--- Log opened Mon Jun 06 00:00:28 2016 | ||
-!- GandalfTheWizar1 [~Eva@112.10.170.58] has quit [Ping timeout: 240 seconds] | 00:04 | |
-!- GandalfTheWizard [~Eva@112.10.170.58] has joined #shogun | 00:05 | |
-!- sonne|osx [~sonne@x4db3bf36.dyn.telefonica.de] has quit [Quit: sonne|osx] | 00:45 | |
-!- lambday [6d9941c0@gateway/web/freenode/ip.109.153.65.192] has quit [Ping timeout: 250 seconds] | 01:44 | |
-!- arianepaola [~ariane@unaffiliated/arianepaola] has quit [Read error: Connection reset by peer] | 04:53 | |
-!- arianepaola [~ariane@unaffiliated/arianepaola] has joined #shogun | 04:58 | |
-!- sonne|osx [~sonne@x4db3bf36.dyn.telefonica.de] has joined #shogun | 05:41 | |
-!- sonne|osx [~sonne@x4db3bf36.dyn.telefonica.de] has quit [Quit: sonne|osx] | 05:57 | |
-!- sanuj [~sanuj@117.203.23.253] has joined #shogun | 07:04 | |
sanuj | Saurabh7, hey | 07:33 |
---|---|---|
@wiking | hi | 07:35 |
@wiking | can i get a standup from both of you | 07:35 |
sanuj | wiking, hi | 07:35 |
@wiking | sanuj & Saurabh7 | 07:35 |
sanuj | wiking, My PR for tags is merge ready (reviewed by sergey) https://github.com/shogun-toolbox/shogun/pull/3221 | 07:36 |
sanuj | was working on this last week | 07:36 |
sanuj | wiking, lisitsyn insists that you or HeikoS review it once before we merge it | 07:37 |
@wiking | k | 07:37 |
@wiking | first of all you should stash your commits into 1 | 07:37 |
sanuj | wiking, there are lots of changes, you want one commit for that? | 07:37 |
@wiking | yes | 07:38 |
sanuj | wiking, neural net cookbook https://github.com/shogun-toolbox/shogun/pull/3246 | 07:38 |
@wiking | s/stash/squash/g | 07:38 |
@wiking | https://ariejan.net/2011/07/05/git-squash-your-latests-commits-into-one/ | 07:38 |
sanuj | i was having some problems with the cookbook | 07:38 |
sanuj | thanks for the link | 07:38 |
sanuj | i'm making a cookbook for mkl also | 07:41 |
sanuj | this was due last week but i got stuck with the nn cookbook | 07:41 |
Saurabh7 | hello | 07:51 |
Saurabh7 | sanuj: hey | 07:51 |
sanuj | Saurabh7, you wrote the mkl notebook? | 07:51 |
sanuj | i have some kernel doubts :) | 07:51 |
Saurabh7 | wiking: finsihed multithread support for unlcoked xval, working on polishing LARS and eigenising it, done for some important routines and added unit tests | 07:52 |
Saurabh7 | also fixed lars benchmark in the frameowrk | 07:53 |
Saurabh7 | sanuj: yep:) | 07:53 |
Saurabh7 | sanuj: whats the problem? | 07:53 |
sanuj | Saurabh7, so via kernels we are trying to project the data in a different space right? | 07:53 |
Saurabh7 | sanuj: yes | 07:54 |
sanuj | so that it is linearly separable in the new space | 07:54 |
Saurabh7 | sanuj: yes tahts the idea | 07:54 |
sanuj | Saurabh7, every kernel takes 2 features as argument | 07:54 |
sanuj | why is that? | 07:54 |
Saurabh7 | sanuj: its basically jsut like distance class | 07:55 |
Saurabh7 | kernel function si computed for kernel(i,j) | 07:55 |
Saurabh7 | i is lhs and j rhs features index | 07:55 |
sanuj | Saurabh7, what do you mean by lhs and rhs features | 07:56 |
sanuj | i saw this in the doc ^ | 07:56 |
sanuj | but it confused me | 07:56 |
Saurabh7 | sanuj: you need two points for calucalting say euclidean dist | 07:58 |
Saurabh7 | same way for kernel | 07:58 |
Saurabh7 | they can come from two different features | 07:58 |
Saurabh7 | matrices | 07:58 |
Saurabh7 | kernel fucntion between ith sample of lhs and jth sample of rhs | 07:59 |
sanuj | oh i see | 07:59 |
sanuj | Saurabh7, if you look at this https://github.com/shogun-toolbox/shogun/blob/develop/examples/undocumented/python_modular/mkl_binclass_modular.py | 08:01 |
sanuj | Saurabh7, so while training we keep lhs = rhs = training_features | 08:02 |
sanuj | and while testing we have lhs = training_features and rhs = testing_features | 08:03 |
Saurabh7 | sanuj: yes | 08:03 |
sanuj | Saurabh7, if we are generating the kernel matrices like this, then what does mkl.train() learns? | 08:04 |
sanuj | the weights of linear combination of kernels? | 08:05 |
Saurabh7 | sanuj: yes optimal weights | 08:05 |
sanuj | okay | 08:06 |
sanuj | Saurabh7, thanks :) | 08:06 |
Saurabh7 | sanuj: cool :) | 08:06 |
-!- sanuj [~sanuj@117.203.23.253] has quit [Ping timeout: 240 seconds] | 09:02 | |
-!- sanuj [~sanuj@117.203.23.253] has joined #shogun | 09:42 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has joined #shogun | 10:42 | |
c4goldsw | Hey, does anyone know if SGVector supports scalar multiplication? I'm browsing through the corresponding .h and .cpp file and I can't find anything, I just want to be certain. | 10:55 |
@wiking | c4goldsw: scale_vector | 10:56 |
@wiking | or | 10:56 |
@wiking | /// Scale vector inplace | 10:56 |
@wiking | void scale(T alpha); | 10:56 |
c4goldsw | wiking thanks! | 10:58 |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 11:01 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has joined #shogun | 11:14 | |
-!- lambday [8028b10a@gateway/web/freenode/ip.128.40.177.10] has joined #shogun | 11:36 | |
-!- mode/#shogun [+o lambday] by ChanServ | 11:36 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 12:11 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has joined #shogun | 12:11 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has quit [Client Quit] | 12:15 | |
-!- besser82 [~besser82@fedora/besser82] has joined #shogun | 12:18 | |
-!- mode/#shogun [+o besser82] by ChanServ | 12:18 | |
sanuj | Saurabh7, there? | 12:28 |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has joined #shogun | 13:49 | |
c4goldsw | lambday, you around? | 13:51 |
@lambday | c4goldsw: yeah | 13:51 |
c4goldsw | would you have time to look at a PR I made? | 13:51 |
@lambday | c4goldsw: sure.. | 13:57 |
c4goldsw | Great - it's not too big: https://github.com/shogun-toolbox/shogun/pull/3249 | 13:57 |
@lambday | c4goldsw: oh you added variance methods for linalg | 13:59 |
c4goldsw | Yes. | 13:59 |
c4goldsw | lambday: I'm just looking for feedback - my main concern is if I'm using eigen to the full extent that I can to take advantage of SIMD operations. | 14:00 |
c4goldsw | my main concern is to ensure that I'm using* | 14:00 |
@lambday | I see. | 14:01 |
@lambday | let me have a look. But did you have a look at https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/mathematics/Statistics.h#L60 ? | 14:01 |
@lambday | I wonder whether the method you added does the same | 14:02 |
@lambday | c4goldsw: | 14:02 |
c4goldsw | It is - Heiko wanted to move some stats methods out of CStatistics into linalg. The issue is here: https://github.com/shogun-toolbox/shogun/issues/3118 | 14:02 |
c4goldsw | lambday: | 14:03 |
@lambday | c4goldsw: ah I see | 14:03 |
c4goldsw | I guess this is being done so that these computations can be done with whatver linalg backend a user uses. | 14:03 |
@lambday | c4goldsw: okay, so first of all, we're changing the structure of linalg a bit.. so all this linalg code probably be refactored in some way.. this is what OXPHOS is working on for her GSoC project.. | 14:04 |
