--- Log opened Tue Jan 28 00:00:13 2014 | ||
-!- thoralf [~thoralf@91-65-140-66-dynip.superkabel.de] has quit [Ping timeout: 272 seconds] | 00:49 | |
-!- zxtx [~zv@129-79-241-148.dhcp-bl.indiana.edu] has quit [Ping timeout: 260 seconds] | 06:29 | |
-!- zxtx [~zv@c-98-223-196-32.hsd1.in.comcast.net] has joined #shogun | 07:12 | |
-!- lisitsyn [~lisitsyn@80.252.20.67] has quit [Read error: Operation timed out] | 09:14 | |
-!- benibadman [~benibadma@94.135.236.129] has joined #shogun | 09:15 | |
-!- benibadm_ [~benibadma@94.135.236.129] has joined #shogun | 09:41 | |
-!- benibadman [~benibadma@94.135.236.129] has quit [Ping timeout: 245 seconds] | 09:45 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 10:00 | |
@wiking | sonney2k: lisitsyn: have u seen yesterday the conversation with naywhayare | 10:02 |
---|---|---|
@wiking | about their benchmarking system? | 10:02 |
lisitsyn | wiking: no let me read logs | 10:03 |
@wiking | lisitsyn: kk | 10:03 |
lisitsyn | wiking: oh it is pretty big ;) | 10:10 |
lisitsyn | tl;dr :D | 10:10 |
lisitsyn | wiking: jk, what is the important point here? | 10:10 |
@wiking | lisitsyn: basically that they've developed some sort benchmarking system | 10:11 |
@wiking | and i've suggested that they should just put that into a github repo | 10:11 |
@wiking | so that we can start hacking it as well | 10:11 |
lisitsyn | wiking: yeah that's opensource :D | 10:11 |
lisitsyn | wiking: we should definitely share libcovertree | 10:11 |
@wiking | yeah sure but now with that svn repo | 10:11 |
lisitsyn | which does not exist yet | 10:11 |
@wiking | it's not so easy to hack... | 10:12 |
lisitsyn | but jl is a kung fu guy of unsupported code | 10:12 |
@wiking | i mean contribute | 10:12 |
@wiking | jl? | 10:12 |
lisitsyn | yeah github is much better in any sense | 10:12 |
@wiking | libcovertree ? | 10:12 |
lisitsyn | wiking: :D | 10:12 |
lisitsyn | wiking: sorry that's another point | 10:12 |
lisitsyn | wiking: so I meant | 10:13 |
lisitsyn | john langford and his covertree | 10:13 |
lisitsyn | which naywhayare hates a bit | 10:13 |
lisitsyn | and actually I tried THRICE to patch it | 10:13 |
@wiking | http://hunch.net/~jl/projects/cover_tree/cover_tree.html? | 10:13 |
lisitsyn | wiking: yeah | 10:13 |
lisitsyn | wiking: we have his code in our repo as well | 10:13 |
@wiking | where? | 10:14 |
lisitsyn | but naywhayare claims it is buggy | 10:14 |
lisitsyn | wiking: some shogun/lib/ I guess | 10:14 |
@wiking | src/shogun/lib/tapkee/neighbors/covertree.hpp | 10:14 |
@wiking | ? | 10:14 |
lisitsyn | wiking: yeah that too | 10:16 |
lisitsyn | :D | 10:16 |
@wiking | ah somewhere else as well? :D | 10:17 |
lisitsyn | wiking: https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/lib/JLCoverTree.h | 10:17 |
lisitsyn | and this one | 10:17 |
@wiking | ah src/shogun/lib/JLCoverTree.h | 10:17 |
@wiking | yeah | 10:17 |
@wiking | why twice? :D | 10:17 |
lisitsyn | wiking: they are a bit different somehow | 10:17 |
@wiking | kk | 10:18 |
lisitsyn | but that's to be solved as well | 10:18 |
@wiking | well ok that's another story | 10:18 |
lisitsyn | wiking: what I meant is | 10:18 |
lisitsyn | we both need fast covertree | 10:18 |
lisitsyn | I mean shogun and mlpack | 10:18 |
lisitsyn | so I'd push push push some shared thing | 10:19 |
@wiking | yeah | 10:19 |
@wiking | sounds reasonable | 10:19 |
lisitsyn | it would be stupid to have two implementations of covertrees | 10:19 |
lisitsyn | :D | 10:19 |
@wiking | well u know | 10:19 |
@wiking | there's like millions of reimplementations :) | 10:19 |
@wiking | in shogun ::PP | 10:19 |
lisitsyn | yeah but no good covertree | 10:19 |
-!- Saurabh7 [~Saurabh7@115.248.130.148] has joined #shogun | 11:47 | |
-!- votjakovr [~votjakovr@188.134.46.30] has joined #shogun | 12:12 | |
