Open in new window / Try shogun cloud
--- Log opened Wed May 21 00:00:57 2014
thoralf|homeabinash_panda: Ping?00:14
-!- pickle27f [~pickle27@216.191.231.218] has quit [Quit: This computer has gone to sleep]00:20
-!- iglesiasg [~iglesiasg@194.151.204.16] has joined #shogun00:49
-!- mode/#shogun [+o iglesiasg] by ChanServ00:49
@iglesiasgHey guys00:51
kislayyo iglesiasg :)01:05
-!- khalednasr [~k.nasr92@41.69.238.139] has quit [Ping timeout: 252 seconds]01:19
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun01:24
shogun-notifier-shogun: Parijat Mazumdar :develop * f7b47a9 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/f7b47a9b44aa46e4d50cfaabe5b8e50e13ad1e4701:24
shogun-notifier-shogun: CART added01:24
shogun-notifier-shogun: Fernando Iglesias :develop * 3dcc48e / / (3 files): https://github.com/shogun-toolbox/shogun/commit/3dcc48e63784f907adad4c7c5504d5d4d196e39b01:24
shogun-notifier-shogun: Merge pull request #2239 from mazumdarparijat/CART01:24
shogun-notifier-shogun:01:24
shogun-notifier-shogun: CART added01:24
shogun-buildbotbuild #2962 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/296201:29
-!- iglesiasg [~iglesiasg@194.151.204.16] has quit [Quit: Leaving]01:32
shogun-buildbotbuild #2299 of bsd1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/229901:35
-!- thoralf|home [~thoralf@91-65-137-195-dynip.superkabel.de] has quit [Quit: Konversation terminated!]01:42
shogun-buildbotbuild #632 of precise - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/precise%20-%20libshogun/builds/63201:49
-!- kislay [~Abhijeet@103.240.205.174] has quit [Ping timeout: 255 seconds]01:53
-!- travis-ci [~travis-ci@ec2-75-101-216-118.compute-1.amazonaws.com] has joined #shogun02:20
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/2564988302:20
-!- travis-ci [~travis-ci@ec2-75-101-216-118.compute-1.amazonaws.com] has left #shogun []02:20
-!- abinash_panda [0e8b2998@gateway/web/freenode/ip.14.139.41.152] has quit [Ping timeout: 240 seconds]02:55
-!- yorkerlin [81614fb3@gateway/web/freenode/ip.129.97.79.179] has quit [Quit: Page closed]02:56
shogun-buildbotbuild #10 of nightly_fedora is complete: Failure [failed git]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_fedora/builds/10  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>, Wu Lin <yorker.lin@gmail.com>, Kevin Hughes <kevinhughes27@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com>03:00
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has joined #shogun03:55
shogun-buildbotbuild #311 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/311  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com>03:58
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]04:24
-!- Saurabh7 [~Saurabh7@117.222.8.201] has joined #shogun05:16
-!- Saurabh7 [~Saurabh7@117.222.8.201] has quit [Ping timeout: 255 seconds]06:00
-!- Saurabh7 [~Saurabh7@117.222.8.201] has joined #shogun06:02
shogun-buildbotbuild #808 of nightly_default is complete: Failure [failed notebooks]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/808  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>, Wu Lin <yorker.lin@gmail.com>, Kevin Hughes <kevinhughes27@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com>06:15
-!- abinash_panda [dce36163@gateway/web/freenode/ip.220.227.97.99] has joined #shogun06:16
abinash_pandaHey thoralf !06:16
-!- abinash_panda [dce36163@gateway/web/freenode/ip.220.227.97.99] has quit []06:25
-!- Saurabh7 [~Saurabh7@117.222.8.201] has quit [Ping timeout: 255 seconds]06:39
-!- Saurabh7 [~Saurabh7@59.88.35.186] has joined #shogun06:52
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has quit [Remote host closed the connection]06:58