@lambday | but let me comment on the usage of the eigen3 part anyway | 14:04 |
-!- sanuj [~sanuj@117.203.23.253] has quit [Ping timeout: 246 seconds] | 14:04 | |
c4goldsw | lambday: Thanks! | 14:05 |
c4goldsw | lambday I read / replied to your comment - we can talk over here if that's easier. | 14:13 |
c4goldsw | lambday: now I understand what you said about replacing sqaureResult | 14:15 |
-!- sanuj [~sanuj@117.203.23.253] has joined #shogun | 14:19 | |
-!- sanuj [~sanuj@117.203.23.253] has quit [Client Quit] | 14:20 | |
-!- sanuj [~sanuj@117.203.23.253] has joined #shogun | 14:21 | |
lisitsyn | lambday: hey | 14:24 |
@lambday | lisitsyn: hello | 14:25 |
lisitsyn | lambday: could you please review sanuj's code? | 14:25 |
@lambday | lisitsyn: which PR you want me to have a look? | 14:25 |
sanuj | lambday, lisitsyn hey | 14:26 |
lisitsyn | in a sec | 14:26 |
sanuj | i'll give you the link | 14:26 |
@lambday | c4goldsw: yeah if possible, we can get rid of these temporaries | 14:26 |
lisitsyn | lambday: https://github.com/shogun-toolbox/shogun/pull/3221 | 14:26 |
lisitsyn | lambday: be super tough as it is critical! :) | 14:26 |
sanuj | yeah this | 14:26 |
@lambday | lisitsyn: thanks.. checking | 14:26 |
@lambday | hah | 14:26 |
sanuj | lambday, when i heiko coming back? | 14:26 |
c4goldsw | lambday sure - thanks for looking over it. | 14:26 |
@lambday | sanuj: should have been back by now | 14:27 |
sanuj | okay | 14:27 |
@lambday | shall I also comment on styling etc? | 14:28 |
sanuj | lisitsyn, ^ | 14:28 |
lisitsyn | lambday: yes sure | 14:28 |
lisitsyn | lambday: we target to merge that so lets polish that | 14:29 |
@lambday | ok | 14:29 |
c4goldsw | lambday another quick question: did the last compute() function also check out? There were no comments. | 14:29 |
sanuj | lisitsyn, wiking told me to squish the commits | 14:29 |
sanuj | i'll do that once there are no more changes to be done | 14:29 |
lisitsyn | sanuj: yes | 14:30 |
lisitsyn | makes sense | 14:30 |
sanuj | lisitsyn, we will merge this in shogun/tags or shogun/develop? | 14:30 |
lisitsyn | sanuj: I thought about develop | 14:30 |
sanuj | okay, the PR is against shogun/tags | 14:30 |
lisitsyn | sanuj: well probably it makes sense to get something complete in shogun/tags | 14:40 |
lisitsyn | like with swig | 14:40 |
sanuj | okay | 14:40 |
@lambday | sanuj: commented on your PR | 15:06 |
@lambday | there are some minor styling issues | 15:06 |
sanuj | yes the spaces | 15:06 |
sanuj | i'll fix those | 15:06 |
@lambday | just a couple of other things | 15:06 |
sanuj | lambday, thanks :) | 15:07 |
@lambday | maybe the constructors for BaseTag and Tag should be explicit | 15:07 |
sanuj | okay | 15:07 |
@lambday | also, whenever you add operator==, please also add operator!= | 15:07 |
@lambday | most important thing that is missing from the PR is class documentation | 15:07 |
@lambday | please add those.. for each method and class | 15:08 |
* lambday afk | 15:09 | |
sanuj | lambday, sure | 15:11 |
lisitsyn | lambday: nice catch on != | 15:11 |
@wiking | mmm | 15:13 |
@wiking | oxphos is not around yet? | 15:13 |
@wiking | arianepaola: how's the setup.py going/ | 15:14 |
@wiking | ? | 15:14 |
* lambday re | 15:36 | |
@wiking | re | 15:36 |
@wiking | fe | 15:36 |
@lambday | heh | 15:37 |
@lambday | lisitsyn: yeah that just caught my attention :D | 15:37 |
sanuj | re | 15:40 |
@lambday | (R)eal (E)vil | 15:41 |
sanuj | :P | 15:42 |
arianepaola | hi wiking, just to let you know that I will be online later today (presentation at uni). | 15:42 |
@lambday | c4goldsw: hey.. yeah I had a quick look.. the thing to care about in these methods is that we want to minimize (if possible, avoid) temporary object creation.. | 15:43 |
c4goldsw | lambday Okee Dokee | 15:44 |
@wiking | arianepaola: k | 15:47 |
-!- sanuj [~sanuj@117.203.23.253] has quit [Ping timeout: 252 seconds] | 15:49 | |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 15:52 | |
-!- OXPHOS [9d8b131c@gateway/web/freenode/ip.157.139.19.28] has joined #shogun | 16:05 | |
@wiking | OXPHOS: yo | 16:05 |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has joined #shogun | 16:06 | |
OXPHOS | wiking: hi | 16:08 |
OXPHOS | wiking: log: linalg working prototype. Now works with Eigen3 and ViennaCL. Also linalg benchmark. Cookbook: multi-class logistic regression and multi-class linear machine | 16:11 |
OXPHOS | lambday: hey | 16:12 |
@lambday | OXPHOS: hello | 16:12 |
@wiking | OXPHOS: should talk about cereal | 16:12 |
OXPHOS | wiking: you mean now? didn't do much last week | 16:13 |
@wiking | i know | 16:17 |
@wiking | but yeah | 16:17 |
@wiking | we should talk | 16:17 |
@wiking | as that part needs to get rolling as well | 16:18 |
-!- HeikoS [~heiko@nat-182-210.internal.eduroam.ucl.ac.uk] has joined #shogun | 16:21 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 16:21 | |
@HeikoS | jojojo | 16:21 |
OXPHOS | wiking: sure! I'll definitely read sanuj's code | 16:21 |
@HeikoS | Saurabh7: hi there | 16:21 |
@HeikoS | OXPHOS: hi! | 16:21 |
OXPHOS | HeikoS: hello! | 16:21 |
c4goldsw | HeikoS Hello | 16:21 |
@HeikoS | c4goldsw: ah hello | 16:21 |
@wiking | OXPHOS: ? what's that has to do with this? :) | 16:21 |
@wiking | i mean with cereal | 16:21 |
OXPHOS | wiking: we're doing it in parameter.h? or u mean sth. more basic | 16:22 |
@wiking | OXPHOS: so your problem was obvious | 16:22 |
@wiking | with sgvector | 16:23 |
@wiking | why did you keep the old load/save vector... because the problem is that since the new one is a template it'll unfortunately match some other code | 16:23 |
@wiking | *function | 16:23 |
@wiking | but | 16:24 |
@wiking | maybe we should use a different naming | 16:26 |
@wiking | like serialize | 16:26 |
@wiking | and deserialize | 16:26 |
-!- sanuj [~sanuj@117.203.18.75] has joined #shogun | 16:26 | |
@wiking | that way it wont have that problem | 16:26 |
lisitsyn | hey | 16:26 |
lisitsyn | wiking: OXPHOS: got time to instant chat on what sanuj is doing? | 16:26 |
@wiking | yesah | 16:26 |
@wiking | i'm about to pass out | 16:26 |
@wiking | but tell me | 16:26 |
lisitsyn | ok simple words | 16:26 |
OXPHOS | lisitsyn: sure | 16:27 |
@HeikoS | wiking: passing out | 16:27 |
lisitsyn | SGObject was base class with nothing | 16:27 |
lisitsyn | fields are added in inherited classes | 16:27 |
lisitsyn | what we're changing now is adding some capability | 16:27 |
lisitsyn | to store parameters of any type | 16:27 |
lisitsyn | basically it is | 16:27 |
lisitsyn | map<Tag, Any> | 16:27 |
lisitsyn | we would need to add serialize/deserialize things for that | 16:28 |
OXPHOS | lisitsyn: so just serealize Tag and any | 16:29 |
OXPHOS | wiking: did you mean change the name of cereal save/load to deserialize? not sure that can be done | 16:29 |
@wiking | why couldnt' it be done? | 16:30 |
@wiking | we could call it 'la' and 'lala' | 16:30 |
OXPHOS | wiking: overload the cereal header? | 16:30 |
@wiking | ? | 16:30 |
@wiking | just a sec | 16:30 |
@wiking | (meeting) | 16:30 |
lisitsyn | OXPHOS: yes we can obtain name and type from Tag | 16:31 |