-!- Saurabh7 [~Saurabh7@115.248.130.148] has quit [Quit: Leaving] | 12:18 | |
-!- HeikoS1 [~heiko@pat-191-250.internal.eduroam.ucl.ac.uk] has joined #shogun | 12:31 | |
votjakovr | HeikoS1: hi! i'm glad to see you :) how is it going with kernel->clone() issue? | 12:50 |
HeikoS1 | votjakovr: hi! good to see you too | 12:50 |
HeikoS1 | ah yeah, I will do that soonish | 12:51 |
HeikoS1 | currently some other things are very pressing regarding next gsoc | 12:51 |
HeikoS1 | votjakovr: how is it going for you=? | 12:51 |
HeikoS1 | votjakovr: thanks for you patch btw! | 12:52 |
HeikoS1 | votjakovr: good work as usual, ... speaking of how are the string examples going? did you find any time for thoise? | 12:53 |
votjakovr | HeikoS1: Oh, yeah :) Btw I'm trying to use GPs with preprocessed features and find a bug in parameters hashing framework | 12:53 |
HeikoS1 | votjakovr: ah bugs bugs ;) | 12:54 |
HeikoS1 | cool, preprcessed features would be nice | 12:55 |
HeikoS1 | I am currently working on categorical data | 12:55 |
HeikoS1 | with diffusion kernels | 12:55 |
HeikoS1 | votjakovr: and btw also using lots of your GPC code for the experiments, it works really nicely | 12:56 |
votjakovr | HeikoS1: cool, but i'd like to speed up them | 12:56 |
HeikoS1 | votjakovr: that would be very helpful indeed | 12:56 |
HeikoS1 | votjakovr: did you think about this shallow copying of the kernel? | 12:57 |
votjakovr | HeikoS1: nope, i'm waiting when you'll try to fix kernel->clone() issue :) | 12:59 |
HeikoS1 | votjakovr: I see | 13:00 |
HeikoS1 | so I will do that next | 13:00 |
votjakovr | HeikoS1: thanks :) | 13:00 |
HeikoS1 | votjakovr: btw what are your plans for the summer? | 13:04 |
votjakovr | HeikoS1: i still don't have plans | 13:07 |
HeikoS1 | votjakovr: are you interested in doing another gsoc? | 13:07 |
votjakovr | HeikoS1: i'm not a student unfortunately | 13:08 |
HeikoS1 | votjakovr: ah too bad | 13:08 |
HeikoS1 | votjakovr: do you want to stay in university or more into proper business world? | 13:09 |
HeikoS1 | do you even want to stay in Russia? | 13:09 |
HeikoS1 | votjakovr: sorry, questions over questions! | 13:09 |
HeikoS1 | votjakovr: in any case, you should stay involved in shogun! :) | 13:09 |
votjakovr | HeikoS1: i'm pretty flexible to change my location, but now i'm in Russia | 13:12 |
HeikoS1 | ever thought about doing msc or phd in machine learning? | 13:13 |
votjakovr | HeikoS1: actually nope :) | 13:13 |
votjakovr | HeikoS1: but it would be cool, i think | 13:13 |
HeikoS1 | votjakovr: so here in europe at least, that would be a quite good match to your skills, just to let you know | 13:14 |
HeikoS1 | I am sure many people would appreciate that | 13:14 |
votjakovr | HeikoS1: Yeah. Btw I think my patch may be merged | 13:45 |
lisitsyn | votjakovr: still in saint p? ;) | 14:23 |
votjakovr | lisitsyn: yeah | 14:26 |
-!- zxtx [~zv@c-98-223-196-32.hsd1.in.comcast.net] has quit [Ping timeout: 260 seconds] | 16:28 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 16:29 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 16:54 | |
shogun-notifier- | shogun: Roman Votyakov :develop * 0e35508 / / (7 files): https://github.com/shogun-toolbox/shogun/commit/0e35508d23b48d8c0e199e408fa55af1a3550c75 | 16:54 |
shogun-notifier- | shogun: various minor fixes in GPs framework | 16:54 |
shogun-notifier- | shogun: Heiko Strathmann :develop * e94c133 / / (7 files): https://github.com/shogun-toolbox/shogun/commit/e94c13395fea2bb1065bcac566e55c2e2497d749 | 16:54 |
shogun-notifier- | shogun: Merge pull request #1829 from votjakovr/feature/gp_refactoring | 16:54 |
shogun-notifier- | shogun: | 16:54 |
shogun-notifier- | shogun: Various minor fixes in GPs framework | 16:54 |
-!- zxtx [~zv@129-79-241-148.dhcp-bl.indiana.edu] has joined #shogun | 17:14 | |
