | 10:58 |
van51 | sonne|work: I also checked a bit the misclassifications | 10:58 |
@iglesiasg | lisitsyn, with member I mean the keywords.parameter | 10:59 |
@lisitsyn | too many messages :D | 10:59 |
van51 | sonne|work: and some of the sentences were trash, because they were generated randomly | 10:59 |
@lisitsyn | sonne|work: Keyword<SGObject*> | 10:59 |
@lisitsyn | sonne|work: I don't see the problem could you please elaborate? | 10:59 |
@lisitsyn | iglesiasg: runtime error | 11:00 |
van51 | sonne|work: I've also started working on RandomFourierDotFeatures | 11:00 |
@iglesiasg | lisitsyn, it would be nicer compile time error I think, but I can live with that too | 11:00 |
sonne|work | lisitsyn: ok so we would do the type checking underneath? I mean if we want a CKernel we would need to do KeyWord<CKernel*> or do SGObject & checks? | 11:01 |
@iglesiasg | maybe what I just said doesn't make sense since this is to be used from python or so... | 11:01 |
@lisitsyn | sonne|work: I'd add checks but we can add keywords for ckernel | 11:01 |
@lisitsyn | sonne|work: runtime error is ok too I think but there is no trouble to add a keyword that needs CKernel* | 11:02 |
@lisitsyn | I would not add keywords for CMulticlassOCAS or whatever to reduce the number of exported keyword classes | 11:03 |
@lisitsyn | but it is totally ok to add Keyword<CKernel*> kernel; I believe | 11:03 |
@lisitsyn | iglesiasg: no it is possible | 11:03 |
@lisitsyn | iglesiasg: you'd need a new class for every keyword though | 11:04 |
@lisitsyn | :D | 11:04 |
@lisitsyn | but you are right compile time doesn't make a lot sense as python just checks the syntax | 11:04 |
sonne|work | lisitsyn: so what is the overhead - I mean we currently have a lot of stuff doing e.g. range checks or so in a for loop from local parameters | 11:07 |
@lisitsyn | sonne|work: it looks for me it is possible to make it quite fast | 11:08 |
@lisitsyn | with hash map | 11:08 |
@lisitsyn | we can ensure each keyword has unique hash | 11:08 |
@lisitsyn | method dispatching is fast (compile time in java for example) | 11:09 |
@lisitsyn | so we just have to look up for the value in map | 11:09 |
@iglesiasg | lisitsyn, good that hash tables are already in the C++ standard | 11:14 |
sonne|work | van51: sounds nice. I would be interested in actual results :) | 11:19 |
van51 | sonne|work: do you mean for the random fourier feats or for the lang detection? | 11:21 |
sonne|work | van51: lang | 11:22 |
@lisitsyn | iglesiasg: finally in 2011.. | 11:23 |
@lisitsyn | iglesiasg: a bit late ain't it? | 11:24 |
@lisitsyn | ;) | 11:24 |
@iglesiasg | lisitsyn, what? | 11:24 |
@iglesiasg | tapkee? | 11:24 |
@lisitsyn | iglesiasg: hash tables I mean | 11:25 |
@lisitsyn | ;) | 11:25 |
@iglesiasg | lisitsyn, aaah ok | 11:25 |
@iglesiasg | yeah | 11:25 |
@lisitsyn | there were in java in 90s | 11:25 |
sonne|work | lisitsyn: seems like there are huge differences http://preshing.com/20110603/hash-table-performance-tests | 11:25 |
@lisitsyn | sonne|work: between what and what? | 11:26 |
sonne|work | hash speed | 11:26 |
sonne|work | table | 11:26 |
@lisitsyn | sonne|work: what we shouldn't use is std::map :) | 11:26 |
@iglesiasg | indeed, unordered_map is the thing | 11:26 |
@iglesiasg | but I mean it doesn't make sense to compare std::map and a hash table, they are just not for the same things | 11:27 |
sonne|work | iglesiasg: what is the difference? | 11:28 |
@lisitsyn | sonne|work: O(ln N) vs O(1) | 11:28 |
@iglesiasg | sonne|work, std::maps are balanced binary search trees and unordered_map is a hash table | 11:28 |
@lisitsyn | red-black tree vs hash tree | 11:28 |
@lisitsyn | err | 11:29 |
@lisitsyn | hash table :D | 11:29 |
@lisitsyn | hash tree haha | 11:29 |
@iglesiasg | :) | 11:29 |
@lisitsyn | iglesiasg: how does hash table work in java? | 11:29 |
@lisitsyn | iglesiasg: interview question time | 11:29 |
@iglesiasg | lisitsyn, haha | 11:29 |
van51 | sonney2k: https://gist.github.com/van51/6339628 | 11:29 |
@lisitsyn | iglesiasg: what happens if the hash function is not that good? a != b but hash(a) == hash(b)? | 11:30 |
@iglesiasg | lisitsyn, linked list | 11:30 |
@lisitsyn | good | 11:30 |
@lisitsyn | ;) | 11:30 |
@lisitsyn | iglesiasg: how to break a hash map in java? | 11:30 |
@iglesiasg | not that good because I don't remember what that is called | 11:30 |
@iglesiasg | arrch | 11:30 |
@iglesiasg | anyway | 11:30 |
@iglesiasg | lisitsyn, break? | 11:30 |
@lisitsyn | iglesiasg: yeah a bit, make it incostistent may be | 11:31 |
@iglesiasg | lisitsyn, what I find weird of hash tables in java is that you don't need to define your hash function | 11:32 |
@lisitsyn | iglesiasg: you do need ;) | 11:33 |
sonne|work | van51: what is combinations again? | 11:33 |
@iglesiasg | lisitsyn, mmm, are you sure? | 11:33 |
@iglesiasg | IIRC there is something done by default | 11:33 |
@lisitsyn | iglesiasg: well they are usually auto-generated | 11:33 |
van51 | sonne|work: how many tokens/ngrams to combin | 11:33 |
@iglesiasg | lisitsyn, exactly, that is what I mean. IMHO, that is not good, at all | 11:33 |
sonne|work | van51: and n-gram length? | 11:34 |
@lisitsyn | iglesiasg: why? you write some POJO and it generates hash for you | 11:34 |
van51 | sonne|work: for these results, 3 | 11:34 |
sonne|work | van51: in any case ~90% is pretty okish already | 11:34 |
@lisitsyn | iglesiasg: I had a problem with entity graph though :D | 11:34 |
@lisitsyn | there were cycles so hash was like %infty time | 11:34 |
van51 | sonne|work: also some sentences were trash so on that specific dataset that I've created I don't think it can go higher | 11:34 |
@iglesiasg | lisitsyn, I have no idea how is the hash function autogenerated. But for things that are not primitive data types, I think that should not be automatically made | 11:35 |
sonne|work | van51: well OK it is already nice. how big is the data set? | 11:35 |
@iglesiasg | lisitsyn, you just told me an example why :) | 11:35 |
sonne|work | van51: I mean if we .xz compress it - could it be stored in shogun-data? | 11:35 |
@lisitsyn | iglesiasg: that's why in the west coast enterprise yo systems you always have some base class | 11:36 |
@lisitsyn | I just use UUID most of the time | 11:36 |
@iglesiasg | lisitsyn, that sounds reasonable | 11:36 |
van51 | sonne|work: it's ~246M | 11:36 |
van51 | sonne|work: w/o compression | 11:36 |
van51 | sonne|work: how much is acceptable? | 11:36 |
@lisitsyn | iglesiasg: it is tiresome to implement these hashes and equals's | 11:37 |
@iglesiasg | lisitsyn, I tried once for my search engines course hash tables that extend to secondary disk. Cool stuff | 11:37 |
@lisitsyn | iglesiasg: some B-tree I guess | 11:37 |
@lisitsyn | ahh | 11:37 |
@lisitsyn | hash table | 11:37 |
@iglesiasg | lisitsyn, no idea how was implemented internally. I should have been curious about that back then | 11:37 |
@lisitsyn | iglesiasg: people like B-trees when it comes to disk IO | 11:38 |
@iglesiasg | lisitsyn, databases are commonly b-trees, right? | 11:38 |
@lisitsyn | iglesiasg: they were (books say that) but I don't know what about now :) | 11:39 |
@iglesiasg | good point :) | 11:40 |
@lisitsyn | there are two guys in the same city with me that develop mysql | 11:40 |
@lisitsyn | no idea what makes them stay here but they kind of working remotely | 11:40 |
sonne|work | van51: well xz -9 compress them :) | 11:41 |
sonne|work | van51: it should be <100MB for sure | 11:41 |
van51 | sonne|work: compressing! I'll let you know | 11:42 |
@iglesiasg | lisitsyn, Eigen question | 11:44 |
@iglesiasg | lisitsyn, sqrt of a matrix. A.array().sqrt() or A.cwiseSqrt() but afterwards A remains unchanged, why? | 11:45 |
@lisitsyn | iglesiasg: you expect some eigenanswer | 11:45 |
@lisitsyn | hmm I think it shouldn't change anything | 11:45 |
@lisitsyn | iglesiasg: A = A.array.sqrt() | 11:46 |
@lisitsyn | you can try that | 11:46 |
@iglesiasg | lisitsyn, I think I tried already, but let me check | 11:47 |
@iglesiasg | lisitsyn, it doesn't compile | 11:47 |
@lisitsyn | A.array() | 11:47 |
@lisitsyn | iglesiasg: A.array() = A.array().sqrt()? | 11:48 |
@lisitsyn | what about that/ | 11:48 |
@iglesiasg | mmm that I didn't come up with | 11:48 |
@iglesiasg | let me see | 11:48 |
@iglesiasg | lisitsyn, yep, that made it | 11:49 |
@iglesiasg | lisitsyn, I got some problems with the data type though | 11:50 |
@iglesiasg | lisitsyn, it was not working before that I was using Matrix3i or Matrix3d, but I changed to MatrixXd and it worked | 11:50 |
@iglesiasg | let me show you what I mean | 11:50 |
@lisitsyn | iglesiasg: Matrix3d is kind of special case I guess | 11:50 |
sonne|work | van51: and? | 11:51 |
van51 | sonne|work: it's still going | 11:51 |
sonne|work | van51: how big is it already then? | 11:52 |
van51 | sonne|work: 52M | 11:52 |
@iglesiasg | lisitsyn, but doing A.array() = A.array().sqrt() is efficiently doing the stuff in place? | 11:52 |
@iglesiasg | lisitsyn, I got the feeling that just A.array().sqrt() (wihtout the operator=) seems more efficient | 11:52 |
van51 | sonne|work: it's done.. 57M | 11:55 |
sonne|work | van51: ok so that is good enough IMHO | 11:55 |
sonne|work | lisitsyn: heh http://www.sanmayce.com/Fastest_Hash/index.html#Yoshimura | 11:59 |