lisitsyn | and name and type from Any | 16:31 |
OXPHOS | lisitsyn: cool | 16:32 |
OXPHOS | wiking: yeah one can change that in cereal/macros.hpp | 16:33 |
sanuj | HeikoS, you are back! | 16:33 |
@HeikoS | sanuj: yes, hi | 16:33 |
@HeikoS | OXPHOS: static calls will work now, I just commented on the kmeans PR | 16:33 |
OXPHOS | HeikoS: thx! | 16:34 |
sanuj | HeikoS, that means maths:init_rand(0) should work now | 16:34 |
sanuj | i'll rebase my gmm PR | 16:35 |
@HeikoS | sanuj: yes :) | 16:35 |
@HeikoS | lets hope these PRs can be merged now | 16:35 |
@HeikoS | we fixed two issues in the meta example system | 16:35 |
sanuj | that's good | 16:36 |
shogun-buildbot | build #2894 of bsd1 - libshogun is complete: Failure [failed configure] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2894 blamelist: Esben Sorig <esben@sorig.eu>, Heiko Strathmann <heiko.strathmann@gmail.com> | 16:38 |
@HeikoS | sanuj: let me know ok? then we can merge these guys asap | 16:40 |
c4goldsw | HeikoS: Hey, would you have time to look at a PR I've made? | 16:40 |
sanuj | HeikoS, remove examples/meta/lextab.py, examples/meta/parsetab.py? | 16:40 |
@HeikoS | c4goldsw: sure, will do soon. I was away for a few days and just going through emails | 16:40 |
c4goldsw | HeikoS: Great, thanks. | 16:40 |
@HeikoS | sanuj: yeah, everytime the meta example grammar is changed, we need to re-create these (happens automatically) | 16:40 |
@HeikoS | sanuj: I will soon fix this behaviour | 16:41 |
sanuj | okay, cool | 16:41 |
shogun-buildbot | build #32 of xenial - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/32 blamelist: Esben Sorig <esben@sorig.eu>, Heiko Strathmann <heiko.strathmann@gmail.com> | 16:41 |
sanuj | HeikoS, getting same error while generating meta examples | 16:58 |
sanuj | http://pastebin.com/prC4FDdx | 16:58 |
sanuj | do you think this might be related to python 3 | 16:58 |
sanuj | you remember i told you about most of my python modular tests failing | 16:58 |
@HeikoS | cd examples/meta/generator/ | 16:58 |
@HeikoS | rm lextab.py* parser.out parsetab.py* | 16:58 |
@HeikoS | this didnt work? | 16:59 |
sanuj | i think that was because of python3 | 16:59 |
@HeikoS | no | 16:59 |
@HeikoS | not python 3 | 16:59 |
sanuj | i removed lextab.py | 16:59 |
sanuj | okay let me try what you said | 16:59 |
@HeikoS | we have a build where it works with Python 3 | 16:59 |
@HeikoS | let me forward you an email | 16:59 |
sanuj | HeikoS, your command worked | 17:00 |
OXPHOS | lambday: hey any idea what may go wrong with global variable in benchmark? | 17:01 |
@HeikoS | sanuj: great! | 17:01 |
@HeikoS | wiking: around? | 17:03 |
-!- c4goldsw [c1a99ae1@gateway/web/cgi-irc/kiwiirc.com/ip.193.169.154.225] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 17:03 | |
Saurabh7 | hello | 17:06 |
Saurabh7 | HeikoS: hi | 17:06 |
@HeikoS | Saurabh7: hey | 17:07 |
@HeikoS | how are things? | 17:07 |
@HeikoS | shall we chat a bit in a moment? | 17:07 |
@HeikoS | just getting a coffee :) | 17:07 |
@HeikoS | Saurabh7: merged the xvalidation btw | 17:07 |
Saurabh7 | ok | 17:07 |
Saurabh7 | great! | 17:07 |
sanuj | HeikoS, will you be around after 45 mins? | 17:08 |
Saurabh7 | HeikoS: ahvent treid that eigen method ou mentioned | 17:08 |
@HeikoS | sanuj: yes | 17:08 |
sanuj | i need to have dinner | 17:08 |
@HeikoS | sanuj: sure, enjoy! | 17:08 |
@HeikoS | Saurabh7: yeah that would be good | 17:08 |
Saurabh7 | but we dont use the whole matrix | 17:08 |
Saurabh7 | X | 17:08 |
@HeikoS | as it really simplifies the code | 17:08 |
Saurabh7 | only a subset of it | 17:08 |
@HeikoS | mmh | 17:08 |
@HeikoS | let me see | 17:08 |
@HeikoS | https://github.com/jbowlan/pyl1ls/blob/master/lars.py#L104 | 17:08 |
@HeikoS | so reading this, they definitely use off the shelf updates | 17:09 |
@HeikoS | where you have a Cholsyky matrix, say R | 17:09 |
@HeikoS | and then you add x*x.T to the original matrix and want the new R | 17:09 |
@HeikoS | Saurabh7: the other thing is | 17:10 |
@HeikoS | Saurabh7: maybe we should not copy the matrix around all the time | 17:10 |
Saurabh7 | HeikoS: yes but in shougn right now its a bit different | 17:10 |
@HeikoS | I see | 17:10 |
@HeikoS | Saurabh7: ok, I am still reading things | 17:10 |
Saurabh7 | its onl checks columns which are in m_active_ | 17:10 |
Saurabh7 | so right now that choolesky_insert is hard coded for it | 17:11 |
@HeikoS | okok | 17:11 |
@HeikoS | cool then | 17:11 |
@HeikoS | Saurabh7: and what are you currently working on there? | 17:12 |
Saurabh7 | but in reality it should be possible ? | 17:12 |
@HeikoS | no it is ok for now | 17:12 |
@HeikoS | lets look at benchmark | 17:12 |
@HeikoS | and then decide | 17:12 |
Saurabh7 | HeikoS: almost done only 2 more solves to be replaced | 17:12 |
@HeikoS | is there an implementation that uses a similar solver? | 17:12 |
@HeikoS | Saurabh7: cool | 17:12 |
@HeikoS | Saurabh7: is the code guarded by unit test? | 17:12 |
Saurabh7 | i added more | 17:12 |
@HeikoS | that is, are you making sure you are not introducing bugs? | 17:12 |
Saurabh7 | for the cholesky insert | 17:12 |
@HeikoS | I think there were some tests before right? | 17:13 |
@HeikoS | for the full algo | 17:13 |
Saurabh7 | so thats definetly good | 17:13 |
Saurabh7 | HeikoS: yes there were | 17:13 |
@HeikoS | okok | 17:13 |
@HeikoS | so as long as they still pass all good | 17:13 |
@HeikoS | and benchmark? | 17:13 |
Saurabh7 | HeikoS: I ran benchamrk framwework | 17:13 |
Saurabh7 | its was bugged i send a fix | 17:13 |
@HeikoS | (havent read all email yet btw) | 17:13 |
@HeikoS | what was bugged? | 17:14 |
Saurabh7 | the way input was handled | 17:14 |
Saurabh7 | HeikoS: anyways the runtimes are very low even for big dataset | 17:14 |
Saurabh7 | I have values I wil psot there | 17:15 |
shogun-buildbot | build #2895 of bsd1 - libshogun is complete: Failure [failed configure] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2895 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com> | 17:15 |
@HeikoS | Saurabh7: what did you compare against? | 17:16 |
@HeikoS | yeah post | 17:16 |
Saurabh7 | scikit | 17:16 |
@HeikoS | Saurabh7: cool very curious about the times, give them :) | 17:16 |
@HeikoS | I will grab a coffee in the meantime | 17:16 |
Saurabh7 | okie | 17:16 |
shogun-buildbot | build #33 of xenial - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/33 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com> | 17:17 |
shogun-buildbot | build #23 of FC23 - libshogun - aarch64 is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/23 blamelist: Esben Sorig <esben@sorig.eu>, Heiko Strathmann <heiko.strathmann@gmail.com> | 17:19 |
@lambday | OXPHOS: what is going wrong with the global variable? | 17:39 |
@HeikoS | Saurabh7: back | 17:39 |
@HeikoS | so what about the times? | 17:39 |
@HeikoS | got them? | 17:39 |
@lambday | OXPHOS: it worked with the unit-tests, right? | 17:39 |
Saurabh7 | HeikoS: posted some | 17:40 |
@HeikoS | ah lemme check | 17:40 |
@HeikoS | which link is it? | 17:40 |