-!- kislay [~Abhijeet@103.240.206.194] has joined #shogun07:05
-!- abinash_panda [d2d43db9@gateway/web/freenode/ip.210.212.61.185] has joined #shogun08:04
-!- abinash_panda [d2d43db9@gateway/web/freenode/ip.210.212.61.185] has quit [Ping timeout: 240 seconds]08:33
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has quit [Quit: Leaving.]08:38
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has joined #shogun08:39
-!- thoralf|home [~thoralf@91-65-137-195-dynip.superkabel.de] has joined #shogun09:13
@besser82thoralf|home, g'Morning!  :D09:14
thoralf|homeMoin.09:14
thoralf|homeYo besser09:14
thoralf|homebesser82: What's up?09:15
@besser82thoralf|home, see query  ;)09:16
thoralf|homeSELECT count(*) FROM irc09:17
thoralf|home12345609:17
-!- jiaolong [9e6d09e4@gateway/web/freenode/ip.158.109.9.228] has joined #shogun09:21
-!- thoralf|home [~thoralf@91-65-137-195-dynip.superkabel.de] has quit [Quit: Konversation terminated!]10:01
-!- kislay [~Abhijeet@103.240.206.194] has quit [Ping timeout: 240 seconds]10:22
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun12:01
-!- mode/#shogun [+o HeikoS] by ChanServ12:01
Saurabh7hey HeikoS !12:06
@HeikoSSaurabh7: hey! how are things12:06
@HeikoSnice to see you  here12:06
Saurabh7i am usually around12:06
Saurabh7but our time dos\esnt match i think12:06
@HeikoSSaurabh7: I haven't been recently since very busy, but I plan to hang around most of the time12:06
Saurabh7:)12:06
@HeikoSSaurabh7: yeah indeed :)12:06
@HeikoSwhat is yours?12:06
Saurabh7HeikoS, right now its 3 30 noon12:07
Saurabh7gmt + 530 i think12:07
@HeikoSah wow here we have 110812:08
@HeikoSam12:08
@HeikoSSaurabh7: I am a bit busy now, but shall we try to talk a bit later today?12:08
@HeikoSI would like to discuss a few things with you12:08
@HeikoSwill you be around for a while?12:08
Saurabh7HeikoS, yup12:09
@HeikoSSaurabh7: cool!12:09
@HeikoSsee you in a few hours then! :)12:09
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]12:09
Saurabh7HeikoS, okay will be here12:09
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun12:13
-!- mode/#shogun [+o HeikoS] by ChanServ12:13
-!- Saurabh7 [~Saurabh7@59.88.35.186] has quit [Ping timeout: 264 seconds]12:15
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]12:24
-!- Saurabh7 [~Saurabh7@117.217.103.228] has joined #shogun12:38
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun12:55
-!- mode/#shogun [+o HeikoS] by ChanServ12:55
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Client Quit]12:58
-!- kislay [~Abhijeet@103.228.76.4] has joined #shogun13:02
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun13:34
-!- mode/#shogun [+o HeikoS] by ChanServ13:34
kislayhey HeikoS .13:37
-!- khalednasr [~k.nasr92@41.69.214.145] has joined #shogun13:45
lisitsyn1khalednasr: hey there14:08
khalednasrlisitsyn1, hey14:08
lisitsyn1khalednasr: how is it going?14:09
khalednasrlisitsyn1, done with the graph stuff, waiting for travis to clear before I send the PR :)14:10
lisitsyn1khalednasr: oh cool14:10
-!- [1]Saurabh7 [~Saurabh7@117.217.103.228] has joined #shogun14:26
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]14:28
-!- Saurabh7 [~Saurabh7@117.217.103.228] has quit [Ping timeout: 252 seconds]14:28
-!- [1]Saurabh7 is now known as Saurabh714:28
khalednasrlisitsyn1, any suggestions as to how I should approach that file-loading thing?14:34
lisitsyn1khalednasr: yeah that's interesting question14:36
lisitsyn1khalednasr: actually we had issue like that with some model file14:36
lisitsyn1just this gsoc iirc14:36
khalednasrwhich model?14:37
lisitsyn1khalednasr: https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/io/UAIFile.cpp14:37