sonne|work | Yoshimura instead of murmur3 ;) | 11:59 |
@iglesiasg | omg what is that | 12:00 |
sonne|work | iglesiasg: looks like an assembly code optimized hash function :) | 12:02 |
@iglesiasg | people are too hardcore | 12:03 |
van51 | sonne|work: should I place it in a new dir? | 12:03 |
sonne|work | van51: yes sure it is a real application.. | 12:04 |
@wiking | fuckiiiing JNI | 12:07 |
van51 | sonne|work: got a warning from github that file exceeds 50M | 12:12 |
sonne|work | van51: well it is only a warning or? | 12:13 |
van51 | sonne|work: but from what I read it should be ok, unless it needs an update | 12:13 |
van51 | sonne|work: it pointed me here : https://help.github.com/articles/working-with-large-files | 12:13 |
sonne|work | van51: you could then put your application under applications/language_detection | 12:15 |
van51 | sonne|work: ok, cool! I'll clean the code a bit first and push later | 12:18 |
sonne|work | van51: another hash http://burtleburtle.net/bob/hash/spooky.html | 12:20 |
sonne|work | van51: or https://code.google.com/p/cityhash/ :) | 12:21 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 12:21 | |
van51 | sonne|work: cityhash looks nice idd | 12:26 |
@lisitsyn | sonne|work: that's totally crazy | 12:30 |
@lisitsyn | unordered_map should be ok for 99% purposes | 12:30 |
@lisitsyn | I mean its default hashing function | 12:30 |
@lisitsyn | however you are doing some fancy things may be it would be ok here | 12:32 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 12:40 | |
@iglesiasg | lisitsyn, sonne|work, HeikoS: about merging the ipython notebooks. How would you like to do it? One pull request at the end once they are finished or several as we go adding more stuff | 12:45 |
@iglesiasg | since I think this cannot really be revised as normal code I guess that maybe one pull request at the end with it is better | 12:45 |
@lisitsyn | iglesiasg: I have no opinion on that :) | 12:55 |
-!- van51 [~van51@ppp-94-66-12-14.home.otenet.gr] has quit [Quit: Leaving.] | 13:01 | |
-!- HeikoS1 [~heiko@nat-188-24.internal.eduroam.ucl.ac.uk] has joined #shogun | 13:02 | |
@lisitsyn | HeikoS1: ! | 13:05 |
@lisitsyn | HeikoS1: HeikoS: need comments from your side on the 'keywords' idea | 13:06 |
HeikoS1 | lisitsyn: hi! | 13:12 |
HeikoS1 | its bank holiday here, excuse my late appearance :) | 13:13 |
-!- foulwall [~user@2001:da8:215:6901:11d1:4711:4290:8f68] has joined #shogun | 13:14 | |
foulwall | /usr/local/include/c++/4.8.1/atomic:1115:57: error: expected initializer before 'noexcept' | 13:17 |
foulwall | atomic_fetch_sub(atomic<_ITp*>* __a, ptrdiff_t __d) noexcept | 13:17 |
foulwall | ^ | 13:17 |
foulwall | In file included from /usr/local/include/shogun/base/SGObject.h:20:0, | 13:17 |
foulwall | from /usr/local/include/shogun/io/File.h:16, | 13:17 |
foulwall | from /usr/local/include/shogun/io/HDF5File.h:18, | 13:17 |
foulwall | from library_hdf5.cpp:5: | 13:17 |
foulwall | /usr/local/include/shogun/lib/RefCount.h: In constructor 'shogun::RefCount::RefCount(int32_t)': | 13:17 |
foulwall | /usr/local/include/shogun/lib/RefCount.h:23:46: error: no matching function for call to 'std::atomic<int>::atomic(int32_t&)' | 13:17 |
foulwall | RefCount(int32_t ref_start=0) : rc(ref_start) {} | 13:17 |
foulwall | ^ | 13:17 |
foulwall | /usr/local/include/shogun/lib/RefCount.h:23:46: note: candidate is: | 13:17 |
HeikoS1 | foulwall: thats c++ issue | 13:17 |
foulwall | In file included from /usr/local/include/shogun/lib/RefCount.h:2:0, | 13:17 |
foulwall | from /usr/local/include/shogun/base/SGObject.h:20, | 13:17 |
HeikoS1 | you have to compile with the latest c++ | 13:17 |
foulwall | from /usr/local/include/shogun/io/File.h:16, | 13:17 |
foulwall | from /usr/local/include/shogun/io/HDF5File.h:18, | 13:18 |
foulwall | from library_hdf5.cpp:5: | 13:18 |
foulwall | /usr/local/include/c++/4.8.1/atomic:599:7: note: std::atomic<int>::atomic() | 13:18 |
foulwall | atomic() noexcept = default; | 13:18 |
foulwall | ^ | 13:18 |
HeikoS1 | -std=c++11 | 13:18 |
foulwall | /usr/local/include/c++/4.8.1/atomic:599:7: note: candidate expects 0 arguments, 1 provided | 13:18 |
foulwall | make: *** [library_hdf5] Error 1 | 13:18 |
foulwall | I compile it with gcc 4.8.1 | 13:18 |
foulwall | ok thanks HeikoS1 | 13:18 |
HeikoS1 | foulwall: add this flag | 13:18 |
HeikoS1 | foulwall: if this is your own program, that should do it | 13:18 |
HeikoS1 | if this is from cmake, no idea whats causing it | 13:18 |
foulwall | from examples/undocumented/libshogun/library_hdf5.cpp | 13:19 |
HeikoS1 | foulwall: yes then compile with the flag | 13:20 |
HeikoS1 | lisitsyn: cool stuff | 13:20 |
foulwall | thanks HeikoS:) | 13:20 |
HeikoS1 | lisitsyn: but the global keywords are unreaslistic | 13:20 |
HeikoS1 | could that be done with namespaces? | 13:20 |
HeikoS1 | lisitsyn: what do you think on | 13:21 |
HeikoS1 | https://github.com/shogun-toolbox/shogun/issues/1460#issuecomment-23235793 | 13:21 |
votjakovr | HeikoS1: hi! The difference between our EP in combination with probit likelihood and EP from GPML is no more than 1e-3. Should i send a PR? | 13:25 |
HeikoS1 | votjakovr: hi ah finally we meet | 13:25 |
@lisitsyn | HeikoS1: lets talk about that | 13:25 |
HeikoS1 | votjakovr: that should be fine, how is it against GPstuff? | 13:25 |
@lisitsyn | HeikoS1: I need to clarify - what's unrealistic in global keywords? | 13:25 |
HeikoS1 | lisitsyn: imagine the parameter keywork "k" | 13:26 |
HeikoS1 | that will be used everywhere | 13:26 |
votjakovr | HeikoS1: i think that this is because of random | 13:26 |
@lisitsyn | HeikoS1: you mean name clash with variables? | 13:26 |
HeikoS1 | votjakovr: could you elaborate a bit? | 13:26 |
HeikoS1 | lisitsyn: yes | 13:26 |
@lisitsyn | HeikoS1: we can put them to inner scope | 13:26 |
@lisitsyn | KNN.k | 13:26 |
@lisitsyn | we can put them to any namespace | 13:27 |
HeikoS1 | lisitsyn: is the KNN generated automatically? | 13:27 |
HeikoS1 | or do people have to add that? | 13:27 |
votjakovr | HeikoS1: since difference between negative log marginal likelihoods is something about 1e-10 | 13:27 |
@lisitsyn | no you'd have to add that | 13:27 |
HeikoS1 | lisitsyn: how do you see that? | 13:27 |
HeikoS1 | is there a way to check that? | 13:27 |
@lisitsyn | HeikoS1: to check what? | 13:27 |
HeikoS1 | lisitsyn: so in the optimal world, devs would just have to register the parameter, and then everything else is generated automatically | 13:28 |
HeikoS1 | votjakovr: ok, but you are also integrating differently right? | 13:28 |
@lisitsyn | HeikoS1: this way you'd need to add a keyword if it is not here | 13:29 |
HeikoS1 | votjakovr: maybe test another dataset to make sure that the difference is low everywhere | 13:29 |
votjakovr | HeikoS1: the order of the local likelihood approximations is random | 13:30 |
HeikoS1 | votjakovr: I see | 13:30 |
@lisitsyn | HeikoS1: I don't get the uniqueness point | 13:30 |
HeikoS1 | lisitsyn: maybe I get it wrong | 13:30 |
HeikoS1 | lisitsyn: could you tell me what I would have to do on writing a new class with parameter "k" ? | 13:30 |
votjakovr | HeikoS1: and GPML has a difference between calls on the same dataset | 13:30 |
HeikoS1 | votjakovr: ah ok, well then this is fine | 13:31 |
@lisitsyn | HeikoS1: define `static const Keyword k;` | 13:31 |
@lisitsyn | and mark somehow that you will use it | 13:31 |
HeikoS1 | lisitsyn: what if another class also has a parameter k? | 13:31 |
votjakovr | HeikoS1: ok, i'll send a PR | 13:31 |
HeikoS1 | votjakovr: nice! | 13:31 |
@lisitsyn | HeikoS1: totally ok | 13:31 |
@lisitsyn | HeikoS1: the keyword doesn't store any value | 13:31 |
@lisitsyn | it is just used to dispatch the type | 13:32 |
HeikoS1 | lisitsyn: ok | 13:32 |
HeikoS1 | lisitsyn: float k, int k, both ok? | 13:32 |
@lisitsyn | ahh | 13:32 |
@lisitsyn | different types? | 13:32 |
HeikoS1 | yes | 13:32 |
@lisitsyn | yes that's a problem | 13:32 |
HeikoS1 | lisitsyn: also keep in mind what our goal was: That one registers a parameter once, and then getters/setters are automatically generated | 13:33 |
@lisitsyn | well that's just name clash | 13:33 |
@lisitsyn | HeikoS1: yes that fits the goal | 13:33 |
HeikoS1 | so it would be best if the SG_ADD, or a similar line would do that | 13:33 |
@lisitsyn | HeikoS1: use(keyword.k); | 13:33 |
@lisitsyn | HeikoS1: that's how I see it | 13:33 |
sonne|work | iglesiasg: I prefer incremental contributions wrt notebooks | 13:33 |
@iglesiasg | sonne|work, all right | 13:34 |
@lisitsyn | HeikoS1: static keyword instance is just a replacement for string | 13:34 |
@lisitsyn | but it can be used to infer the type | 13:34 |
@lisitsyn | that's the only difference | 13:34 |
sonne|work | lisitsyn: sure this is more about what van51 is doing :) | 13:34 |
@lisitsyn | sonne|work: yeah then makes sense with such a massive hashing | 13:34 |
HeikoS1 | I am confused ;) | 13:35 |
@lisitsyn | HeikoS1: but do you think it is a common case with float k and int k? | 13:35 |
@lisitsyn | I mean it is much better to call it n_neighbors | 13:35 |
HeikoS1 | lisitsyn: I just think we should not sacrifice flexibility | 13:35 |
HeikoS1 | lisitsyn: because currently we can do that | 13:35 |
HeikoS1 | lisitsyn: but say this is not a problem | 13:35 |