Saurabh7 | need to add scikit for one dataset too | 17:40 |
Saurabh7 | https://github.com/shogun-toolbox/shogun/pull/3243 | 17:40 |
OXPHOS | lambday: yes! | 17:43 |
sanuj | HeikoS, can you merge https://github.com/shogun-toolbox/shogun-data/pull/96 | 17:43 |
OXPHOS | lambday: if I do sg_linalg.set_cpu_backend(..), it gives segementation fault | 17:44 |
@lambday | OXPHOS: did you call init_shogun() ? | 17:44 |
@HeikoS | sanuj: done! | 17:44 |
sanuj | thanks | 17:44 |
OXPHOS | lambday: ...nooooooo | 17:44 |
@lambday | OXPHOS: there you go :) | 17:45 |
@lambday | OXPHOS: do one thing. In the constructor of the struct Data, call init_shogun_with_defaults().. and in the destructor, call exit_shogun() :) | 17:45 |
OXPHOS | lambday: sure. thanks! | 17:45 |
@lambday | OXPHOS: nw :) | 17:46 |
@HeikoS | Saurabh7: these look good | 17:46 |
@HeikoS | are there also some for which it takes longer than 10s | 17:46 |
@HeikoS | Saurabh7: another question: does our LARS allow to extract the regularisation path easily? | 17:46 |
@HeikoS | Saurabh7: and what are the next steps? | 17:47 |
Saurabh7 | HeikoS: mayb i can find one for which bigger runtimes, but htey migh tnot be meaningful dataset | 17:48 |
Saurabh7 | HeikoS: yes i think so | 17:48 |
@HeikoS | no please use a meaningful one :) | 17:48 |
@HeikoS | Saurabh7: you can also just generate data yourself | 17:48 |
@HeikoS | which is LASSO friendly | 17:48 |
@wiking | OXPHOS: so basically with the tags any you could store all the params right? | 17:48 |
@HeikoS | that is: a linear regression function with many dimensions | 17:48 |
@wiking | lisitsyn: ^ | 17:48 |
@HeikoS | but with only a few that contribute to the ys | 17:48 |
Saurabh7 | HeikoS: ok lets comapre on sythetic big one then | 17:49 |
OXPHOS | wiking: yes i think so | 17:49 |
@HeikoS | Saurabh7: do you see how I mean this? | 17:49 |
@HeikoS | Saurabh7: maybe prepare a notebook that generates such data and share, then I can comment | 17:49 |
@wiking | OXPHOS: ok so that'd be enough to have the serialization of that map | 17:49 |
sanuj | HeikoS, i forgot to do Math:init_random(1) ==> Math:init_random(0) | 17:49 |
sanuj | i hope that is fine | 17:49 |
Saurabh7 | HeikoS: only few variates should be included in final ? | 17:49 |
@HeikoS | sanuj: dont worry aobut it | 17:50 |
@HeikoS | Saurabh7: so the model is | 17:50 |
@HeikoS | y=w*x | 17:50 |
@wiking | looooooooooooool | 17:50 |
* wiking reacts on the runtime of lars | 17:50 | |
@HeikoS | wiking: why? :) | 17:50 |
@wiking | HeikoS: the diff | 17:51 |
@HeikoS | Saurabh7: and x and w are high dimensional | 17:51 |
@HeikoS | but w has mostly zeros | 17:51 |
@wiking | HeikoS: btw do we wanna do the cholesky from eigen, not linalg? | 17:51 |
@HeikoS | Saurabh7: so you generate x from say a gaussian | 17:51 |
@HeikoS | pick w to have only few non-zero components | 17:51 |
@HeikoS | and then generate y | 17:51 |
@HeikoS | wiking: its good right? | 17:51 |
@HeikoS | wiking: but for these small problems, the difference is not too meaningful | 17:52 |
@HeikoS | wiking: I am curious to see this for problems that take like a minute (or more) for shogun | 17:52 |
@HeikoS | wiking: we want to use linalg | 17:52 |
@HeikoS | but as it is not ready | 17:52 |
@HeikoS | we can use eigen for now | 17:52 |
@wiking | what do you mean not ready? | 17:52 |
@wiking | i mean we could add to the current linalg no? | 17:52 |
@HeikoS | wiking: yeah | 17:52 |
@HeikoS | I think the factorisation is using linalg | 17:52 |
@wiking | i mean the function call will not change | 17:52 |
@HeikoS | true | 17:53 |
@wiking | just how it is implemented in the backend | 17:53 |
@HeikoS | Saurabh7: ^ | 17:53 |
@wiking | which is fine | 17:53 |
@HeikoS | yeah I agree | 17:53 |
@wiking | but we dont wanna rewrite algos like lars | 17:53 |
@HeikoS | Saurabh7: so we need a triangular solve in linalg | 17:53 |
@HeikoS | wiking: doing this atm :) | 17:53 |
@HeikoS | since it was all hand-implemented | 17:53 |
@HeikoS | all the dot products | 17:53 |
@HeikoS | using lapack and loops | 17:53 |
@wiking | we shoudl rather implement a cholesky as part of linalg | 17:53 |
@wiking | hehehe | 17:53 |
@HeikoS | wiking: it is there an dused | 17:54 |
@wiking | i mean we could move those | 17:54 |
@HeikoS | but not the triangular solve | 17:54 |
@wiking | to linalg :P | 17:54 |
@wiking | but i guess nobody wants to support that | 17:54 |
@HeikoS | nop :) | 17:54 |
@HeikoS | there actually is some hybrid cholesky algorithms | 17:54 |
@HeikoS | that use GPU and CPU in a combined manner | 17:54 |
Saurabh7 | HeikoS: I am using eigen solve, jsut need to extract specific cols from X | 17:54 |
@HeikoS | Saurabh7: you are using a triangular solve in eigen, no? | 17:55 |
Saurabh7 | yesyes | 17:55 |
@HeikoS | so that should be in linalg | 17:55 |
Saurabh7 | ok linalg | 17:55 |
Saurabh7 | isee | 17:55 |
@HeikoS | but lets do it in a second step | 17:55 |
@HeikoS | first finish all the rest | 17:55 |
@HeikoS | as changing is easier then | 17:55 |
@HeikoS | Saurabh7: any chance that multicore helps in the LARS implementation you recon? | 17:56 |
@HeikoS | probably not right? | 17:56 |
Saurabh7 | was thinking about it | 17:56 |
Saurabh7 | not anythign obvious tho | 17:56 |
Saurabh7 | maybe some of the dot products | 17:57 |
@HeikoS | Saurabh7: so for the benchmark | 18:00 |
@HeikoS | can you try this for multiple D, N, and #num_non_zero | 18:00 |
@HeikoS | and post numbers for each? | 18:00 |
@HeikoS | identifying where shogun wins more and where less? | 18:00 |
@HeikoS | and then I will tweet "shogun 10x faster LARS than sklearn with 10x less developers" :D | 18:01 |
@HeikoS | wiking: ^ | 18:01 |
Saurabh7 | HeikoS: sure :) | 18:02 |
Saurabh7 | gotta run for dinner now , might be back a bit later | 18:02 |
Saurabh7 | HeikoS: should be up by tmrw morn | 18:02 |
@HeikoS | Saurabh7: cool, will catch up then | 18:03 |
shogun-buildbot | build #24 of FC23 - libshogun - aarch64 is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/24 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com> | 18:05 |
@wiking | just a sec | 18:07 |
@wiking | :>>>> | 18:07 |
@wiking | we need to start autogen the benchmarks | 18:07 |
@HeikoS | yep definitely | 18:08 |
@HeikoS | to support such claims | 18:08 |
@HeikoS | that is also something nice to show to the amazon people | 18:08 |
@HeikoS | since they employ a few sklearn devs ;) | 18:08 |
@HeikoS | wiking: any eta for that? | 18:08 |
@wiking | hahaha | 18:09 |
lisitsyn | shogun has yandex devs! | 18:09 |
@wiking | lisitsyn: devs? | 18:09 |
@wiking | you mean dev | 18:09 |
lisitsyn | DEV | 18:09 |
@wiking | :D | 18:09 |
lisitsyn | :D | 18:09 |
@HeikoS | it is THE DEV | 18:09 |
@HeikoS | (tm) | 18:09 |
@wiking | i'm gonna look into it | 18:09 |
@wiking | and start running benchmarks | 18:10 |
@wiking | now that we move off everything | 18:10 |
@HeikoS | wiking: yeah the bot should be idling these days | 18:10 |
@wiking | from the servers | 18:10 |
@wiking | not 100% | 18:10 |
@HeikoS | the aws seems to work quite smooth | 18:10 |