lisitsyn1this is the thing14:38
lisitsyn1I don't know whether it is very good designed already14:38
lisitsyn1but afaik it is something similar to what you'd need14:38
khalednasrcool! I'll take a look at it14:39
-!- travis-ci [~travis-ci@ec2-54-237-169-250.compute-1.amazonaws.com] has joined #shogun14:48
travis-ci[travis-ci] it's khalednasr's turn to pay the next round of drinks for the massacre he caused in khalednasr/shogun: http://travis-ci.org/khalednasr/shogun/builds/2568746714:48
-!- travis-ci [~travis-ci@ec2-54-237-169-250.compute-1.amazonaws.com] has left #shogun []14:48
khalednasrlisitsyn1, I took a look at the UAI file stuff, It would work if we only wanted to specify the connectivity of the network, but we also need to specify the type of layers and the hyper parameters14:59
khalednasrlisitsyn1, maybe I could use JSON?14:59
lisitsyn1khalednasr: nono I mean it is something that serves similar purpose14:59
khalednasrlisitsyn1, I see. what do you think about using json or xml?15:02
lisitsyn1khalednasr: I am quite ok with either of them15:02
khalednasrlisitsyn1, cool15:02
lisitsyn1khalednasr: the question is15:02
lisitsyn1you know we have json15:03
lisitsyn1but it is something that serializes our shogun internal things into json15:03
lisitsyn1and back15:03
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun15:04
-!- mode/#shogun [+o HeikoS] by ChanServ15:04
khalednasryeah I saw it15:04
-!- Saurabh7 [~Saurabh7@117.217.103.228] has quit [Quit: HydraIRC -> http://www.hydrairc.com <- The alternative IRC client]15:09
-!- lambday [67157f4f@gateway/web/freenode/ip.103.21.127.79] has joined #shogun15:14
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Ping timeout: 252 seconds]15:15
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun15:15
shogun-notifier-shogun: lambday :feature/linalg * 115cc99 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/115cc9998f55d8856ac9eb61b778a386924bf10115:15
shogun-notifier-shogun: added rowwise and colwise sum methods in linalg15:15
shogun-notifier-shogun: Soumyajit De :feature/linalg * a676eb3 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/a676eb30d42f9562833f82936e743c5749d6ce6315:15
shogun-notifier-shogun: Merge pull request #2235 from lambday/feature/linalg15:15
shogun-notifier-shogun:15:15
shogun-notifier-shogun: added rowwise and colwise sum methods in linalg15:15
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun15:18
-!- mode/#shogun [+o HeikoS] by ChanServ15:19
-!- khalednasr [~k.nasr92@41.69.214.145] has quit [Ping timeout: 252 seconds]15:24
-!- khalednasr [~k.nasr92@41.69.237.19] has joined #shogun15:25
@HeikoSkislay: hey, sorry for not getting back earlier, was in a meeting. Now going out for lunch but back soon!15:29
kislayHeikoS, oh. np!15:29
-!- cuphrody [cuphrody@211.69.205.236] has joined #shogun15:35
-!- Saurabh7 [~Saurabh7@59.88.33.86] has joined #shogun15:45
-!- Saurabh7 [~Saurabh7@59.88.33.86] has quit [Ping timeout: 276 seconds]15:55
-!- Saurabh7 [~Saurabh7@117.222.11.53] has joined #shogun16:21
-!- pickle27f [~pickle27@216.191.231.218] has joined #shogun16:23
-!- Saurabh7 [~Saurabh7@117.222.11.53] has quit [Ping timeout: 265 seconds]16:27
-!- iglesiasg [~iglesiasg@2001:6b0:1:1041:25ea:1e69:30f2:e06c] has joined #shogun16:46
-!- mode/#shogun [+o iglesiasg] by ChanServ16:46
@iglesiasgwow back to .se16:46
lisitsyn1iglesiasg: you back to .se? ;)16:47
@iglesiasglisitsyn1, http://www.sais.se/sais2014/16:47
@iglesiasglisitsyn1, workshop starting tomorrow at my old uni16:47
lisitsyn1ahh cool16:48
thoralfHey :)16:57
* thoralf just got a confirmation that we can do the SHOGUN-Hands-On on Sunday, 27th of July at ReasearchGate.16:59