@lisitsyn | HeikoS1: well it is just a suggestion :) | 13:36 |
sonne|work | lisitsyn: do you have gcc 4.8? | 13:36 |
@lisitsyn | sonne|work: not really | 13:36 |
sonne|work | foulwall: hmmhh is this the latest clone on what arch? | 13:37 |
HeikoS1 | lisitsyn: I like it, just not sure whether I understand it | 13:37 |
@lisitsyn | HeikoS1: what is unclear? let me explain :) | 13:37 |
@lisitsyn | I am afraid I am losing some point so I'd like to discuss it | 13:37 |
HeikoS1 | lisitsyn: yes cool | 13:37 |
HeikoS1 | lisitsyn: I guess an example would be cool | 13:38 |
HeikoS1 | where a parameter is registered | 13:38 |
HeikoS1 | and then how to do get/set | 13:38 |
@lisitsyn | HeikoS1:well it is not really about parameter registration | 13:38 |
foulwall | sonne|work: latest clone and on debian, in examples/undocument/libshogun, I add -std=C++11 in libshogun's Makefile and now it works fine | 13:39 |
HeikoS1 | lisitsyn: I know, but thats what I dont get yet | 13:39 |
@lisitsyn | HeikoS1: in sgobject we'd have a map<Keyword,any> | 13:39 |
@lisitsyn | HeikoS1: so once you register parameter you add something to the map | 13:39 |
sonne|work | wiking: are we compiling examples with same flags like c++11 etc like libshogun? | 13:40 |
@lisitsyn | say you call float get(Keyword<float>) | 13:40 |
@lisitsyn | it finds the keyword, casts the any value to float | 13:40 |
@lisitsyn | and returns it | 13:40 |
foulwall | sonne|work: there's no c++11 flags in libshogun's makefile | 13:40 |
sonne|work | foulwall: yeah but you build with cmake now right? | 13:40 |
HeikoS1 | lisitsyn: ok | 13:40 |
votjakovr | HeikoS1: i've just sent it :) | 13:41 |
HeikoS1 | lisitsyn: should we start a feature branch with that? | 13:41 |
@lisitsyn | HeikoS1: I don't know | 13:41 |
HeikoS1 | votjakovr: reading :) | 13:41 |
@lisitsyn | HeikoS1: I can try | 13:41 |
foulwall | sonne|work: Ah, for examples I built with make. I'll try cmake | 13:41 |
@lisitsyn | HeikoS1: it doesn't sound like a big amount of work actually | 13:41 |
HeikoS1 | lisitsyn: that would be even better :) | 13:41 |
@lisitsyn | HeikoS1: the issue that bothers me | 13:42 |
@lisitsyn | and you too probably | 13:42 |
@lisitsyn | is there to place there keywords or whatever | 13:42 |
@lisitsyn | these keywords* | 13:42 |
@lisitsyn | where* | 13:42 |
@lisitsyn | :D | 13:42 |
@lisitsyn | shit | 13:42 |
@lisitsyn | HeikoS1: so the class scope or namespace | 13:42 |
@lisitsyn | or globally | 13:42 |
@lisitsyn | I don't know | 13:43 |
@lisitsyn | class scope is bad because you write the class name too many times | 13:43 |
@lisitsyn | namespace scope is ok but a bit messy | 13:43 |
HeikoS1 | lisitsyn: ehm what? | 13:44 |
@lisitsyn | global scope can lead to name clashes as people do `from modshogun import *` sometimes | 13:44 |
@lisitsyn | HeikoS1: class scope is KNN.k | 13:44 |
HeikoS1 | lisitsyn: I see | 13:44 |
@lisitsyn | by namespace scope I mean kw.k | 13:44 |
@lisitsyn | so modshogun.kw.k | 13:44 |
@lisitsyn | like that | 13:44 |
HeikoS1 | no class scope should be better | 13:44 |
HeikoS1 | dont you think? | 13:44 |
@lisitsyn | well depends | 13:44 |
@lisitsyn | HeikoS1: actually quite a few parameters are shared | 13:45 |
@lisitsyn | n_neighbors, n_iters | 13:45 |
@lisitsyn | width, etc | 13:45 |
HeikoS1 | yep | 13:46 |
@lisitsyn | HeikoS1: I expect it speeds up compilation | 13:46 |
HeikoS1 | lisitsyn: thats good | 13:46 |
HeikoS1 | lisitsyn: a question: will this allow to manually override getters/setters? | 13:47 |
@lisitsyn | HeikoS1: but I don't know how much | 13:47 |
HeikoS1 | lisitsyn: since the idea also was to use this for safer model-selection calls | 13:47 |
@lisitsyn | HeikoS1: there should be a way | 13:47 |
HeikoS1 | you know that for example one can make sure stter only accepts pos. values of an integer | 13:47 |
@lisitsyn | HeikoS1: I think we need some 'checkers' for that | 13:48 |
@lisitsyn | so you attach Positive checker to parameter | 13:48 |
@lisitsyn | it is a rare case you really need something custom | 13:48 |
-!- foulwall [~user@2001:da8:215:6901:11d1:4711:4290:8f68] has quit [Ping timeout: 264 seconds] | 13:48 | |
@lisitsyn | HeikoS1: static const Keyword<float> width("Width of gaussian kernel", positive + non_zero); | 13:49 |
HeikoS1 | lisitsyn: no wait | 13:49 |
HeikoS1 | the point why I suggested that originally was | 13:49 |
HeikoS1 | that in model-selection, I can only modify the parameters directly | 13:50 |
HeikoS1 | but sometimes, one has some post methods that need to be called or statements executed | 13:50 |
@lisitsyn | HeikoS1: this is obvious observer pattern for me I think it can be done | 13:50 |
HeikoS1 | lisitsyn: lets thing of an example | 13:51 |
HeikoS1 | ah yes | 13:51 |
HeikoS1 | Gaussian kernel | 13:51 |
HeikoS1 | the diagonal is precomputed | 13:51 |
@lisitsyn | we just add some NOP method that is called once parameter is changed | 13:51 |
HeikoS1 | lisitsyn: actually, the setter does not even update it | 13:52 |
HeikoS1 | argh | 13:52 |
HeikoS1 | lisitsyn: ok | 13:52 |
HeikoS1 | that would do it | 13:52 |
@lisitsyn | HeikoS1: yeah got it | 13:52 |
HeikoS1 | so currnetly GaussianKernel::init(lhs,rhs);GaussianKernel.set_width(other_value); GaussianKernel::get_kernel_matrix | 13:53 |
HeikoS1 | returns wrong results | 13:53 |
@lisitsyn | why? | 13:53 |
HeikoS1 | ah no | 13:53 |
HeikoS1 | precompute squared just computes dot products | 13:53 |
HeikoS1 | sorry | 13:53 |
@lisitsyn | HeikoS1: I think there is nothing infeasible here but good you mention it | 13:54 |
HeikoS1 | votjakovr: I will wait for travis and then merge, good work! | 13:54 |
HeikoS1 | lisitsyn: oh one more thing | 13:56 |
@lisitsyn | HeikoS1: yes? | 13:56 |
HeikoS1 | lisitsyn: is it possible for the model-selection framework to use that? | 13:56 |
votjakovr | HeikoS1: good :) | 13:56 |
HeikoS1 | lisitsyn: since in there, one has only access to the registered parameters | 13:56 |
@lisitsyn | HeikoS1: too generic question ;) | 13:56 |
HeikoS1 | as shogun parameters | 13:56 |
HeikoS1 | so TParameter | 13:56 |
@lisitsyn | HeikoS1: it doesn't allow *unregistered* parameters | 13:56 |
@lisitsyn | you just can't use it if it is not in map | 13:57 |
HeikoS1 | lisitsyn: no I mean | 13:57 |
HeikoS1 | from the source code, I need to call things | 13:57 |
HeikoS1 | let me show you how I do it currently | 13:57 |
HeikoS1 | ParameterCombination::apply_to_modsel_parameter | 13:58 |
HeikoS1 | lisitsyn: and then this uses Parameter::set_from_parameters | 13:59 |
HeikoS1 | in there I even have to do a type distinction | 13:59 |
@lisitsyn | HeikoS1: I don't really get it | 14:00 |
HeikoS1 | the grid search iterates over all possible parameter combinations | 14:00 |
HeikoS1 | and then applies them to SGObjects | 14:00 |
@lisitsyn | yes | 14:00 |
HeikoS1 | this is currently done via just copying the memory | 14:00 |
HeikoS1 | or pointer etc | 14:00 |
HeikoS1 | will it be possible to do this with the setters/getters? | 14:01 |
@lisitsyn | HeikoS1: yes probably if I get what you mean | 14:01 |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 14:02 | |
shogun-notifier- | shogun: Roman Votyakov :develop * fa3a7c9 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/fa3a7c92bbd0c0e78b3b0c9d8d9f16fb69869295 | 14:02 |
shogun-notifier- | shogun: add expectation propagation inference method | 14:02 |
shogun-notifier- | shogun: Heiko Strathmann :develop * 8aee21e / / (3 files): https://github.com/shogun-toolbox/shogun/commit/8aee21eed0b16d18b348976da5bb51e11fe63325 | 14:02 |
shogun-notifier- | shogun: Merge pull request #1462 from votjakovr/feature/gp_binary_classification | 14:02 |
shogun-notifier- | shogun: | 14:02 |
shogun-notifier- | shogun: Add expectation propagation inference method | 14:02 |
HeikoS1 | lisitsyn: though, as I said, I would totally drop grid-search | 14:02 |
HeikoS1 | lisitsyn: the effort of using our framework is larger than doing it by hand if cross-validation is available | 14:02 |
@lisitsyn | HeikoS1: yeah that's why I try to think how to do that better way | 14:02 |
HeikoS1 | lisitsyn: still | 14:03 |
HeikoS1 | the amount of work to maintain that is just insane | 14:03 |
HeikoS1 | and it sooo easy to do it by hand | 14:03 |
@lisitsyn | HeikoS1: I believe it is too low-level | 14:03 |
@lisitsyn | pointers etc | 14:03 |
@lisitsyn | indexes | 14:03 |
@lisitsyn | enums | 14:04 |
@lisitsyn | but I am not sure | 14:04 |
HeikoS1 | lisitsyn: exactly | 14:04 |
@lisitsyn | if it could be done better | 14:04 |
HeikoS1 | the whole parameter tree | 14:04 |
HeikoS1 | paramter combination | 14:04 |
HeikoS1 | etc | 14:04 |
HeikoS1 | lisitsyn: I dont really see another way to do it | 14:05 |
HeikoS1 | lisitsyn: the problem is the way how shogun parameters are registered | 14:05 |
@lisitsyn | yeah you would do it better if you knew :) | 14:05 |
HeikoS1 | and creating parameter data from "nowhere" is just hard | 14:05 |
HeikoS1 | lisitsyn: also, no debugging of model-selection | 14:05 |
HeikoS1 | remember this x-validation output class? | 14:06 |
@lisitsyn | yes | 14:06 |
HeikoS1 | we are lacking that for grid-search | 14:06 |
HeikoS1 | its easy to do, but again, its effort | 14:06 |
HeikoS1 | and if people do this by hand, its not | 14:06 |
HeikoS1 | but later | 14:07 |