@wiking | the first tests are still on the boths | 18:10 |
@wiking | HeikoS: had some hiccups | 18:10 |
@wiking | but yeah | 18:10 |
@HeikoS | wiking: oh one more thing I wanted to ask about | 18:10 |
@wiking | i'll move the devian tests as well | 18:10 |
@HeikoS | release and gpl | 18:10 |
@wiking | *debian | 18:10 |
@wiking | yes | 18:10 |
@HeikoS | I created a repo | 18:10 |
@wiking | saw it | 18:10 |
@HeikoS | but before I move things | 18:11 |
@HeikoS | should discuss briefly | 18:11 |
@HeikoS | how to modify the build etc | 18:11 |
@HeikoS | and dir structure | 18:11 |
@HeikoS | etc | 18:11 |
@wiking | well | 18:11 |
@wiking | i reckon we should just have an extra dir under src/shogun | 18:12 |
@wiking | that is a submodule | 18:12 |
@HeikoS | yeah | 18:12 |
@HeikoS | but is it flat? | 18:12 |
@HeikoS | or does it mirror the dir structure | 18:12 |
@wiking | mmm doesn't matter | 18:12 |
@wiking | could be anything | 18:12 |
@HeikoS | all files in there plain is a a bit ugly | 18:12 |
@wiking | i guess it should be a bit hierarch | 18:12 |
@wiking | but maybe per project | 18:13 |
@wiking | and not per type | 18:13 |
@wiking | so not like | 18:13 |
@wiking | 'classifier' | 18:13 |
@wiking | but something like: 'implementation_x' | 18:13 |
@HeikoS | I see | 18:13 |
@HeikoS | yeah maybe | 18:13 |
@wiking | but htat's just an idea | 18:13 |
@HeikoS | and we can just incrementally move thing | 18:14 |
@HeikoS | 1.) move file to gpl repo 2.) delete from main repo and chnage paths 3.) check build | 18:14 |
@HeikoS | 4.) repeat | 18:14 |
@HeikoS | wiking: we will keep the switch in cmake I guess | 18:15 |
@wiking | yes | 18:15 |
sanuj | HeikoS, i tried --> all_layers = all_layers.linear(50); all_layers = all_layers.softmax(2); all_layers = all_layers.done() | 18:19 |
sanuj | the translation fails | 18:19 |
@HeikoS | sanuj: whats the error? | 18:19 |
@HeikoS | sanuj: does that work in Python? | 18:19 |
lisitsyn | lambday: hey | 18:20 |
@HeikoS | translation itself? | 18:20 |
@HeikoS | sanuj: can you share the code and details, then I can check it locally | 18:20 |
lisitsyn | lambday: what's the matter with spaces :D | 18:20 |
lisitsyn | ain't it super ugly to have | 18:20 |
lisitsyn | x=y | 18:20 |
sanuj | HeikoS, there is no error, the cpp file generated is incomplete because of problems in translation | 18:20 |
sanuj | i'll send you the code | 18:20 |
@HeikoS | sanuj: I see | 18:20 |
@HeikoS | yeah lemme check | 18:20 |
sanuj | it works in python | 18:20 |
@HeikoS | sanuj: sanity check | 18:21 |
@HeikoS | meta examples dont use semicolons | 18:21 |
@HeikoS | put every statement in nits own line | 18:21 |
sanuj | HeikoS, i think it is because these functions return a pointer to NeuralLayers | 18:21 |
sanuj | yeah, i didn't put semi colons | 18:22 |
@HeikoS | translation doesnt care about semantics | 18:22 |
@HeikoS | so it cannot fail due to semantics | 18:22 |
sanuj | HeikoS, i had tried get_layer_parameters also | 18:22 |
sanuj | it returned SGVector<>* | 18:22 |
@HeikoS | sanuj: send it | 18:22 |
sanuj | and failed | 18:22 |
@HeikoS | sanuj: that is not SWIG compativlew | 18:22 |
@HeikoS | compatible | 18:22 |
sanuj | yeah, i'll send | 18:23 |
@HeikoS | maybe start with a python example and then translate that to .sg | 18:23 |
sanuj | okay | 18:23 |
sanuj | HeikoS, btw, when i run "make test" or "ctest", most of the python-modular tests fail http://pastebin.com/6RZdijXf | 18:24 |
sanuj | we had discussed this earlier | 18:24 |
sanuj | my PYTHONPATH = /home/sanuj/Projects/shogun_cb/buildpython/src/interfaces/python_modular | 18:24 |
sanuj | and LD_LIBRARY_PATH = /home/sanuj/Projects/shogun_cb/buildpython/src/shogun | 18:24 |
sanuj | i'm sure this is because of python3 | 18:27 |
sanuj | "python distance_tanimoto_modular.py" but "python3 distance_tanimoto_modular.py | 18:27 |
sanuj | " doesn't | 18:27 |
@HeikoS | sanuj: what happens if you run any of the examples manually? | 18:27 |
@HeikoS | with what error? | 18:27 |
sanuj | HeikoS, ImportError: dynamic module does not define init function (PyInit__modshogun) | 18:28 |
OXPHOS | HeikoS: I'm current doing in conf.py: 'cookbook_classifier_linearsvm': ('doc/cookbook/html/examples/classifier/linear_svm.html', ''), | 18:28 |
OXPHOS | HeikoS: too specific? | 18:28 |
@HeikoS | sanuj: that is your python setup I am quite sure | 18:29 |
@HeikoS | sanuj: not sure how to fix it | 18:29 |
@HeikoS | did you ask google? | 18:29 |
@HeikoS | wiking might know as well ^ | 18:29 |
@HeikoS | OXPHOS: sorry, what? : | 18:29 |
@HeikoS | :) | 18:29 |
OXPHOS | HeikoS: u want a hyperlink from "see the linear SVM cookbook" right? | 18:30 |
OXPHOS | the only way I can think of is 'cookbook_classifier_linearsvm': ('doc/cookbook/html/examples/classifier/linear_svm.html', ''), | 18:30 |
OXPHOS | and in .sg :cookbook_classifier_linearsvm:`see the linear SVM cookbook` | 18:30 |
OXPHOS | .rst | 18:31 |
@HeikoS | yeah hyperlink would be good | 18:31 |
sanuj | wiking, ping pong | 18:31 |
@HeikoS | OXPHOS: let me check, just a sec | 18:32 |
@HeikoS | OXPHOS: whats the link again? | 18:32 |
sanuj | HeikoS, but why is "make test" using python3 | 18:32 |
@wiking | pong pong | 18:32 |
OXPHOS | HeikoS: https://github.com/shogun-toolbox/shogun/pull/3242 ? | 18:32 |
sanuj | wiking, on running python modular examples with python3 i get "ImportError: dynamic module does not define init function (PyInit__modshogun)" | 18:32 |
OXPHOS | HeikoS I don't want to mislead sanuj since you mentioned.. | 18:32 |
@wiking | sanuj: what's the problem with python3 | 18:33 |
sanuj | wiking, most of the python-modular examples fail when i do "make test", i think it is because of python3 | 18:33 |
@wiking | works here http://buildbot.shogun-toolbox.org/builders/deb4%20-%20python3/builds/862 | 18:33 |
@HeikoS | OXPHOS: ah now I see | 18:34 |
sanuj | why am i getting that error | 18:34 |
@HeikoS | OXPHOS: so if you "make cookbook" this, what does the link point to? | 18:34 |
@wiking | sanuj: good question | 18:34 |
@HeikoS | sanuj: I suspect it is you python setup, and that something is messed up between python2/3 | 18:35 |
@HeikoS | your shogun is not only compiled against one version or so | 18:35 |
sanuj | wiking, and why is "make test" using python3 and not python 2.7 | 18:35 |
OXPHOS | HeikoS: actually is is "http://localhost:8000/doc/cookbook/html/examples/classifier/doc/cookbook/html/examples/classifier/linear_svm.htmlSee%20the%20linear%20SVM%20cookbook" | 18:36 |
@HeikoS | sanuj: make test uses the python interpreter that swig finds | 18:36 |
@wiking | sanuj: i need to see your cmake output | 18:36 |
@HeikoS | OXPHOS: this is a bit more problematic, I just reliased, so the "latest" can stay for now | 18:36 |
@wiking | |plz email | 18:36 |
@HeikoS | OXPHOS: we need to change that later | 18:36 |
sanuj | HeikoS, my cmakecache.txt has PYTHON_PACKAGES_PATH lib/python2.7/dist-packages | 18:36 |
@wiking | as im about to sleep | 18:36 |
@HeikoS | OXPHOS: so ignore my comment | 18:36 |
sanuj | wiking, okay :) | 18:36 |