thoralfBerlin-Mitte, close to Hauptbahnhof and other transports.16:59
thoralfI'll send a more official announcement for this soon.17:01
@iglesiasgthoralf, great!17:03
lisitsyn1ah berlin berlin17:05
lisitsyn1ah guys I have to announce I probably won't be there :(17:05
@wikingunfortunately possible me neither17:05
lisitsyn1it seems I am moving to moscow17:07
lisitsyn1just around this date17:07
lisitsyn1will have to do a lot of stuff17:07
* wiking moving as well17:08
lisitsyn1wiking: where to? :)17:08
@wikingor seems like at least17:08
@wikinglisitsyn1: heheh dont want to jinx it17:08
lisitsyn1japan?17:08
lisitsyn1:D17:08
@wikinglisitsyn1: close...17:08
lisitsyn1really?17:08
@wikingyep, but i really dont want to jinx it17:09
lisitsyn1wiking: sure ;)17:09
thoralfwiking going to Baidu? ;)17:10
-!- lisitsyn1 [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.]17:11
@HeikoSthoralf: nice!!!17:17
thoralfHeikoS: I think we should wait a few weeks to announce this.17:17
thoralfHeikoS: First announce workshop.17:18
thoralfwait 2 weeks17:18
thoralfAnnounce hands-on.17:18
thoralfJust spreading it a bit over time.17:18
-!- Saurabh7 [~Saurabh7@117.200.183.38] has joined #shogun17:26
@HeikoSthoralf: yep!17:27
@HeikoSthoralf: a man, time overkill here, would love to push this more17:28
@HeikoSthoralf: could you update the google doc?17:28
thoralfHeikoS: Don't have my google credentials at work... maybe on the weekend.17:29
thoralfSorry.17:29
@HeikoSthoralf: no worries, just in general we should push this a bit in the near future17:30
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]18:15
-!- pickle27f [~pickle27@216.191.231.218] has quit [Quit: This computer has gone to sleep]18:24
@HeikoSlambday: hey18:27
@HeikoSjust reading your patch18:27
lambdayHeikoS: hi!18:27
lambdayHeikoS: yeah just read your comments18:27
lambdayHeikoS: so I added sum_rowwise, sum_colwise as you suggested18:27
@HeikoSlambday: cool18:28
lambdaygot this feeling that you'd like it this way :)18:28
@HeikoSlambday: haha :)18:28
@HeikoSlambday: so lets step a bit back and get the big picture of the current state now ....18:28
lambdayHeikoS: regarding square, though, I'm not sure!18:28
@HeikoSlambday: ok, why?18:28
@HeikoSlambday: this square thing is nice to have, but it will create a new matrix right?18:28
lambdayHeikoS: yeah that's true18:29
@HeikoSlambday: ok18:29
lambdayfor a very brief period thogh18:29
@HeikoSlambday: so, what is missing for the MMD stuff to work?18:29
@HeikoSlambday: but allocating takes time18:29
@HeikoSlambday: whereas summing up squares does not18:29
lambdayHeikoS: alright but that18:30
lambday's a way too specific task to go to linalg, no?18:30
lambdayumm.. although squared_sum can be done18:31
-!- cuphrody [cuphrody@211.69.205.236] has quit [Ping timeout: 240 seconds]18:31
lambdayso - I'll change that!18:31
@HeikoSlambday: yeah sum up squares is sueful18:31
@HeikoSuseful18:31
lambdayHeikoS: as of now, nothing is missing for mmd to work - in fact I just modified quadratic time mmd to compute statistic and variance on the fly18:32
@HeikoSlambday: ok cool18:32
@HeikoSlambday: what about all those different MMD null approximations?18:32
@HeikoSlambday: thats where we left the thing a while ago18:32
lambdayHeikoS: yeah! - so as we discussed, I added these enums18:32
@HeikoSlambday: cool18:33
lambdayso, there are 2 enums for MMD_SPECTRUM now18:33
@HeikoSlambday: nice18:33
@HeikoSlambday: do you have an estimate for how long these things will take to wrap up? then dino and I can plan a bit18:33
@HeikoSlambday: we will probably start writing some experiments with all of this later in June18:33