HeikoS1 | ill fix the xml serialization for now :) | 14:07 |
@lisitsyn | HeikoS1: alright let me think about it in background daemon | 14:08 |
@lisitsyn | as usual | 14:08 |
HeikoS1 | ok :) | 14:08 |
HeikoS1 | lisitsyn: cool ideas you had though, I would love to see this happen | 14:08 |
@lisitsyn | you may notice I disappeared for a month and then came with some idea | 14:08 |
@lisitsyn | :D | 14:08 |
HeikoS1 | haha :) | 14:08 |
HeikoS1 | lisitsyn: btw do you have any idea why we use cstrings rather than string? | 14:08 |
@lisitsyn | HeikoS1: cstrings - const char*? | 14:09 |
HeikoS1 | yes | 14:09 |
@lisitsyn | no idea | 14:09 |
HeikoS1 | lisitsyn: just so dirty | 14:09 |
@lisitsyn | we are in std haters club | 14:09 |
HeikoS1 | haha :) | 14:09 |
thoralf | :D | 14:09 |
@lisitsyn | the first rule | 14:09 |
@lisitsyn | is never use std:: | 14:09 |
thoralf | shogun is implemented in C with classes :) | 14:10 |
HeikoS1 | thoralf: indeed :) | 14:10 |
thoralf | lisitsyn: Rule 1: Whenever you can reinvent the wheel: Do it. | 14:10 |
@lisitsyn | HeikoS1: tech-wise we also need shared_ptr | 14:10 |
@lisitsyn | thoralf: yes true | 14:11 |
@lisitsyn | the wheel is not round enough | 14:11 |
thoralf | lisitsyn: SGVector (or better: void *) instead of std::vector | 14:11 |
@lisitsyn | thoralf: sgvector is actually something else than std::vector now | 14:12 |
thoralf | lisitsyn: I know. | 14:12 |
@lisitsyn | I hate everything about it though | 14:12 |
@lisitsyn | all these cmaps and etc | 14:12 |
thoralf | lisitsyn: Elaborate? | 14:12 |
@lisitsyn | dynamicobjectarray | 14:12 |
@lisitsyn | or how it is called | 14:12 |
@lisitsyn | I wish we didn't had that code | 14:12 |
HeikoS1 | lisitsyn: go for it! kill it | 14:13 |
HeikoS1 | remove things step by step :) | 14:13 |
HeikoS1 | lisitsyn: we are currently on a good run to make things better, so this is another one on the list | 14:13 |
HeikoS1 | lisitsyn: and you are the expert ;) | 14:13 |
thoralf | lisitsyn: I'd be happy to use a thin std::vector wrapper instead of sgvector. :) | 14:13 |
@lisitsyn | thoralf: why? | 14:14 |
thoralf | lisitsyn: Just to simplify code. | 14:14 |
shogun-buildbot | build #1527 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1527 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 14:14 |
@lisitsyn | thoralf: I'd like to see Eigen::VectorXd here | 14:14 |
thoralf | lisitsyn: Even better. | 14:15 |
thoralf | Stop. No reinventing? No! | 14:15 |
@lisitsyn | thoralf: opencv uses just Mat | 14:15 |
@lisitsyn | stupid naming but engineering wise it looks better | 14:16 |
HeikoS1 | lisitsyn: btw we should also have a wrapper class that makes a CSGObject out of SG Types | 14:17 |
HeikoS1 | to be able to use them in lists etc | 14:17 |
@lisitsyn | HeikoS1: what lists? | 14:17 |
HeikoS1 | like a list of vectors | 14:17 |
HeikoS1 | different sized vectors | 14:17 |
HeikoS1 | sorry | 14:18 |
HeikoS1 | not a list | 14:18 |
HeikoS1 | an array that dynamically grows | 14:18 |
@lisitsyn | std::vector? | 14:18 |
@lisitsyn | ;) | 14:18 |
HeikoS1 | lisitsyn: yes | 14:18 |
HeikoS1 | but serializable | 14:18 |
HeikoS1 | lisitsyn: a shogun type | 14:18 |
@lisitsyn | ahh I have no idea about it | 14:18 |
HeikoS1 | we could even define a wrapper class for std::vector | 14:18 |
thoralf | lisitsyn: Anyway, I think that SGSparse* is a good thing. :) | 14:19 |
HeikoS1 | which IS a CSGObject | 14:19 |
votjakovr | HeikoS1: wow EP works much faster than Laplace! | 14:25 |
HeikoS1 | votjakovr: really? | 14:25 |
HeikoS1 | votjakovr: thats weird | 14:25 |
HeikoS1 | usually its more expensive | 14:25 |
HeikoS1 | I wonder why that is | 14:25 |
votjakovr | HeikoS1: yep i know, but i've tried graphical example with EP and it is subjectively faster | 14:26 |
HeikoS1 | votjakovr: ok, well thats great then | 14:27 |
HeikoS1 | votjakovr: we can always tune later | 14:27 |
HeikoS1 | votjakovr: did you see my pm? | 14:27 |
votjakovr | HeikoS1: i think in future we could make a benchmark test and try to optimize bottlenecks | 14:28 |
votjakovr | HeikoS1: pm? | 14:29 |
HeikoS1 | votjakovr: yep that would be awesome! also to have benchmarks which other people can use | 14:29 |
HeikoS1 | votjakovr: my private message in irc | 14:29 |
HeikoS1 | votjakovr: btw did you already start on the notebook? | 14:33 |
HeikoS1 | I liked your graphical examples a lot | 14:33 |
HeikoS1 | they should be in there definitely | 14:33 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Read error: Connection reset by peer] | 14:40 | |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has joined #shogun | 14:40 | |
travis-ci | [travis-ci] it's Heiko Strathmann'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/10621112 | 14:40 |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has left #shogun [] | 14:40 | |
@iglesiasg | what's up with travis? | 14:47 |
@iglesiasg | it seems that the ppa for octave change killed it | 14:47 |
HeikoS1 | yep | 14:48 |
shogun-buildbot | build #1610 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/1610 | 14:48 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 14:50 | |
shogun-buildbot | build #992 of rpm1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/992 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 14:51 |
shogun-buildbot | build #1611 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1611 blamelist: Roman Votyakov <votjakovr@gmail.com> | 14:53 |
@iglesiasg | HeikoS1: is someone trying to fix that? | 14:53 |
HeikoS1 | iglesiasg: not aware of :) | 14:53 |
HeikoS1 | I am fixing other things currently | 14:53 |
shogun-buildbot | build #1316 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1316 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 14:54 |
shogun-buildbot | build #1528 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1528 blamelist: Roman Votyakov <votjakovr@gmail.com> | 14:57 |
@iglesiasg | sonne|work, why is the second ppa for octave 3.6 added? | 14:58 |
sonne|work | iglesiasg: remove it - it didn't work and I cannot from work | 14:58 |
@iglesiasg | sonne|work, ok | 14:59 |
@iglesiasg | sonne|work, it starts with sudo sudo add-apt instead of sudo add-apt | 14:59 |
@iglesiasg | maybe that is the problem | 14:59 |
sonne|work | iglesiasg: don't think so but still a bug | 15:00 |
@iglesiasg | yeah.. the trace looks unrelated to that, but I will try without sudo twice first just in case | 15:01 |
shogun-notifier- | shogun: Fernando Iglesias :develop * ebf640f / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/ebf640f341ee65b4996d8640f35570f5b3e33196 | 15:03 |
shogun-notifier- | shogun: Remove double sudo in travis octave 3.6 second ppa. | 15:03 |
@iglesiasg | nah, it didn't fix it | 15:05 |
HeikoS1 | iglesiasg: please dont use "m_" as parameter names | 15:07 |
HeikoS1 | I will change that in you LMNN class | 15:07 |
shogun-notifier- | shogun: Fernando Iglesias :develop * 765e145 / .travis.yml: https://github.com/shogun-toolbox/shogun/commit/765e145c23292f9cd8f8b678aaecffb330e71cf1 | 15:07 |
shogun-notifier- | shogun: Remove octave 3.6 second ppa in travis. | 15:07 |
@iglesiasg | HeikoS1, eeeh all right, but why not? | 15:07 |
@iglesiasg | I think it is a good practice to diff the members somehow | 15:08 |
HeikoS1 | thats more of a style thing for the member variables | 15:08 |
HeikoS1 | but the variable name should be the parameter name | 15:08 |
@iglesiasg | what do you mean? | 15:08 |
@iglesiasg | I am just using it for the members, right? | 15:08 |
HeikoS1 | yep | 15:09 |
HeikoS1 | so CClass:m_k is registered as "k" | 15:09 |
HeikoS1 | not a big thing, I just saw it | 15:09 |
@iglesiasg | aaaah ok | 15:09 |
@iglesiasg | just when doing SG_ADD | 15:09 |
HeikoS1 | yes :) | 15:09 |
@iglesiasg | not changing the name of attributes | 15:09 |
@iglesiasg | sure :) | 15:09 |
@iglesiasg | HeikoS1: out of curiosity, why? I mean, for what are used later these names? | 15:10 |
@iglesiasg | I guess that for the SGObject clone is a reason | 15:10 |
HeikoS1 | iglesiasg: just whenever the parameter is accessed written or read | 15:11 |
HeikoS1 | also for modelselection | 15:12 |
HeikoS1 | so simpler is better there | 15:12 |
HeikoS1 | oh man | 15:12 |
HeikoS1 | iglesiasg: I justed looked into the PCA class | 15:13 |
HeikoS1 | thats pure horror | 15:13 |
@iglesiasg | yeah, a little bit :) | 15:13 |
@iglesiasg | it took me a little while to discover how to get the transform matrix when I wanted to use it for PCA | 15:13 |
@iglesiasg | there were some arguments to tune and so | 15:14 |
HeikoS1 | PCA is a three liner why is that so complicated ...grrrr | 15:14 |
@iglesiasg | hehe yeah | 15:14 |
HeikoS1 | but no eigen was around back then | 15:15 |
@iglesiasg | I mean, it is good that it can be tunable, the PCA we have does extra stuff | 15:15 |
HeikoS1 | true | 15:15 |
@iglesiasg | but I think it could be way more readable using Eigen and still keeping those functionalities | 15:15 |
HeikoS1 | iglesiasg: feel free to change :) also a unit test might be nice for such a basic thing | 15:16 |
@iglesiasg | but at the end of the day, it works just fine so I am not sure to the point it pays off to do it | 15:16 |
@iglesiasg | HeikoS1, unit test is a good idea indeed | 15:16 |
@iglesiasg | HeikoS1: BTW, we should set up a shogun-team meeting for the NIPS workshop | 15:17 |
