--- Log opened Wed Apr 08 00:00:34 2015 | ||
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has joined #shogun | 00:31 | |
travis-ci | it's Soumyajit De'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/57548889 | 00:31 |
---|---|---|
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has left #shogun [] | 00:31 | |
shogun-notifier- | shogun: lambday :develop * f4c4bf7 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/f4c4bf74f17fc60f6641e9534246dfe79e4a7f33 | 00:46 |
shogun-notifier- | shogun: Added matrix-type independent add in linalg | 00:46 |
shogun-notifier- | shogun: Soumyajit De :develop * 701191d / / (3 files): https://github.com/shogun-toolbox/shogun/commit/701191d6ec6bb7703aa1d147cf55aa9e2f4da242 | 00:46 |
shogun-notifier- | shogun: Merge pull request #2811 from lambday/feature/linalg | 00:46 |
shogun-notifier- | shogun: | 00:46 |
shogun-notifier- | shogun: Added matrix-type independent add in linalg | 00:46 |
shogun-buildbot | build #282 of trusty - libshogun - viennacl is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/282 blamelist: lambday <heavensdevil6909@gmail.com> | 00:50 |
shogun-buildbot | build #283 of trusty - libshogun - viennacl is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/283 blamelist: Soumyajit De <heavensdevil6909@gmail.com> | 00:52 |
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has joined #shogun | 00:57 | |
travis-ci | it's Soumyajit De'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/57561631 | 00:57 |
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has left #shogun [] | 00:57 | |
-!- PirosB3 [~pirosb3@host33-99-dynamic.181-80-r.retail.telecomitalia.it] has quit [Quit: PirosB3] | 01:02 | |
@lambday | lisitsyn: there? | 01:04 |
lisitsyn | lambday: hey | 01:04 |
@lambday | lisitsyn: how are things? :D | 01:05 |
lisitsyn | lambday: fine | 01:05 |
lisitsyn | what about you? | 01:05 |
lisitsyn | currently mapreducing some stuff around :) | 01:05 |
@lambday | lisitsyn: haha that's great! | 01:06 |
@lambday | lisitsyn: I have been working on hypothesis testing a bit | 01:06 |
@lambday | and lately (thanks to wu lin) doing some stuffs in linalg as well | 01:06 |
lisitsyn | hmm cool | 01:06 |
lisitsyn | I am on my way to refactor neural nets a bit | 01:07 |
@lambday | lisitsyn: one thing I want your opinion on | 01:07 |
lisitsyn | sure | 01:07 |
@lambday | lisitsyn: check this benchmark https://github.com/shogun-toolbox/shogun/pull/2806 | 01:07 |
@lambday | it comes from a discussion here https://github.com/shogun-toolbox/shogun/issues/2801 | 01:07 |
lisitsyn | oh something big | 01:08 |
@lambday | so long story short - | 01:08 |
@lambday | wu lin was using linalg methods without specifying any backend | 01:08 |
@lambday | and he's using SGMatrix | 01:08 |
lisitsyn | okay I've seen there is a bug | 01:08 |
lisitsyn | some forgotten loading right? | 01:09 |
@lambday | so if someone sets viennacl for global backend, then we're doing all the CPU-GPU data transfer | 01:09 |
@lambday | which stabs the performance at the heart | 01:09 |
lisitsyn | yeah that's obvious | 01:09 |
@lambday | lisitsyn: that is fixed - we just needed a variant that returned the result matrix/vector | 01:10 |
@lambday | ok so what I was thinking is | 01:10 |
@lambday | how about a cmake switch to make SGMatrix as GPU one shogun-wise? | 01:10 |
@lambday | we can just inherit from a separate base | 01:10 |
@lambday | based on some preprocessor flag | 01:11 |
@lambday | that way, we can milk the performance boost to the fullest when GPU is used | 01:11 |
@lambday | this is totally possible even now with explicitly using CGPUMatrix+ViennaCL backed | 01:11 |
lisitsyn | hmm let me think | 01:12 |
@lambday | but how about we use the same SGMatrix everywhere - with the backend changed in a switch | 01:12 |
@lambday | I am mainly worried because you see in the benchmark the penalty is unbearable | 01:13 |
lisitsyn | I see no crime in such switch for sure | 01:13 |
lisitsyn | but I have doubts we can hide gpu | 01:13 |
@lambday | but when we use GPU memory then gain makes me drool :D | 01:13 |