lambdayHeikoS: well, quadratic time mmd I'll finish by tonight or tomorrow18:34
lambdaythen BTest and linear time mmd - by next week!18:34
lambdayactually I'll make a doc to keep track of things18:34
@HeikoSlambday: ok, no rush, I am just asking18:34
@HeikoSlambday: yeah a doc would be great18:34
@HeikoSlambday: since GSoC also started officially now, we should soon transition to the feature selection stuff :)18:35
lambdayHeikoS: since I deviated from this for a while I lost track :/18:35
@HeikoSlambday: haha :)18:35
@HeikoSlambday: yeah no worries, you are doing great work here18:35
@HeikoSlambday: I already might have a customer for feature selection stuff, so that is nice18:35
lambdayHeikoS: wow that's cool! :d18:35
@HeikoSlambday: do you need any input on something?18:35
lambdayHeikoS: umm.... can't think of any right now - if I remmebe something, will mail you and dino?18:36
@HeikoSlambday: yeah sure18:37
lambdayHeikoS: so, we had a whole bunch of tasks planned - I'll reschedule things a bit after finishing with MMD... as per earlier estimation this would have been done long ago while I was doing btest :(18:37
@HeikoSlambday: yeah I know18:37
@HeikoSI am actually thinking of postponing a few things thats why I am asking you18:37
@HeikoSlambday: but also great to have, so its fine if you spend another week on those things18:38
@HeikoSthen later in the summer we will make use of those things and maybe request some updates/find problems ,etc18:38
lambdayHeikoS: yeah that sounds good18:38
lambdaywell I was kind of worried to catch up on the schedule that I planned18:38
@HeikoSwhich of the feature selection things would you like to start with?18:39
@HeikoSlambday: dont worry, rather take time to do things properly, thats worth more :)18:39
lambdayHeikoS: we can start with smola's paper, right? feature selection via dependence maximization18:40
-!- pickle27f [~pickle27@216.191.231.218] has joined #shogun18:40
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]18:41
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun18:41
-!- mode/#shogun [+o HeikoS] by ChanServ18:41
@HeikoSlambday: BAHSIC you mean?18:42
lambdayHeikoS: yeah18:42
@HeikoSlambday: about the Block stuff18:42
@HeikoSlambday: (sorry to jump back here=)18:42
lambdayno worries18:43
@HeikoSlambday: if one computes the sum of a matrix block18:43
@HeikoSdoes this copy the block around?18:43
lambdayHeikoS: nah18:43
@HeikoSor how does this work?18:43
@HeikoSlambday: sorry for asking (I could also read code, but since you are here thats faster)18:43
lambdaywell, as of now I just used eigen3, which refers to blocks in the orig matrix via those block() calls18:44
lambdayHeikoS: so, no copying18:44
@HeikoSlambday: and SGMatrix?18:44
-!- Saurabh7 [~Saurabh7@117.200.183.38] has quit [Ping timeout: 252 seconds]18:44
@HeikoSah sorry18:44
@HeikoSso that creates a map first18:44
@HeikoSlambday: so the naive implementation could just map indices then?18:45
@HeikoSlike a subset?18:45
lambdayHeikoS: yeah that's the best way I guess18:45
lambdayHeikoS: I haven't added naive implementation though18:45
@HeikoSlambday: cool!18:45
@HeikoSlambday: no need for now18:45
@HeikoS(entrance task :) )18:45
lambdayhahaha :D18:45
@HeikoSmaybe make a list of future entrance tasks somewhere18:45
@HeikoSseriously, I always forget them, its important to grow those on the fly18:46
lambdayumm yeah might be good - we always forget stuffs18:46
@HeikoSand its a good  entrance task18:46
@HeikoSone learns a lot18:46
@HeikoSlambday: ok then18:46
@HeikoSBAHSIC is the first algo for feature selection18:46
@HeikoSlambday: some design questions naturally appear18:46
@HeikoSlambday: Preprocessor18:47