HeikoS1 | iglesiasg: yep totally | 15:17 |
HeikoS1 | wiking is the guy to do :) | 15:17 |
shogun-buildbot | build #993 of rpm1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/993 | 15:17 |
shogun-buildbot | build #1529 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1529 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 15:18 |
@iglesiasg | I don't mind setting up the doodle, I will do it this evening probably | 15:18 |
-!- pickle27 [~kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun | 15:18 | |
lisitsyn | HeikoS1: I thought you wanted to go? ;) | 15:19 |
lisitsyn | it can be any of us though, I know I am off | 15:20 |
-!- HeikoS1 [~heiko@nat-188-24.internal.eduroam.ucl.ac.uk] has quit [Ping timeout: 245 seconds] | 15:22 | |
@wiking | HeikoS: ? | 15:28 |
-!- HeikoS1 [~heiko@nat-188-24.internal.eduroam.ucl.ac.uk] has joined #shogun | 15:28 | |
@wiking | ah yeah the meeting | 15:28 |
@iglesiasg | yeah | 15:28 |
@wiking | i'll send around a doodle today | 15:28 |
@iglesiasg | cool! | 15:28 |
* wiking is a bit busy today with errands... no commits today :P | 15:29 | |
@iglesiasg | wiking, I am excited about this NIPS thing, it would be cool if we can do something for it | 15:29 |
@wiking | yeah indeed | 15:29 |
@iglesiasg | wiking, we have been mistreating travis :-O | 15:29 |
@wiking | iglesiasg: heheh i saw... | 15:29 |
@wiking | fixit :) | 15:29 |
@iglesiasg | it seems to be fine again though | 15:30 |
@iglesiasg | oh no fuck | 15:30 |
HeikoS1 | tada | 15:30 |
HeikoS1 | XML serialization is lossy | 15:30 |
@wiking | iglesiasg: it's fine... i've restarted the job... it was just some randome | 15:30 |
@wiking | error ;P | 15:30 |
@iglesiasg | wiking, yes, MulticlassOCASTest.train, it is not the first time it fails I think mm | 15:31 |
@wiking | yeah | 15:32 |
@wiking | that unittest should be a bit modified + random seed should be fixed imho | 15:33 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * 2741b0c / src/shogun/ (7 files): https://github.com/shogun-toolbox/shogun/commit/2741b0c8c10e49012fa51a6485387f2f506ba51a | 15:37 |
shogun-notifier- | shogun: removed spaces from parameter names that are registered | 15:37 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * e11517f / src/shogun/base/Parameter.cpp: https://github.com/shogun-toolbox/shogun/commit/e11517f0012428b96614094cd3a56ddcf20342af | 15:37 |
shogun-notifier- | shogun: only allow alnum and underscore for registered parameter names | 15:37 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * 9759434 / tests/unit/io/SerializationXML_unittest.cc.jinja2: https://github.com/shogun-toolbox/shogun/commit/97594343003d94a63293737877a0135a2dbeee01 | 15:37 |
shogun-notifier- | shogun: introduce accuracy for xml serialization equal tests | 15:37 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * a022433 / tests/unit/CMakeLists.txt: https://github.com/shogun-toolbox/shogun/commit/a0224338eabf701264a66dfd9b0551de00b8871a | 15:37 |
shogun-notifier- | shogun: activate XML serialization since it works | 15:37 |
HeikoS1 | wiking: XML works! | 15:37 |
HeikoS1 | wiking: json last one | 15:38 |
shogun-buildbot | build #1317 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1317 blamelist: Roman Votyakov <votjakovr@gmail.com> | 15:38 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has joined #shogun | 15:42 | |
travis-ci | [travis-ci] it's Fernando Iglesias'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/10622771 | 15:42 |
-!- travis-ci [~travis-ci@ec2-23-23-19-157.compute-1.amazonaws.com] has left #shogun [] | 15:42 | |
shogun-notifier- | shogun: Roman Votyakov :develop * 3171861 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/31718619c67e1c5543305c1add80b613fbdf68cf | 15:44 |
shogun-notifier- | shogun: add expectation propagation to modular interfaces | 15:44 |
shogun-notifier- | shogun: Heiko Strathmann :develop * a447d63 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/a447d636d7cc5bd1233076d90cb6d8ebbc91211b | 15:44 |
shogun-notifier- | shogun: Merge pull request #1463 from votjakovr/feature/gp_binary_classification | 15:44 |
shogun-notifier- | shogun: | 15:44 |
shogun-notifier- | shogun: Add expectation propagation to modular interfaces | 15:44 |
-!- pickle27 [~kevin@d67-193-243-174.home3.cgocable.net] has left #shogun [] | 15:47 | |
-!- thoralf [~thoralf@enki.zib.de] has quit [Read error: Connection reset by peer] | 15:47 | |
-!- thoralf [~thoralf@enki.zib.de] has joined #shogun | 15:49 | |
thoralf | Yo. | 15:49 |
HeikoS1 | yoyo | 15:50 |
thoralf | :) | 15:50 |
thoralf | yolo | 15:50 |
HeikoS1 | krkr ;) | 15:51 |
thoralf | HeikoS1: I don't speak russian. ;) | 15:52 |
HeikoS1 | haha :D | 15:52 |
shogun-buildbot | build #1530 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1530 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 15:57 |
votjakovr | HeikoS1: what system is used on travis-ci? I'd like to enable nlopt package on travis-ci, since gradient model selection depends on it. | 15:58 |
HeikoS1 | votjakovr: best ask wiking for that, I think you can just sudo-apt get install the usual stuff | 15:58 |
HeikoS1 | have a look how we do it for eigen etc | 15:59 |
@wiking | votjakovr: is nlopt available as ubuntu package? | 15:59 |
@wiking | votjakovr: i'm afraid that it's only available in raring http://packages.ubuntu.com/raring/libnlopt-dev | 16:01 |
votjakovr | wiking: yep, since 12.10 | 16:01 |
@wiking | votjakovr: yeah and travis-ci is not 12.10 | 16:01 |
@wiking | it's like 12.04 | 16:01 |
@wiking | so you have to write a bundle script for it | 16:01 |
@wiking | see cmake/external/ | 16:01 |
votjakovr | wiking: ok, thanks, i'll do it like there :) | 16:02 |
@wiking | and try to write a patch accordingly | 16:02 |
@wiking | bundle it as a static lib | 16:02 |
@wiking | so just write it send a PR tag me and i'll review it | 16:03 |
lisitsyn | thoralf: russian sounds a bit different ;) | 16:03 |
votjakovr | wiking: ok, thanks | 16:03 |
thoralf | lisitsyn: Yeah. From chinese perspective japanese is different. From european it's not. ;) | 16:04 |
-!- lambday [67157f4c@gateway/web/freenode/ip.103.21.127.76] has joined #shogun | 16:06 | |
lisitsyn | thoralf: iglesiasg had to listen russian quite a lot in berlin :D | 16:06 |
@iglesiasg | haha not that much :P | 16:06 |
lisitsyn | iglesiasg: all the time you walked with us barbarians | 16:07 |
HeikoS1 | can anyone help me compiling with json? | 16:07 |
lisitsyn | iglesiasg: oh I remember the FBI thing | 16:07 |
@iglesiasg | lisitsyn, barbarians? I wouldn't say that taking into account the Russian embassy in Berlin hehe | 16:07 |
HeikoS1 | what do I need to pass to my program compiler if I used Cmake to fetch it? | 16:07 |
@iglesiasg | HeikoS1, do you need to give an argument to cmake? | 16:08 |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has joined #shogun | 16:08 | |
travis-ci | [travis-ci] it's Fernando Iglesias'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/10622920 | 16:08 |
-!- travis-ci [~travis-ci@ec2-23-23-71-239.compute-1.amazonaws.com] has left #shogun [] | 16:08 | |
lisitsyn | iglesiasg: hahah yes it takes a half of under den linden | 16:08 |
@iglesiasg | HeikoS1, in that case, have you tried taking a look using ccmake? | 16:08 |
HeikoS1 | so I told cmake to bundle json | 16:08 |
HeikoS1 | and now I have a little .cpp file that I would like to compile against shogun with json | 16:08 |
HeikoS1 | so the unit tests work with it | 16:09 |
HeikoS1 | but I would like to compile my own program | 16:10 |
@iglesiasg | -l shogun? | 16:10 |
@iglesiasg | maybe you need to include a directory using -L | 16:10 |
@iglesiasg | if I remember correctly I needed to use -L from the moment we switched to cmake | 16:10 |
shogun-buildbot | build #1612 of deb3 - modular_interfaces is complete: Failure [failed test ruby_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1612 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 16:10 |
HeikoS1 | so first weird thing is that HAVE_JSON is 0 | 16:11 |
HeikoS1 | so I manually set it to 1 | 16:11 |
HeikoS1 | then I added -I path/to/json | 16:11 |
HeikoS1 | and -ljson | 16:11 |
HeikoS1 | how do I again tell the compiler where the lib is? | 16:11 |
HeikoS1 | l L ? | 16:11 |
HeikoS1 | -L? | 16:11 |
@iglesiasg | -L for the path to the dir where the lib is, yes | 16:11 |
@iglesiasg | and then -l for the lib | 16:11 |
HeikoS1 | but it is in a std place | 16:12 |
HeikoS1 | :/ | 16:12 |
@iglesiasg | HeikoS1, but if HAVE_JSON is 0, it sounds like shogun didn't compile JSON stuff | 16:12 |
HeikoS1 | mmh | 16:12 |
HeikoS1 | lets see | 16:12 |
HeikoS1 | iglesiasg: the unit tests with json run | 16:12 |
@iglesiasg | mmmm I think that is a bit weird :) | 16:13 |
@iglesiasg | are they guarded with HAVE_JSON? | 16:13 |
HeikoS1 | maybe ill use my local json | 16:14 |
HeikoS1 | since i have it installed | 16:14 |
HeikoS1 | so recompiling | 16:14 |
shogun-buildbot | build #1318 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1318 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 16:21 |
HeikoS1 | wiking: is cmake done with json? | 16:22 |
HeikoS1 | seems that there are some weird things going on | 16:22 |
HeikoS1 | mmh on the other hand the unit tests are running ... | 16:23 |
HeikoS1 | ah confused | 16:23 |