OXPHOS | HeikoS: okay..actually I thought it would be easy ;) thanks | 18:37 |
@HeikoS | OXPHOS: we will need to have some notion of the current shogun version in there and that put that in the link | 18:38 |
@HeikoS | but we dont have that yet | 18:38 |
@HeikoS | will work on that after gsoc | 18:38 |
@HeikoS | since low priority | 18:38 |
@HeikoS | OXPHOS: did i comment on anything else on the linear machine? | 18:38 |
@lambday | lisitsyn: I totally agree. I myself never use x=y :D but in Shogun we don't use that : https://github.com/shogun-toolbox/shogun/wiki/Code-style | 18:39 |
lisitsyn | lambday: but we don't really follow that | 18:39 |
OXPHOS | HeikoS: yes. some phrases and the creation of the instance | 18:39 |
lisitsyn | sanuj: could you please ignore lambday :D :D :D | 18:39 |
@lambday | lisitsyn: then it shouldn't be in coding styles :D | 18:39 |
@lambday | lisitsyn: I try to follow that when pushing code to shogun anyway :D | 18:39 |
lisitsyn | lets change it! | 18:40 |
lisitsyn | x=y | 18:40 |
lisitsyn | wtf | 18:40 |
OXPHOS | HeikoS: you mean we keep full address like http://. How about the rename stuff? linear_svm --> see the linear SVM cookbook | 18:40 |
@lambday | lisitsyn: fine with me. | 18:41 |
@HeikoS | x=y? | 18:41 |
@HeikoS | what? | 18:41 |
@lambday | no spaces between operators and operands | 18:41 |
lisitsyn | HeikoS: lambday wants no spaces around = | 18:41 |
lisitsyn | :) | 18:41 |
@HeikoS | lisitsyn: thats what we have written down | 18:41 |
@HeikoS | lisitsyn: should change it then | 18:41 |
@HeikoS | OXPHOS: yeah keep full address to the "latest" | 18:41 |
@HeikoS | rename works just like in html's href (I think) | 18:42 |
sanuj | lisitsyn, lambday rofl | 18:44 |
OXPHOS | HeikoS: I think you didnt get me about kmeans..now the dataset and kmeans are both updated. but the kmeans integration dataset need to be merged first | 18:48 |
@HeikoS | I see | 18:48 |
@HeikoS | so I should merge? | 18:49 |
@HeikoS | sorry too many emails ;) | 18:49 |
OXPHOS | HeikoS: yep thanks | 18:49 |
OXPHOS | HeikoS: I can tell that :) | 18:49 |
@HeikoS | kmeans merged | 18:49 |
OXPHOS | HeikoS: thanks! | 18:49 |
@HeikoS | OXPHOS: often it helps if you just tell me what to do "HeikoS merge this [link]" ;) | 18:49 |
@HeikoS | OXPHOS: your other data pr doesnt have any output | 18:50 |
@HeikoS | need to extract something | 18:50 |
OXPHOS | HeikoS: sure;) I'll look into that. Yes I just generated some lables | 18:50 |
@HeikoS | cool | 18:51 |
@HeikoS | anyone here knows whether JULIA supports SWIG these days? | 18:53 |
sanuj | HeikoS, gmm is failing https://travis-ci.org/shogun-toolbox/shogun/jobs/135628446#L2632 | 19:00 |
sanuj | for ruby | 19:00 |
sanuj | can't say about java and lua, the build terminated | 19:00 |
@HeikoS | checking | 19:06 |
@HeikoS | ah static call fun | 19:06 |
@HeikoS | checking | 19:06 |
@HeikoS | sanuj: I think ruby is easy to fix | 19:08 |
@HeikoS | will send a patch | 19:08 |
sanuj | HeikoS, okay thanks | 19:08 |
@HeikoS | restarting lua | 19:08 |
@HeikoS | can you ping me on that? | 19:08 |
sanuj | HeikoS, i'm going to sleep | 19:09 |
@HeikoS | sleep well | 19:09 |
sanuj | HeikoS, can you review my tags PR | 19:09 |
@HeikoS | sanuj: I can, what to focus on? | 19:09 |
@HeikoS | sanuj: is this for merging already? | 19:09 |
sanuj | it's ready to be merged apart from some comment by lambday | 19:09 |
@HeikoS | okok | 19:09 |
sanuj | HeikoS, i'll look at it tomorrow | 19:09 |
sanuj | https://github.com/shogun-toolbox/shogun/pull/3221 | 19:10 |
@HeikoS | cool, see you! | 19:10 |
sanuj | comment in the PR | 19:10 |
-!- sanuj [~sanuj@117.203.18.75] has quit [Ping timeout: 276 seconds] | 19:15 | |
-!- GSqU [~GSqU@104.156.238.186] has joined #shogun | 19:39 | |
-!- GSqU [~GSqU@104.156.238.186] has left #shogun [] | 19:40 | |
-!- lambday [8028b10a@gateway/web/freenode/ip.128.40.177.10] has quit [] | 19:54 | |
OXPHOS | HeikoS: Hey, I didn't find anything to extract for multi-class logistic regression and multi-class linear machine | 19:55 |
OXPHOS | via the multi-class classifier directly | 19:56 |
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 260 seconds] | 20:25 | |
-!- besser82 [~besser82@fedora/besser82] has joined #shogun | 20:33 | |
-!- mode/#shogun [+o besser82] by ChanServ | 20:33 | |
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 246 seconds] | 20:38 | |
-!- besser82 [~besser82@fedora/besser82] has joined #shogun | 20:40 | |
-!- mode/#shogun [+o besser82] by ChanServ | 20:40 | |
@HeikoS | rcurtin: jo | 20:43 |
@HeikoS | OXPHOS: e.g. the labels | 20:43 |
@HeikoS | OXPHOS: you can extract the binary labels for each class, or the multiclass label vector | 20:44 |
rcurtin | HeikoS: hello there | 20:46 |
rcurtin | in a meeting, might be slightly slow to respond | 20:46 |
@HeikoS | rcurtin: hi man, how are things? | 20:46 |
rcurtin | GSoC makes me busy, I always forget just how much work it is :) | 20:47 |
@HeikoS | rcurtin: haha same here | 20:48 |
@HeikoS | rcurtin: there is a PR for the benchmark thing on lars, just wanted to ask to merge as it fixes a bug | 20:48 |
rcurtin | sure, is it ready to go? you just made a comment on it pointing out that it might not be the right thing | 20:48 |
@HeikoS | lemme check | 20:50 |
@HeikoS | rcurtin: dont see your comment | 20:50 |
rcurtin | I did not make a comment, you did :) | 20:51 |
rcurtin | "Not sure I get this. Definitely, it only makes sense to compare against [this one](http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LassoLars.html) " | 20:51 |
@HeikoS | ah sorry | 20:52 |
@HeikoS | Saurabh7: around? | 20:54 |
@HeikoS | rcurtin: no merge pls :) | 20:57 |
@HeikoS | rcurtin: commented | 20:57 |
@HeikoS | rcurtin: how is gsoc going for you guys? | 20:57 |
rcurtin | HeikoS: it is going well, lots of code starting to come in now | 20:58 |
@HeikoS | rcurtin: good code? ;) | 20:58 |
rcurtin | yeah, it is great so far, but it takes a long time to check through it all and make sure it is good | 20:58 |
rcurtin | I think the first few PRs are especially hard because you have to check style and all kinds of other little things before you trust the students fully and check their code in less detail | 20:59 |
rcurtin | it looks like this channel is quite busy, I would guess you guys are having a similar experience :) | 21:00 |
@HeikoS | rcurtin: yeah indeed | 21:00 |
@HeikoS | rcurtin: yeah very busy | 21:00 |
@HeikoS | we are going for tighter mentoring this year | 21:00 |
@HeikoS | as the projects are quite invovled | 21:00 |
@HeikoS | not just extensions | 21:00 |
@HeikoS | rcurtin: you are working on a csv parser I see ?:) | 21:01 |
-!- travis-ci [~travis-ci@ec2-54-205-226-52.compute-1.amazonaws.com] has joined #shogun | 21:03 | |
travis-ci | it's lambday's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/135647014 | 21:03 |
-!- travis-ci [~travis-ci@ec2-54-205-226-52.compute-1.amazonaws.com] has left #shogun [] | 21:03 | |
rcurtin | HeikoS: yeah, tighter mentoring definitely pays off | 21:04 |