@HeikoSvariable-selection-preprocessor18:47
@HeikoSwe dont have any of those18:47
lambdayHeikoS: yeah - I have to read the paper in more details to get how do they do it.. currently not having much idea how do add these in preprocessors...18:49
@HeikoSlambday: its just backward elimination of features using HSIC as a measure of dependence18:49
@HeikoSlambday: super easy18:49
@HeikoSlambday: and design is independent of paper18:49
@HeikoSlambday: maybe dino and I can share some thoughts, we discussed this a while ago18:50
@HeikoSwill try to squeeze this in18:50
lambdayalright - so basically among different feature selection algo, the measure of dependence is what differs, right?18:50
@HeikoSlambday: yes18:53
@HeikoSthe algorithm always removes the feature that is least informative18:54
lambdayHeikoS: so the different independence measures that we planned to have, will be used by this feature selection framework18:54
lambdaysorry for sounding super obvious - just trying to get clear thoughts :)18:56
-!- Saurabh7 [~Saurabh7@117.200.183.38] has joined #shogun18:57
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Read error: No route to host]19:00
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun19:00
-!- mode/#shogun [+o HeikoS] by ChanServ19:00
@HeikoSlambday: yes exactly19:00
@HeikoSlambday: sorry got disconnected19:00
@HeikoSlambday: yes these measure can then be combined with different ways to select fetaures19:00
lambdayHeikoS: alright... so, this would all be done via some subclass(es) of preprocessor19:01
-!- iglesiasg [~iglesiasg@2001:6b0:1:1041:25ea:1e69:30f2:e06c] has quit [Quit: Leaving]19:02
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun19:02
shogun-notifier-shogun: Heiko Strathmann :develop * 9d77bda / src/shogun/kernel/ANOVAKernel.cpp,src/shogun/kernel/ANOVAKernel.h: https://github.com/shogun-toolbox/shogun/commit/9d77bda196015fd125c9010c5ed5e5de78e996b919:02
shogun-notifier-shogun: added obtain_from_generic method19:02
shogun-notifier-shogun: Heiko Strathmann :develop * c23f2ec / src/shogun/kernel/ANOVAKernel.cpp,src/shogun/kernel/ANOVAKernel.h: https://github.com/shogun-toolbox/shogun/commit/c23f2ecf1e16850da3b685be37364162807def1f19:02
shogun-notifier-shogun: Merge pull request #2242 from karlnapf/develop19:02
shogun-notifier-shogun:19:02
shogun-notifier-shogun: added CANOVAKernel::obtain_from_generic19:02
@HeikoSlambday: yes19:04
@HeikoSlambday: so we need one that selects a number of variables from the input features19:04
lambdayHeikoS: okay... so, something inside apply_to_feature_matrix (under DensePreprocessor) would return a feature matrix with selected features19:07
@HeikoSlambday: I guess yes19:08
@HeikoSlambday: in fact we can think about avoiding the copy19:08
lambdayHeikoS: use subsets instead19:08
@HeikoSlambday: we already have subsets for the features19:08
@HeikoSlambday: but that might be tricky19:08
@HeikoSlambday: not sure whether its worth it or too messy19:08
@HeikoSlambday: a copy is definitely the first choiu19:08
@HeikoSchoice19:08
lambdayalright - yeah we gotta think a bit about how this would work19:09
@HeikoSlambday: ok cool19:09
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]19:11
-!- HeikoS [~heiko@0542590b.skybroadband.com] has joined #shogun19:11
-!- mode/#shogun [+o HeikoS] by ChanServ19:11
lambdayHeikoS: okay - so, we need one hierarchy of classes FeatureSelection - some base class - subclasses use different dependence measures - provides means for applying the subsetting19:14
@HeikoSlambday: yes!19:14
lambdayHeikoS: and then one preprocessor having an instance of this feature selection inside it19:14