-!- foulwall [~user@2001:da8:215:c252:39e3:9dc6:650c:106a] has joined #shogun | 16:26 | |
foulwall | ls | 16:27 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 16:31 | |
shogun-buildbot | build #1501 of deb2 - static_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1501 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 16:43 |
shogun-buildbot | build #1614 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1614 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 16:44 |
shogun-buildbot | build #1613 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/1613 | 16:47 |
shogun-buildbot | build #1531 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1531 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 16:48 |
votjakovr | HeikoS1: how accurately should be an estimate value of marginal likelihood with num_importance_samples=10000? i have the difference less than 1e-3 for EP. Is it ok? | 16:53 |
-!- thoralf [~thoralf@enki.zib.de] has quit [Quit: Konversation terminated!] | 16:55 | |
votjakovr | HeikoS1: in contrast difference for Laplace approximation is something about 1e-1 | 16:58 |
votjakovr | HeikoS1: i played a little bit with your InferenceMethod.get_log_ml_estimate_binary_logit_laplace unit test :) | 16:59 |
shogun-buildbot | build #1319 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1319 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 17:04 |
HeikoS1 | votjakovr: hi sorry I was out for some food | 17:05 |
HeikoS1 | votjakovr: the EP should converge faster than the laplace | 17:05 |
HeikoS1 | votjakovr: the number of samples one needs to get an accurate estimate depends on the dimension of the GP | 17:06 |
HeikoS1 | so for more data points, once needs a larger number | 17:06 |
HeikoS1 | votjakovr: its good to hear that the EP does a better job, this means less samples needed | 17:06 |
HeikoS1 | although the sampling should not be too expensive, if you look at the code, youll see it | 17:07 |
HeikoS1 | computing the approximation is more expensive | 17:07 |
HeikoS1 | votjakovr: btw | 17:08 |
HeikoS1 | there is *no* way of computing the exact marginal likelihood | 17:08 |
HeikoS1 | so no comparison | 17:08 |
HeikoS1 | you are comparing to an approximation | 17:08 |
votjakovr | HeikoS1: yeah | 17:08 |
HeikoS1 | the sampling is the only way to get an exact estimate | 17:09 |
HeikoS1 | the cool thing it allows to sample from the marginal posterior over the hyperparameters | 17:09 |
HeikoS1 | which is useful for model-selection the Bayesian way (not single point estimates but averaging over distributions) | 17:09 |
votjakovr | HeikoS1: yep, i know :) | 17:10 |
HeikoS1 | votjakovr: if you are interested, we can add methods for that to shogun later this year | 17:11 |
votjakovr | HeikoS1: something like mcmc? | 17:12 |
HeikoS1 | votjakovr: well that is a lot of work (though I am also planning doing that= | 17:12 |
HeikoS1 | but for now I meant just given one has samples, average predictions over them | 17:12 |
HeikoS1 | so monte carlo integration | 17:13 |
HeikoS1 | then, sampling can work with that | 17:13 |
HeikoS1 | mcmc is a big thing with many options | 17:13 |
HeikoS1 | so not that easy to offer black-box methods there | 17:13 |
HeikoS1 | but it is possible | 17:13 |
HeikoS1 | mcmc might be a gsoc project next year | 17:14 |
votjakovr | HeikoS1: cool :) | 17:15 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: WeeChat 0.4.0] | 17:16 | |
-!- pickle27 [~kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun | 17:31 | |
shogun-buildbot | build #1320 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1320 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 17:46 |
-!- travis-ci [~travis-ci@ec2-174-129-166-170.compute-1.amazonaws.com] has joined #shogun | 17:48 | |
travis-ci | [travis-ci] it's Heiko Strathmann'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/10624218 | 17:48 |
-!- travis-ci [~travis-ci@ec2-174-129-166-170.compute-1.amazonaws.com] has left #shogun [] | 17:48 | |
-!- foulwall` [~user@2001:da8:215:c252:3025:2d13:68db:4199] has joined #shogun | 17:55 | |
lambday | HeikoS1: there? | 17:57 |
-!- foulwall [~user@2001:da8:215:c252:39e3:9dc6:650c:106a] has quit [Ping timeout: 264 seconds] | 17:57 | |
HeikoS1 | lambday: yes, hi! | 17:58 |
lambday | just got your mail | 17:58 |
lambday | umm...so far I remember, it was an warning, right? | 17:58 |
lambday | or I put error :-/ | 17:58 |
HeikoS1 | lambday: ah yes, I am currently trying to fix json serialization tests | 17:58 |
lambday | HeikoS1: I am not too sure but I think we can't serialize for complex with json...had some trouble | 17:59 |
HeikoS1 | lambday: cant we just make two floats? | 17:59 |
HeikoS1 | I remember now :) | 17:59 |
lambday | yes | 17:59 |
lambday | but it was left that time... | 17:59 |
HeikoS1 | I see | 17:59 |
lambday | as in, if someone needs this, then only we'll add support | 18:00 |
HeikoS1 | ok dont worry too much, just keep in mind to fix it | 18:00 |
HeikoS1 | it breaks serialization tests | 18:00 |
lambday | and now is the time | 18:00 |
HeikoS1 | yep :D | 18:00 |
lambday | alright | 18:00 |
lambday | :D | 18:00 |
lambday | oh and matrix power's sparsity pattern works fine | 18:00 |
-!- van51 [~van51@ppp-94-66-12-14.home.otenet.gr] has joined #shogun | 18:00 | |
lambday | just added a unit test | 18:00 |
lambday | now finishing the probing sampler unit test | 18:01 |
HeikoS1 | hey, cool! | 18:01 |
lambday | then will send PR | 18:01 |
HeikoS1 | cool, how long does the graph colouring take? | 18:01 |
lambday | haven't tested that yet :-/ | 18:01 |
lambday | only testing with small ones till now | 18:02 |
lambday | with them its pretty fast | 18:02 |
HeikoS1 | ok | 18:02 |
HeikoS1 | really curious | 18:02 |
HeikoS1 | for the ozone matrix | 18:02 |
lambday | yes once I add the unit-test then will try with the ozone one | 18:03 |
HeikoS1 | I am excited :) | 18:03 |
lambday | HeikoS1: so, there are these distance-1 distance-2 colorings and various coloring methods, I am setting them as params | 18:04 |
lambday | don't wanna make it fixed | 18:04 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * e48b2b4 / tests/unit/io/ (2 files): https://github.com/shogun-toolbox/shogun/commit/e48b2b43c7667f12e6538e9808e21aaeeeb6c8fd | 18:05 |
shogun-notifier- | shogun: added accuracy for template classes | 18:05 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * 0610fb9 / tests/unit/io/SerializationJSON_unittest.cc.jinja2: https://github.com/shogun-toolbox/shogun/commit/0610fb952688155bcf938f386c45ac693e175ac0 | 18:05 |
shogun-notifier- | shogun: json seems to have a low accuracy | 18:05 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * 92f01a9 / tests/unit/CMakeLists.txt: https://github.com/shogun-toolbox/shogun/commit/92f01a96b686120c2c158928b20326101df07e8f | 18:05 |
shogun-notifier- | shogun: activate json since most things are working | 18:05 |
shogun-notifier- | shogun: Heiko Strathmann :feature/SerialUTests * 3602ec7 / src/shogun/io/SerializableJsonFile.cpp: https://github.com/shogun-toolbox/shogun/commit/3602ec74749d86ec71d6af711a969d34b71f16c2 | 18:05 |
shogun-notifier- | shogun: replace warning by error since otherwise shogun runs into uninitialized memory problems | 18:05 |
HeikoS1 | lambday: thats good | 18:06 |
HeikoS1 | lambday: we can experiment with that | 18:06 |
HeikoS1 | we basically just need *a* colouring | 18:06 |
HeikoS1 | in my experiments I used to compute it once and then even fix, even if the matrix changed | 18:06 |
lambday | alright | 18:07 |
lambday | we'll have default values for them, but keeping the option to experiment later on might be good :) | 18:07 |
HeikoS1 | lambday: totally, we can experiment ourselves and then change the default values | 18:20 |
@wiking | HeikoS1: i would ditch the json serialization | 18:29 |
HeikoS1 | wiking: why? | 18:29 |
@wiking | HeikoS1: and go with msgpack or avro | 18:29 |
HeikoS1 | wiking: btw accuracy is 1e-6 | 18:29 |
@wiking | HeikoS1: because that's some homemade hack | 18:29 |
@wiking | HeikoS1: instead of using some standard stuf | 18:29 |
@wiking | f | 18:29 |
HeikoS1 | wiking: up for that | 18:29 |
HeikoS1 | sonney2k wrote that not so long ago | 18:30 |
HeikoS1 | wiking: feel free to remove I dont object | 18:30 |
HeikoS1 | wiking: all other things are working now | 18:30 |
HeikoS1 | so branch ready to merge once this is done | 18:30 |
HeikoS1 | wiking: json is used by many languages though | 18:31 |
HeikoS1 | ipython for example | 18:31 |
@wiking | HeikoS1: yes | 18:31 |
HeikoS1 | but yeah as said, I am happy to drop things | 18:31 |
@wiking | HeikoS1: but it cannot be used as it is now | 18:31 |
@wiking | in any other parts | 18:31 |
@wiking | because one would need to know how to deserialize | 18:32 |
HeikoS1 | wiking: really? that sounds really nice | 18:32 |
HeikoS1 | haha, | 18:32 |
HeikoS1 | so then lets delete it | 18:32 |
@wiking | i mean json is just a syntax | 18:32 |
@wiking | one needs to know how to deserialize it | 18:32 |
@wiking | and at least msgpack and avro is out there already for a while | 18:32 |
@wiking | been used for a longer time now | 18:32 |
HeikoS1 | well for that we need to write another class | 18:32 |
@wiking | hence we really dont need to test it if it works at all | 18:32 |