lisitsyn | I mean you always have to care whether you transfer data or not | 01:13 |
lisitsyn | btw check torch | 01:14 |
lisitsyn | https://github.com/torch/torch7 | 01:14 |
@lambday | lisitsyn: checking.. | 01:14 |
lisitsyn | neural guys love to use it | 01:14 |
@lambday | lisitsyn: what I was saying is something like | 01:14 |
@lambday | struct CPUMatrix | 01:14 |
@lambday | { | 01:15 |
@lambday | } | 01:15 |
@lambday | ; | 01:15 |
@lambday | struct GPUMatrix {}; | 01:15 |
@lambday | #ifdef USE_CPU | 01:15 |
lisitsyn | oh you finally talk C++ now I understand ;) | 01:15 |
@lambday | lol | 01:15 |
@lambday | you got the idea | 01:15 |
@lambday | so two bases | 01:15 |
lisitsyn | can we get more flexible with less compile time stuff? | 01:16 |
@lambday | what do you see the problem with that? | 01:16 |
lisitsyn | I am a bit worried about switching everything | 01:16 |
lisitsyn | I mean if sgmatrix is in gpu | 01:16 |
lisitsyn | some stuff would break right? | 01:16 |
@lambday | I'm thinking what would that be.. | 01:17 |
lisitsyn | most of code is not written for gpu so we should expect bugs here | 01:17 |
@lambday | well that makes sense.. | 01:18 |
@lambday | so devs should explicitly use gpumatrix if they wanna use one.. | 01:19 |
lisitsyn | lambday: it seems to me that sgmatrix should be kind of dynamic about that | 01:19 |
lisitsyn | maybe there is some solution to make it a bit more dynamic | 01:19 |
@lambday | lisitsyn: what do you mean by dynamic? | 01:19 |
lisitsyn | like sometimes it is gpu but sometimes it is cpu | 01:19 |
@lambday | lisitsyn: that's what I meant.. | 01:19 |
@lambday | #ifdef USE_CPU | 01:19 |
lisitsyn | I mean not globally | 01:19 |
@lambday | lisitsyn: okay.. so some param | 01:20 |
lisitsyn | well say in neural nets I have all these parameters in sgmatrix | 01:20 |
lisitsyn | that's cool to have it as gpu matrix | 01:20 |
lisitsyn | but some other algorithms do not expect gpu | 01:20 |
@lambday | lisitsyn: yeah that makes sense... | 01:21 |
lisitsyn | this doesn't fit your template linalg idea though | 01:21 |
@lambday | lisitsyn: what I don't like is that now we have two different types of matrices | 01:21 |
lisitsyn | that's ok they are really different | 01:21 |
lisitsyn | it seems there is no magic to treat them in the same way | 01:21 |
@lambday | yep.. apparently so :( | 01:22 |
@lambday | in linalg we're anyway treating them differently | 01:22 |
lisitsyn | yeah | 01:22 |
lisitsyn | I think sgmatrix should be like tensor in torch | 01:22 |
lisitsyn | they have some clean api it seems | 01:23 |
lisitsyn | btw what about sgmatrix->sgtensor? ;) | 01:23 |
@lambday | lol we do have that multidimensional array thing right? | 01:23 |
@lambday | maybe make that more like tension | 01:23 |
@lambday | oops | 01:23 |
@lambday | tensor | 01:23 |
@lambday | totally didn't mean tension :D | 01:24 |
lisitsyn | hah | 01:24 |
lisitsyn | that's actually super cool thing | 01:24 |
lisitsyn | that they can generalize | 01:24 |
lisitsyn | say you have RGB image | 01:24 |
@lambday | I'm assuming that it is efficient as well | 01:24 |
lisitsyn | and then you got RGB-D and its 3d | 01:25 |
lisitsyn | nothing works | 01:25 |
lisitsyn | :D | 01:25 |
@lambday | voxels ? :D | 01:25 |
lisitsyn | yeah like these | 01:25 |
lisitsyn | and when aliens get control of humanity | 01:25 |
lisitsyn | we get to 4d | 01:25 |
lisitsyn | would be bad to rewrite code again | 01:26 |
@lambday | gotta check it then! | 01:26 |
lisitsyn | aha | 01:26 |
lisitsyn | ohh these MR tasks go so slow | 01:27 |
@lambday | it's in C ! | 01:27 |
lisitsyn | yeah lua + C | 01:27 |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has joined #shogun | 02:18 | |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has quit [Remote host closed the connection] | 02:29 | |
-!- lambday [6a33f00b@gateway/web/freenode/ip.106.51.240.11] has quit [Ping timeout: 246 seconds] | 03:40 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 03:46 | |