rcurtin | after some of our 2014 projects, it was like mid-2015 until I realized some fairly obvious issues with the usability of the code that was produced | 21:04 |
@HeikoS | yeah | 21:04 |
@HeikoS | same | 21:04 |
@HeikoS | and you also want people to stay around afterwards | 21:05 |
@HeikoS | more likely if they are integrated more | 21:05 |
rcurtin | we are indeed looking into writing a parser ourselves, because Armadillo's loading functionality doesn't really handle missing values and other things that are important in ML | 21:05 |
@HeikoS | and more fun for everyone | 21:05 |
rcurtin | I agree, definitely | 21:05 |
rcurtin | boost::spirit seems really cool, but quite crazy | 21:05 |
rcurtin | I don't like maintaining a parser ourselves, since it's just more code to take care off that can have bugs that will be our problem (instead of some other project's), but I am not sure we have a choice in this case | 21:05 |
@HeikoS | rcurtin: you fine with boost as a dependency? | 21:06 |
@HeikoS | we're avoiding that here | 21:06 |
@HeikoS | with some pretty strong opinions on that | 21:07 |
@HeikoS | but our csv loader is buggy | 21:07 |
@HeikoS | ;) | 21:07 |
rcurtin | HeikoS: yeah, I think the primary objection to boost is how huge it is, right? | 21:07 |
@HeikoS | yes | 21:07 |
@HeikoS | creates issues for portability | 21:07 |
@HeikoS | it is a nightmare already for us | 21:07 |
rcurtin | portability to what? | 21:07 |
@HeikoS | keeping all the builds green | 21:07 |
@HeikoS | different OS | 21:08 |
rcurtin | we have no issues with windows/os x/linux/freebsd | 21:08 |
@HeikoS | with different boost versions etc | 21:08 |
rcurtin | anything embedded, probably you might have issues | 21:08 |
rcurtin | ah, yeah, we avoid that by being careful with the minimum accepted boost version and testing across boost versions | 21:08 |
rcurtin | although, we don't really have a formal framework in place for that | 21:08 |
@HeikoS | hehe | 21:08 |
@HeikoS | btw have you seen our cookbooks? | 21:08 |
rcurtin | I am *still* waiting for my company to mount our new jenkins server that they bought for us (I have been waiting since October) | 21:08 |
rcurtin | I have glanced at some in the past, but not in depth | 21:08 |
@HeikoS | we are making every student write two a week atm | 21:08 |
@HeikoS | check shogun.ml/cookbook/latest | 21:08 |
rcurtin | these are like little tutorials? | 21:09 |
@HeikoS | no | 21:09 |
@HeikoS | more like: API examples, integration test, and code snippets at once | 21:09 |
@HeikoS | rcurtin: hey that is cool | 21:09 |
rcurtin | ah, yeah, that's right, they are also integration tests | 21:09 |
rcurtin | that is very nice, most people just want snippets to copy/paste and try | 21:09 |
@HeikoS | yeah all executable code (in all languages) | 21:09 |
@HeikoS | so travis runs all of them | 21:09 |
@HeikoS | but the pages only show a subset of the full listing | 21:09 |
@HeikoS | we recently got some amazon credits, and moved our buildbot there | 21:10 |
@HeikoS | now we have more space on our own servers for benchmarks | 21:10 |
rcurtin | did they give them to you for free because you are an open source project? | 21:10 |
@HeikoS | so will probably run your framework there | 21:10 |
rcurtin | I seem to remember that they offer credits for stuff like that | 21:10 |
@HeikoS | rcurtin: kind of, we applied | 21:10 |
@HeikoS | and knew some people | 21:10 |
@HeikoS | yeah | 21:10 |
@HeikoS | pretty cool | 21:10 |
rcurtin | knowing some people is always the best way :) | 21:10 |
@HeikoS | we are not nearly using all of them | 21:10 |
@HeikoS | yep ;) | 21:11 |
@HeikoS | the old shogunners | 21:11 |
@HeikoS | they all did their phds together hehe | 21:11 |
@HeikoS | I am way too young | 21:11 |
@HeikoS | the others even more | 21:11 |
rcurtin | where did soeren end up? is he still at tomtom, or did he go to amazon? | 21:11 |
rcurtin | he was the only old shogunner I ever met :) | 21:11 |
rcurtin | is it shogunner or shoguneer? :) | 21:12 |
shogun-buildbot | build #2896 of bsd1 - libshogun is complete: Failure [failed configure] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2896 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:12 |
@HeikoS | tomtom | 21:12 |
@HeikoS | there is gunnar | 21:12 |
@HeikoS | he supervised sören | 21:12 |
rcurtin | ah, okay, I did not realize he went to amazon | 21:13 |
shogun-buildbot | build #34 of xenial - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/34 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:14 |
@HeikoS | no he didnt | 21:14 |
rcurtin | oh, okay, my comprehension level is low I guess :) | 21:14 |
@HeikoS | but both of them know the head of german amazon research | 21:14 |
@HeikoS | yeah sorry I am too vague | 21:14 |
@HeikoS | aws is really unreliable | 21:14 |
rcurtin | :( | 21:15 |
@HeikoS | I think viktor told you already | 21:15 |
@HeikoS | but it is ok for buildbotting | 21:15 |
rcurtin | do you find the builds are randomly failing or something? | 21:15 |
@HeikoS | I think they were | 21:15 |
@HeikoS | and stalling | 21:16 |
rcurtin | I remember we had a discussion about whether or not AWS was suitable for benchmarking, I think he disagreed with my opinion, but that is okay, my opinion is often wrong :) | 21:16 |
@HeikoS | disappearing | 21:16 |
rcurtin | that's really frustrating, but we have had the same problem with Jenkins sometimes | 21:16 |
@HeikoS | I think he agrees with you | 21:16 |
@HeikoS | not suitable for benchmarking | 21:16 |
rcurtin | I can't remember, too much has happened since then :) | 21:16 |
@HeikoS | at that thats whatI understood | 21:16 |
@HeikoS | hehe | 21:16 |
rcurtin | our whole build setup is stalled on Symantec getting this damn server mounted and publicly accessible | 21:16 |
@HeikoS | I see | 21:16 |
@HeikoS | hard ? | 21:16 |
rcurtin | but it's just, there is so much paperwork to fill out, and the IT people are so slow, that we have been waiting for months | 21:17 |
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 250 seconds] | 21:17 | |
rcurtin | that's the thing, it shouldn't be difficult! we just want two ports open (HTTP and SSH) but they have a giant list of requirements, the system has to be scanned, we have to submit forms for a firewall change request | 21:17 |
rcurtin | but they are not running the scan for some reason, and the request is not going through as a result | 21:17 |
rcurtin | it's really stupid, too much process overhead at this company | 21:17 |
rcurtin | fortunately I do not have to deal with the IT people directly, I have someone helping me, but even so, it is quite frustrating | 21:17 |
@HeikoS | phew I can imagine this is nerve wrecking | 21:19 |
rcurtin | it's irritating, because a lot of stuff can't proceed until that is done | 21:21 |
rcurtin | I have enough to do in the mean time, but it's gone from being a monthly "I wonder when they will have that done" to a weekly "I wonder what the progress is" to a daily "why the hell is it not up yet" | 21:21 |
@HeikoS | hehe | 21:22 |