@wiking | HeikoS1: as said in my comments | 18:33 |
@wiking | i've already started msgpack | 18:33 |
HeikoS1 | wiking: ah well ok | 18:33 |
HeikoS1 | then go | 18:33 |
HeikoS1 | wiking: could you then delete the old things and update our feature branch and merge it? | 18:33 |
HeikoS1 | since rest works | 18:33 |
HeikoS1 | serialization though leaks memory sometimes | 18:33 |
HeikoS1 | json also does :) | 18:33 |
@wiking | HeikoS1: hehe ok i'll check on it later today | 18:34 |
@wiking | i gotta fix some shit still | 18:34 |
HeikoS1 | wiking: pretty cool with the automated tests now | 18:34 |
HeikoS1 | was a lot work work but I think its worth it | 18:34 |
@wiking | yeah i mean serialization is necessary | 18:34 |
HeikoS1 | fixed a lot of bugs which would have scared of people | 18:34 |
HeikoS1 | wiking: what do you think about different versions of shogun files? | 18:35 |
HeikoS1 | should we support that? | 18:35 |
HeikoS1 | or only "latest" version | 18:35 |
@wiking | HeikoS1: well... yeah that's a good question | 18:35 |
@wiking | but yeah let's first have one stable version ;) | 18:35 |
HeikoS1 | wiking: the migration framework in principle is able to translate older versions | 18:35 |
HeikoS1 | but its a horror to maintain, the code is hard to understand etc etc | 18:35 |
HeikoS1 | I would just go for "not supported" in this case | 18:36 |
HeikoS1 | and tell users to save their scripts so that they can re-generate their instances | 18:36 |
HeikoS1 | if we had more people that would be feasible | 18:36 |
HeikoS1 | I worked on that for a lot of hours and it doesnt even work now | 18:36 |
HeikoS1 | its hard to use, everyon ignores it etc | 18:36 |
HeikoS1 | so rather drop | 18:36 |
@wiking | heheh indeed :) | 18:37 |
HeikoS1 | ok thats next step after your msgpack is there | 18:37 |
HeikoS1 | ok cool, I will stop shogunning now and do something else, let me know once the serailU brauch is merged | 18:37 |
@wiking | HeikoS1: will do | 18:37 |
-!- foulwall` [~user@2001:da8:215:c252:3025:2d13:68db:4199] has quit [Ping timeout: 264 seconds] | 18:56 | |
-!- sonne|osx [~sonne@89.204.135.161] has joined #shogun | 19:06 | |
-!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has joined #shogun | 19:10 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 19:10 | |
sonne|osx | iglesiasg: so the travis stuff works - hmhh weird but excellent :) | 19:11 |
@iglesiasg | sonne|osx, I just removed the second ppa, so it is basically as you left it before that last commit | 19:11 |
@iglesiasg | as you said, the sudo sudo thing didn't trigger the problem | 19:11 |
sonne|osx | iglesiasg: yeah but it wasn't building before | 19:12 |
sonne|osx | no idea | 19:12 |
@iglesiasg | mmm I think it was, let me show you | 19:12 |
@iglesiasg | sonne|osx, https://travis-ci.org/shogun-toolbox/shogun/builds | 19:12 |
@iglesiasg | after the answer "yes" it was all good | 19:13 |
-!- travis-ci [~travis-ci@ec2-174-129-166-170.compute-1.amazonaws.com] has joined #shogun | 19:14 | |
travis-ci | [travis-ci] it's Heiko Strathmann'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/10629960 | 19:14 |
-!- travis-ci [~travis-ci@ec2-174-129-166-170.compute-1.amazonaws.com] has left #shogun [] | 19:14 | |
@iglesiasg | but now we are bad again xD | 19:15 |
@iglesiasg | there is something wrong factor graph | 19:15 |
@iglesiasg | let's see if hushell comes and can fix it some time soon | 19:15 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 64b8e3e / CMakeLists.txt: https://github.com/shogun-toolbox/shogun/commit/64b8e3e9152179404ff2dea48799636807c421b9 | 19:24 |
shogun-notifier- | shogun: improve cmake summary | 19:24 |
sonne|osx | iglesiasg: ohh well :) | 19:25 |
-!- sonne|osx [~sonne@89.204.135.161] has quit [Quit: sonne|osx] | 19:27 | |
shogun-buildbot | build #1502 of deb2 - static_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/1502 blamelist: Soeren Sonnenburg <sonne@debian.org> | 19:28 |
-!- foulwall` [~user@2001:da8:215:c252:ad2d:8912:7f6d:5445] has joined #shogun | 19:30 | |
shogun-buildbot | build #1532 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1532 blamelist: Soeren Sonnenburg <sonne@debian.org> | 19:33 |
lambday | sorry I was away :( | 19:40 |
shogun-notifier- | shogun-data: van51 :master * f3c9732 / lang_detection/lang_dataset.dat.xz,lang_detection/lang_label.txt.xz: https://github.com/shogun-toolbox/shogun-data/commit/f3c9732a63e57a60e525d43268e1d5a6f8f874e6 | 19:49 |
shogun-notifier- | shogun-data: Added lang_detection data | 19:49 |
shogun-notifier- | shogun-data: Soeren Sonnenburg :master * a48aa9c / lang_detection/lang_dataset.dat.xz,lang_detection/lang_label.txt.xz: https://github.com/shogun-toolbox/shogun-data/commit/a48aa9c34b3661a1a302dea9230feac62617ce13 | 19:49 |
shogun-notifier- | shogun-data: Merge pull request #13 from van51/develop | 19:49 |
shogun-notifier- | shogun-data: | 19:49 |
shogun-notifier- | shogun-data: Added lang_detection data | 19:49 |
shogun-notifier- | shogun: Kevin :develop * cc30a5a / src/interfaces/octave_modular/swig_typemaps.i: https://github.com/shogun-toolbox/shogun/commit/cc30a5ac885fdd52ffc6228ed03fda47ea40a719 | 19:51 |
shogun-notifier- | shogun: SGNDArray octave in typemap fix | 19:51 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 2611cb1 / src/interfaces/octave_modular/swig_typemaps.i: https://github.com/shogun-toolbox/shogun/commit/2611cb1d63cc417917e81db4a75344c667ba1a75 | 19:51 |
shogun-notifier- | shogun: Merge pull request #1465 from pickle27/develop | 19:51 |
shogun-notifier- | shogun: | 19:51 |
shogun-notifier- | shogun: SGNDArray octave in typemap fix | 19:51 |
-!- iglesiasg [~iglesias@2001:6b0:1:1da0:19a:206d:eb6b:cdad] has quit [Quit: Ex-Chat] | 19:56 | |
-!- foulwall` [~user@2001:da8:215:c252:ad2d:8912:7f6d:5445] has quit [Ping timeout: 245 seconds] | 19:58 | |
@sonney2k | van51, once you have cleaned up your stuff you should talk to foulwall about how to do that in a demo :) | 19:59 |
van51 | sonney2k: sure :) | 19:59 |
van51 | sonney2k: I tried to start it on my own, but I got lost a bit | 19:59 |
-!- HeikoS1 [~heiko@nat-188-24.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 20:01 | |
shogun-buildbot | build #1321 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1321 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:12 |
shogun-buildbot | build #1615 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/1615 | 20:15 |
van51 | sonney2k: I 've made a PR about random fourier feats | 20:16 |
van51 | sonney2k: start looking into it when you have time! | 20:16 |
@wiking | HeikoS: gsoc meeting or shogun meeting? | 20:23 |
@wiking | ah gsoc shogun | 20:23 |
@wiking | ok | 20:23 |
@lisitsyn | wiking: wedding again? ;) | 20:29 |
@wiking | yeah | 20:30 |
@wiking | not mine :) | 20:30 |
shogun-buildbot | build #1616 of deb3 - modular_interfaces is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1616 blamelist: Soeren Sonnenburg <sonne@debian.org>, Kevin <kevinhughes27@gmail.com> | 20:42 |
shogun-buildbot | build #1503 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/1503 | 20:45 |
shogun-buildbot | build #1533 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1533 blamelist: Soeren Sonnenburg <sonne@debian.org>, Kevin <kevinhughes27@gmail.com> | 20:47 |
-!- thoralf [~thoralf@91-64-2-67-dynip.superkabel.de] has joined #shogun | 20:54 | |
thoralf | Heyho. | 20:55 |
-!- van51 [~van51@ppp-94-66-12-14.home.otenet.gr] has quit [Quit: Leaving.] | 21:07 | |
@sonney2k | wiking, you serious? | 21:13 |
@sonney2k | wiking, yours? | 21:13 |
@sonney2k | ahh ok good :) | 21:13 |
@lisitsyn | sonney2k: he had his in 2012 ;) | 21:17 |
shogun-notifier- | shogun: Thoralf Klein :develop * 6112665 / tests/unit/lib/SGSparseVector_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/61126659534bf8c185f318b0c8cfc74b0d555104 | 21:17 |
shogun-notifier- | shogun: Unit test for future changes on SGSparseVector. | 21:17 |
shogun-notifier- | shogun: Thoralf Klein :develop * b1e20ea / src/shogun/ (4 files): https://github.com/shogun-toolbox/shogun/commit/b1e20ea6253a02b35653dd6d2fc829ac6f0bbb1f | 21:17 |
shogun-notifier- | shogun: Avoid copy/malloc in SGSparseVector::sort_feature() : | 21:17 |
shogun-notifier- | shogun: * changed qsort_index to sort inline | 21:17 |
shogun-notifier- | shogun: * changed compression to work in line (and filter zero-entries) | 21:17 |
shogun-notifier- | shogun: * added switch "stable_pointers" to prevent re-alloc | 21:17 |
shogun-notifier- | shogun: * added unit-tests for new behaviour (in previous commit) | 21:17 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 41d7a03 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/41d7a0319758b64a507eb011e778c49c7f861ec6 | 21:17 |
shogun-notifier- | shogun: Merge pull request #1467 from tklein23/sort_features_refactoring | 21:17 |
shogun-notifier- | shogun: | 21:17 |
shogun-notifier- | shogun: Last bit of sort_features() refactoring | 21:17 |
@sonney2k | lisitsyn, yeah that is why :D | 21:18 |
@sonney2k | lisitsyn, was thinking he is a serial marriage type of guy | 21:18 |
-!- iglesiasg [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 21:19 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 21:19 | |