shogun-buildbot | build #47 of deb4 - python3 is complete: Failure [failed test python modular] Build details are at http://buildbot.shogun-toolbox.org/builders/deb4%20-%20python3/builds/47 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 17:26 |
shogun-buildbot | build #410 of precise - libshogun is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/precise%20-%20libshogun/builds/410 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Roman Votyakov <votjakovr@gmail.com> | 17:38 |
-!- votjakovr [~votjakovr@188.134.46.30] has quit [Quit: WeeChat 0.4.0] | 17:47 | |
-!- travis-ci [~travis-ci@ec2-54-211-233-235.compute-1.amazonaws.com] has joined #shogun | 17:53 | |
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/17774414 | 17:53 |
-!- travis-ci [~travis-ci@ec2-54-211-233-235.compute-1.amazonaws.com] has left #shogun [] | 17:53 | |
-!- benibadm_ [~benibadma@94.135.236.129] has quit [Ping timeout: 252 seconds] | 17:58 | |
-!- lisitsyn [~lisitsyn@80.252.20.67] has joined #shogun | 18:18 | |
-!- thoralf [~thoralf@91-65-136-138-dynip.superkabel.de] has joined #shogun | 19:18 | |
thoralf | Heyho. | 19:18 |
-!- HeikoS1 [~heiko@pat-191-250.internal.eduroam.ucl.ac.uk] has quit [Ping timeout: 272 seconds] | 19:31 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 19:55 | |
-!- zxtx [~zv@129-79-241-148.dhcp-bl.indiana.edu] has quit [Ping timeout: 272 seconds] | 20:32 | |
-!- thoralf [~thoralf@91-65-136-138-dynip.superkabel.de] has quit [Read error: Operation timed out] | 20:50 | |
-!- thoralf [~thoralf@91-65-136-138-dynip.superkabel.de] has joined #shogun | 21:26 | |
@wiking | yo | 21:41 |
-!- zxtx [~zv@129-79-241-148.dhcp-bl.indiana.edu] has joined #shogun | 22:27 | |
-!- FSCV [~FSCV@50.7.50.60] has joined #shogun | 22:39 | |
thoralf | wiking: I found a case NCBM gets stuck due to numerical issues... and I discovered a few places where NaNs occur and result in non-terminating QP solver. (e.g. in the line search) | 22:43 |
thoralf | Interested in this or are you already over it? ;) | 22:43 |
@wiking | nono | 23:27 |
@wiking | just send the issue | 23:27 |
@wiking | and i'll give it a look | 23:28 |
thoralf | Cool. On my way. | 23:36 |
thoralf | https://github.com/shogun-toolbox/shogun/issues/1830 | 23:51 |
thoralf | The numerical stuff is at least indirectly related to the problem. | 23:52 |
thoralf | Hot-fixing only a few of the points turns out to solve the problem at least for my example. | 23:53 |
thoralf | But what makes me more thoughtful is that your code creates something that cannot be solved by our QP solver. | 23:54 |
@wiking | ? | 23:55 |
@wiking | what do u mean? | 23:56 |
@wiking | or what exactly do u mean about 'creates something that cannot be solved by our QP solver.'? | 23:56 |
thoralf | The qp solver does not converge. | 23:56 |
thoralf | It simply got stuck (not the first time though) | 23:56 |
@wiking | you mean ncbm gets stuck | 23:57 |
@wiking | right? | 23:57 |
thoralf | Probably NaNs in the OP. | 23:57 |
@wiking | k | 23:57 |
@wiking | will try to check | 23:57 |
@wiking | have a new qp solver idea | 23:57 |
@wiking | but no time to do it :< | 23:57 |
thoralf | Varying parameters in the problem yields *either* to non-converging qp solver or forever iterating ncbm. | 23:58 |
@wiking | btw is your problem concave? | 23:58 |
@wiking | i.e. non-convex | 23:58 |
thoralf | It's a toy example. Maybe. | 23:58 |
thoralf | Or, probably. | 23:58 |
thoralf | Never thought of, but good point. | 23:59 |
@wiking | since ncbm should be used mostly for non-convex problems | 23:59 |
@wiking | i mean that's what it's for actually | 23:59 |
@wiking | although that shouldn't mean that it cannot solve a standard convex problem | 23:59 |
--- Log closed Wed Jan 29 00:00:08 2014 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!