@HeikoS | rcurtin: in that sense open source moves faster than companies | 21:22 |
@HeikoS | at least if somebody who is motivated is around | 21:23 |
rcurtin | truth | 21:23 |
@HeikoS | rcurtin: well good luck with the rest of the first half of gsoc | 21:27 |
@HeikoS | nice that the benchmarking is now a bit of a joint venture! | 21:27 |
rcurtin | thanks, I am looking forward to what we come out with :) | 21:29 |
rcurtin | both for the first half of GSoC and the benchmarking system :) | 21:30 |
shogun-buildbot | build #864 of deb4 - python3 is complete: Failure [failed install test python modular] Build details are at http://buildbot.shogun-toolbox.org/builders/deb4%20-%20python3/builds/864 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:47 |
shogun-buildbot | build #2897 of bsd1 - libshogun is complete: Failure [failed configure] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2897 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:49 |
shogun-buildbot | build #35 of xenial - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/35 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:52 |
-!- besser82 [~besser82@fedora/besser82] has joined #shogun | 21:53 | |
-!- mode/#shogun [+o besser82] by ChanServ | 21:53 | |
shogun-buildbot | build #25 of FC23 - libshogun - aarch64 is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/25 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 21:53 |
@HeikoS | shogun-buildbot: force build 'deb4 - python3' | 21:54 |
@HeikoS | force build --branch=develop 'deb4 - python3' | 21:54 |
@HeikoS | shogun-buildbot: force build --branch=develop 'deb4 - python3' | 21:54 |
shogun-buildbot | The build has been queued, I'll give a shout when it starts | 21:54 |
-!- HeikoS [~heiko@nat-182-210.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 21:56 | |
-!- lambday [6d9941c0@gateway/web/freenode/ip.109.153.65.192] has joined #shogun | 21:59 | |
-!- mode/#shogun [+o lambday] by ChanServ | 21:59 | |
@lambday | OXPHOS: hola | 22:00 |
@lambday | OXPHOS: is the global sg_linalg working now? | 22:00 |
@lambday | in the benchmark? | 22:00 |
@lambday | OXPHOS: also, did you get OpenCL to work locally? | 22:00 |
OXPHOS | lambday: should be..Im about to run it | 22:01 |
OXPHOS | lambday: it works, just didn't work with -l | 22:01 |
@lambday | OXPHOS: can you check your viennacl cmake output? it should be able to detect whether OpenCL is working or not | 22:01 |
@lambday | even Shogun's cmake should be able to do that | 22:02 |
OXPHOS | Is it in CMakeOutput.log? | 22:04 |
OXPHOS | lambday ^ | 22:04 |
OXPHOS | apparently it didn't recognize init_shogun_with_defaults(); | 22:05 |
OXPHOS | lemme see | 22:05 |
OXPHOS | works now. #include | 22:07 |
@lambday | OXPHOS: yeah.. init.h | 22:07 |
@lambday | OXPHOS: also, about the benchmark, me and Heiko discussed a bit today.. | 22:08 |
@lambday | OXPHOS: so basically what would be interesting to see is that, how much we pay for the flexibility | 22:08 |
@lambday | OXPHOS: please check if you can see this -http://cosketch.com/Rooms/kymwluv | 22:08 |
OXPHOS | yes lol | 22:09 |
@lambday | we expect a plot like this | 22:09 |
-!- sonne|osx [~sonne@x4db3bf36.dyn.telefonica.de] has joined #shogun | 22:09 | |
@lambday | pardon my drawing skills :D | 22:09 |
OXPHOS | it's cute :) | 22:09 |
@lambday | haha | 22:09 |
@lambday | so notice the diagram.. the overhead for the flexibility should go down as we increase the dimension of the vector | 22:10 |
@lambday | so, in practice, as long as the vector has a sufficiently large dimension, we don't pay anything extra.. | 22:11 |
@lambday | we want to know what that dimension is | 22:11 |
@lambday | see the point? | 22:11 |
@lambday | haha doodling :D | 22:11 |
lisitsyn | how's my drawing | 22:11 |
@lambday | better than mine :D | 22:11 |
@lambday | that was supposed to be a ;) | 22:12 |
OXPHOS | lambday: a. SGVector -map to Eigen Vector - call Eigen dot ? | 22:12 |
@lambday | OXPHOS: yeah.. | 22:13 |
OXPHOS | but this is what we're doing in CPUBackend.dot now.. | 22:13 |
@lambday | so you should include the creation of map in the benchmark as well.. | 22:13 |
@lambday | OXPHOS: of course.. that's the point | 22:13 |
lisitsyn | the pigder! | 22:14 |
lisitsyn | spipig! | 22:14 |
@lambday | lisitsyn: that looks like captain kirk | 22:14 |
@lambday | piggified version | 22:14 |
@lambday | lisitsyn: hahaha.. | 22:15 |
OXPHOS | haha | 22:15 |
lisitsyn | DONE | 22:16 |
@lambday | lisitsyn: I did something too xD | 22:17 |
@lambday | lisitsyn: hahaha | 22:17 |
@lambday | OXPHOS: so, do you copy? over. | 22:19 |
OXPHOS | lambday: roger that | 22:19 |
shogun-buildbot | build forced [ETA 5m47s] | 22:20 |
shogun-buildbot | I'll give a shout when the build finishes | 22:20 |
@lambday | shogun-buildbot: don't shout, just dance | 22:20 |
@lambday | shogun-buildbot: dance | 22:20 |
shogun-buildbot | <(^.^<) | 22:20 |
shogun-buildbot | <(^.^)> | 22:20 |
shogun-buildbot | (>^.^)> | 22:20 |
shogun-buildbot | (7^.^)7 | 22:20 |
shogun-buildbot | (>^.^<) | 22:20 |
@lambday | lisitsyn: so about the SGObject.. the subclasses register params, right? | 22:21 |
@lambday | and then the serialization works with the hashmap from the base class? | 22:21 |
shogun-buildbot | build #865 of deb4 - python3 is complete: Success [build successful] Build details are at http://buildbot.shogun-toolbox.org/builders/deb4%20-%20python3/builds/865 | 22:25 |
shogun-buildbot | build #2898 of bsd1 - libshogun is complete: Failure [failed configure] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2898 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 22:26 |
shogun-buildbot | build #36 of xenial - libshogun is complete: Failure [failed compile] Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/36 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 22:28 |
-!- lambday [6d9941c0@gateway/web/freenode/ip.109.153.65.192] has quit [Ping timeout: 250 seconds] | 22:30 | |
shogun-buildbot | build #26 of FC23 - libshogun - aarch64 is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/26 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 22:34 |
lisitsyn | lambday: no need to register anything, they are directly in the map | 22:40 |
lisitsyn | subclasses could set up default values | 22:40 |
-!- sonne|osx [~sonne@x4db3bf36.dyn.telefonica.de] has quit [Quit: sonne|osx] | 23:09 | |
shogun-buildbot | build #27 of FC23 - libshogun - aarch64 is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/27 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 23:13 |
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun | 23:17 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 23:17 | |
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Client Quit] | 23:18 | |
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 246 seconds] | 23:40 | |
-!- lambday [6d9941c0@gateway/web/freenode/ip.109.153.65.192] has joined #shogun | 23:44 | |
-!- mode/#shogun [+o lambday] by ChanServ | 23:44 | |
--- Log closed Tue Jun 07 00:00:29 2016 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!