shogun-buildbot | build #995 of nightly_default is complete: Failure [failed notebooks] Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/995 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, lambday <heavensdevil6909@gmail.com>, Wu Lin <yorker.lin@gmail.com>, Soumyajit De <heavensdevil6909@gmail.com> | 04:04 |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has joined #shogun | 04:11 | |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has quit [Remote host closed the connection] | 04:27 | |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has joined #shogun | 05:22 | |
-!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has quit [Remote host closed the connection] | 05:51 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 06:26 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 244 seconds] | 06:37 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 08:16 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 256 seconds] | 08:38 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 08:56 | |
-!- HeikoS [~heiko@05453a2b.skybroadband.com] has joined #shogun | 10:40 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 10:40 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 246 seconds] | 11:34 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 12:11 | |
shogun-notifier- | shogun: Wu Lin :develop * f3a5315 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/f3a53156500cd286d3f39e1c3133a86b436b6f9f | 12:11 |
shogun-notifier- | shogun: added the FITC Laplace method using lbfgs | 12:11 |
shogun-notifier- | shogun: Wu Lin :develop * 7c1d31f / src/shogun/machine/gp/ (2 files): https://github.com/shogun-toolbox/shogun/commit/7c1d31f9df70e021577c4750d674eedd6a222902 | 12:11 |
shogun-notifier- | shogun: add comments | 12:11 |
shogun-notifier- | shogun: Wu Lin :develop * 92b55e5 / tests/unit/machine/gp/SingleFITCLaplacianInferenceMethodWithLBFGS_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/92b55e52f12810776b3d6d4f059ebe6d02a8e4a6 | 12:11 |
shogun-notifier- | shogun: remove commented codes | 12:11 |
shogun-notifier- | shogun: Heiko Strathmann :develop * 202951f / / (3 files): https://github.com/shogun-toolbox/shogun/commit/202951f26b6c2e022d7f9cba38a50a06b3b11818 | 12:11 |
shogun-notifier- | shogun: Merge pull request #2804 from yorkerlin/fitc-lbfgs | 12:11 |
shogun-notifier- | shogun: | 12:11 |
shogun-notifier- | shogun: Fitc lbfgs | 12:11 |
shogun-buildbot | build #284 of trusty - libshogun - viennacl is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/284 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Wu Lin <yorker.lin@gmail.com> | 12:16 |
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has joined #shogun | 12:22 | |
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/57622020 | 12:22 |
-!- travis-ci [~travis-ci@ec2-54-226-91-37.compute-1.amazonaws.com] has left #shogun [] | 12:22 | |
shogun-buildbot | build #3402 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/3402 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Wu Lin <yorker.lin@gmail.com> | 12:57 |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 13:44 | |
-!- HeikoS [~heiko@05453a2b.skybroadband.com] has quit [Quit: Leaving.] | 14:20 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 15:11 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 256 seconds] | 15:56 | |
-!- HeikoS [~heiko@laptop-net.gatsby.ucl.ac.uk] has joined #shogun | 16:38 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 16:38 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 17:55 | |
-!- PirosB3 [~pirosb3@host33-99-dynamic.181-80-r.retail.telecomitalia.it] has joined #shogun | 18:02 | |
-!- HeikoS [~heiko@laptop-net.gatsby.ucl.ac.uk] has quit [Ping timeout: 244 seconds] | 20:29 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 244 seconds] | 20:42 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 22:12 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Read error: Connection reset by peer] | 22:12 | |
-!- kshitijk [~kshitijk@220.227.149.86] has joined #shogun | 22:13 | |
-!- kshitijk [~kshitijk@220.227.149.86] has quit [Ping timeout: 246 seconds] | 22:38 | |
--- Log closed Thu Apr 09 00:00:35 2015 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!