shogun-buildbot | build #1898 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1898 blamelist: Thoralf Klein <thoralf.klein@zib.de> | 21:22 |
@wiking | sonney2k: heheheh was thinking? :)) | 21:23 |
* wiking has a deadline :S | 21:26 | |
-!- iglesiasg [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Connection reset by peer] | 21:26 | |
shogun-buildbot | build #1322 of cyg1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1322 blamelist: Soeren Sonnenburg <sonne@debian.org>, Kevin <kevinhughes27@gmail.com> | 21:26 |
-!- iglesiasg [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 21:27 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 21:27 | |
@wiking | iglesiasg lisitsyn HeikoS up for tomorrow's early meeting? :) | 21:27 |
@lisitsyn | early is? | 21:27 |
@wiking | mmmm check doodle | 21:27 |
@lisitsyn | 12 UTC? | 21:28 |
@wiking | 10-2 CET | 21:28 |
@iglesiasg | wiking, was going to check right now | 21:28 |
@lisitsyn | waaat? | 21:28 |
@iglesiasg | but yeah, sure | 21:28 |
@wiking | lisitsyn: not that long | 21:28 |
@wiking | lisitsyn: anytime between that | 21:28 |
@lisitsyn | what is CET | 21:28 |
@lisitsyn | ;D | 21:28 |
@wiking | central european time | 21:28 |
@wiking | GMT+2 atm | 21:28 |
@lisitsyn | GMT+1? | 21:28 |
@wiking | i thnk | 21:28 |
@lisitsyn | agh shit | 21:28 |
@wiking | mmm CEST | 21:28 |
@wiking | it's now powersaving mode | 21:29 |
@wiking | so i guess it's GMT+2 | 21:29 |
@lisitsyn | I hate time zones | 21:29 |
@lisitsyn | hate hate | 21:29 |
@wiking | lisitsyn: UTC/GMT +2 hours | 21:29 |
@lisitsyn | so it should be 12-14 for me | 21:29 |
@wiking | should be 12-4 yeah | 21:30 |
@wiking | anytime between that | 21:30 |
@lisitsyn | okay I am ok | 21:30 |
@wiking | put on doodle which interval is good for ya | 21:30 |
@lisitsyn | let me put it | 21:30 |
@sonney2k | lisitsyn, GMT+2 I know that for sure. I only ever work w/ UTC | 21:30 |
@lisitsyn | sonney2k: I jump from gmt+4 to gmt+5 back and forth all the time | 21:31 |
@lisitsyn | due to mind disabled government | 21:31 |
@lisitsyn | :D | 21:31 |
@lisitsyn | so this makes me hate it more and more | 21:32 |
-!- hushell [~hushell@c-98-232-178-161.hsd1.or.comcast.net] has joined #shogun | 21:32 | |
@wiking | sonney2k: server status? comment on msgpack? :D | 21:33 |
@sonney2k | who is batman? | 21:35 |
@sonney2k | wiking, msgpack? | 21:35 |
@sonney2k | wiking, server still no news.. waiting for action | 21:35 |
-!- iglesiasg is now known as batman_ | 21:37 | |
@sonney2k | wiking, messagepack? sure feel free to write another reader/writer. All you have to do is supply a SerializableReader/Writer | 21:39 |
@wiking | sonney2k: i know ... i was just thnking that since we have some problems with json itself... why not switch to some more standard lib... | 21:41 |
@wiking | and drop the native json serialization | 21:41 |
@sonney2k | wiking, well we are not forced to use json and it is good to have | 21:41 |
@sonney2k | wiking, well json is a standard many people use | 21:42 |
@sonney2k | yes it has flaws | 21:42 |
@sonney2k | but java/.js web-people use it | 21:42 |
@wiking | yeah but that's why i'm saying that let's use a lib for that | 21:43 |
@wiking | like msgpack | 21:43 |
@wiking | or avro | 21:43 |
-!- batman_ is now known as iglesiasg | 21:45 | |
@sonney2k | wiking, not sure avro would work? Doesn't it need static schemas? | 21:47 |
@wiking | sonney2k: autogen them? :) | 21:49 |
shogun-buildbot | build #1899 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1899 blamelist: Soeren Sonnenburg <sonne@debian.org>, Thoralf Klein <thoralf.klein@zib.de> | 21:58 |
thoralf | sonney2k: Damn. I'm already using c++11 in my tests, but this buildbot isn't. | 21:59 |
thoralf | sonney2k: How do you handle this usually? ifndef xxx? | 21:59 |
@sonney2k | thoralf, yes | 22:03 |
-!- lambday [67157f4c@gateway/web/freenode/ip.103.21.127.76] has quit [Ping timeout: 250 seconds] | 22:04 | |
hushell | Did you get this error? tests/unit/lib/SGSparseVector_unittest.cc:215:73: warning: '>>' operator will be treated as two right angle brackets in C++0x [-Wc++0x-compat] | 22:08 |
thoralf | hushell: Its my fault. | 22:09 |
thoralf | hushell: Your compiler does not know c++09/c++11? | 22:09 |
hushell | thoralf: I think it should support, I am using g++4.6 | 22:10 |
hushell | thoralf: I haven't fetch for few days. It works fine before | 22:11 |
thoralf | hushell: Do a checkout of something before the last merge. ;) | 22:12 |
thoralf | sonney2k, wiking: How can I disable c++11 while building? I'd like to test my changes locally... | 22:16 |
@wiking | thoralf: :D | 22:21 |
hushell | thoralf: Maybe you edit SET(HAVE_CXX11 0) in CMakeLists.txt | 22:21 |
@sonney2k | thoralf, ask wiking he is the god of hell fire | 22:25 |
thoralf | hushell: Hmm... but then I also would need to remove the appended options from CXX_FLAGS. | 22:25 |
* thoralf removed the whole block from CMakeLists.txt | 22:26 | |
thoralf | Should work for now. :D | 22:26 |
@sonney2k | thoralf, heh | 22:26 |
@sonney2k | I totally don't get why some stuff in cmake cannot be -DTURNED_OFF=TRUE! | 22:27 |
thoralf | sonney2k: god of hell fire? sounds like Prodigy or sth similar? | 22:27 |
@lisitsyn | F-G INSTIGATOR | 22:30 |
@lisitsyn | I AM FIRESTARTER | 22:30 |
@lisitsyn | thoralf: the songs of my childhood! | 22:32 |
thoralf | Prodigy: Fire/jericho: "I am the god of Hell fire, and I bring you (fire)" :D | 22:33 |
@sonney2k | thoralf, can you fix the test quickly or shall I? | 22:33 |
thoralf | sonney2k: I'm on it. Will take a big unless I fix it by nesting two tests with #ifndef ... | 22:34 |
thoralf | s/bit/bit/ | 22:34 |
@sonney2k | thoralf, well you can just do > > (space!) and all good | 22:36 |
@sonney2k | no need to ifdef | 22:37 |
thoralf | sonney2k: No, it's also about the use of auto and "extended initializers". | 22:37 |
thoralf | sonney2k: I tried insering spaces and it raised many new errors. ;) | 22:38 |
@iglesiasg | pickle27, did you find many attacks to the English language in the notebook? :) | 22:38 |
pickle27 | iglesiasg: I don't think so, I read mainly for content, do you want me to read it again for language? | 22:39 |
@sonney2k | thoralf, you! | 22:39 |
@sonney2k | hmmhh | 22:39 |
@sonney2k | the cygwin stuff runs fine from cmdline | 22:39 |
@sonney2k | no idea why it crashes w/ the buildbot | 22:39 |
@sonney2k | wiking, running the installer for visual studio express on the bb now | 22:43 |
-!- iglesiasg [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 268 seconds] | 22:43 | |
@sonney2k | why does this need 5GB? | 22:45 |
thoralf | sonney2k: https://github.com/shogun-toolbox/shogun/pull/1468 - but we should wait for travis. I removed the tests, but I'm working on a real fix. | 22:45 |
* sonney2k is going to regret this | 22:45 | |
* thoralf fell in love with c++09/11's "auto" type declaration - and now has to remove it. | 22:54 | |
thoralf | What a pity! | 22:54 |
-!- travis-ci [~travis-ci@ec2-107-22-100-33.compute-1.amazonaws.com] has joined #shogun | 23:03 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/10636281 | 23:03 |
-!- travis-ci [~travis-ci@ec2-107-22-100-33.compute-1.amazonaws.com] has left #shogun [] | 23:03 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 23:13 | |
@sonney2k | microsoft visual studio express is loading user settings ... this may take a few minutes | 23:18 |
@sonney2k | express! | 23:18 |
@sonney2k | man once installed it wants to update itself | 23:19 |
@sonney2k | brilliant technology | 23:19 |
@sonney2k | wiking, now this crap is installed - so now what? | 23:19 |
@lisitsyn | hahahaha | 23:21 |
@lisitsyn | sonney2k: I guess you are cmaking stuff - just generate cmake project | 23:23 |
@lisitsyn | err | 23:23 |
@lisitsyn | VS project I mean | 23:23 |
@HeikoS | wiking, around | 23:27 |
@HeikoS | wiking, I can do until 1130 cest tomorrow | 23:28 |
@HeikoS | wiking, so 10 start would be good for me. 9 is too early | 23:28 |
@HeikoS | 1030 even better | 23:29 |
@HeikoS | So exactly the hour in between | 23:30 |
@HeikoS | The two existing ones | 23:30 |
shogun-notifier- | shogun: Thoralf Klein :develop * a01097d / tests/unit/lib/SGSparseVector_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/a01097da3f42b4e2a0f8e8bddc78d14c6b7b3ed0 | 23:31 |
shogun-notifier- | shogun: Removed two big unit tests to fix the build. (Used too many c++09/11 features like "auto" and extended initializers.) | 23:31 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 4e3362c / tests/unit/lib/SGSparseVector_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/4e3362c82d68b9c7fafea6879db2c1a7da9bf63d | 23:31 |
shogun-notifier- | shogun: Merge pull request #1468 from tklein23/build_hotfix | 23:31 |
shogun-notifier- | shogun: | 23:31 |
shogun-notifier- | shogun: Removed two unit tests to fix the build... | 23:31 |
shogun-buildbot | build #1534 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1534 blamelist: Soeren Sonnenburg <sonne@debian.org> | 23:53 |
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun | 23:54 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 23:54 | |
@iglesiasg | pickle27: nah, checking for language sounds too boring, I guess it is at least understandable which is the idea :) | 23:55 |
--- Log closed Tue Aug 27 00:00:32 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!