@HeikoSlambday: I think it would be nice if it was so modular that we can just plug in different measures into different algorithms19:14
lambdayHeikoS: umm... yeah that's better - so may be we provide some method in those dependence measures only :/19:15
lambdayHeikoS: then feature selection works with those methods19:15
lambdaynot sure though - just thinking19:15
@HeikoSlambday: no wait, the preprocessor does the feature selection19:16
@HeikoSand it gets an instance of the dependence measure19:16
lambdayHeikoS: yeah - so, it all works with C<dependence-measure-class>::sample_null() ??19:18
@HeikoSlambday: mmh goo dpoint19:19
@HeikoSlambday: I am not sure wheather we need that19:19
lambdayI mean are the null samples the only things that we need to do feature selection?19:19
lambdayHeikoS: I'm not sure - may be I'll read the paper today and try to get an idea :/19:19
@HeikoSfor BAHSIC, one just computes the statistic doesnt one? (not sure currently, hehe)19:19
@HeikoSlambday: see also arthurs lecture slides19:20
@HeikoShttp://www.gatsby.ucl.ac.uk/~gretton/coursefiles/rkhscourse.html19:20
@HeikoSlceture 4,5,6 slides19:20
lambdayHeikoS: checking...19:20
lambday(I took printed copies of those :D)19:20
@HeikoSslide 17719:21
@HeikoShehe19:21
lambdayHeikoS: yeah - just the arg max of statistic19:23
lambdaythen remove that one19:23
lambdayah got it19:23
lambdayso, other algorithms would work this way only - then we can have it this way - FeatureSelection ---> DensePreprocessor, have an instance of IndependenceTest, overrides apply_to_feature_matrix, inside this it uses IndependenceTest::compute_statistic and remove the irrelevent ones iteratively19:26
lambdayso this loop would go inside that19:26
@HeikoSyeah, so that should work with any dependence measure right? :) backwards elemination19:26
@HeikoSexactly19:26
lambdayHeikoS: yeah - although I am stuck at (Why backwords?) :D19:27
lambdayI mean, if backword elimination works for all then this simple structure would suffice19:28
@HeikoSlambday: if you go forward and start adding features, you will miss pairwise dependencies19:28
@HeikoSlambday: or what did you mean?19:28
lambdayaaahhh19:28
lambdayyeah19:28
lambdaygot your pt19:28
lambdayHeikoS: yeah I meant that only19:28
lambdaycool!19:30
lambdayHeikoS: alright! feeling a bit relaxed.. got the main points - I'll try to finish quadratic time mmd (almost done - except spectrum)..19:34
@HeikoSlambday: cool! yeah relax, this will be fun, no need to freak out, rather enjoy, its cool stuff! :)19:34
lambdayHeikoS: I'll read the soon and start with this issue first - so we'll have the framework ready - later we can add different dependence measures19:35
lambdayHeikoS: oh another thing - will it work similarly for conditional dependence measures?19:36
-!- jiaolong [9e6d09e4@gateway/web/freenode/ip.158.109.9.228] has quit []19:38
@HeikoSlambday: good point!19:40
@HeikoSlambday: when you set up the doc, put those things in19:40
@HeikoSwe can then give feedback and discuss19:40
lambdayHeikoS: alright that sounds good!19:40
@HeikoSlambday: I agree that we should draft base classes first, without implementations,19:41
lambdayyeah!19:41
-!- lisitsyn [~qdrgsm@80.252.20.67] has joined #shogun19:48
-!- Saurabh7 [~Saurabh7@117.200.183.38] has quit [Ping timeout: 252 seconds]19:58
-!- Saurabh7 [~Saurabh7@117.200.183.180] has joined #shogun20:01
@HeikoSSaurabh7: hey!20:01
Saurabh7HeikoS, hi!20:01
Saurabh7HeikoS, did you get my msg?20:01
@HeikoSSaurabh7: do you have a minute to discuss?20:01
@HeikoSnot yet, currently working throgh a batch of emails :)20:01
Saurabh7HeikoS, no i pinged you here20:02
Saurabh7i think got dc first20:02
@HeikoSSaurabh7: ah no20:02
Saurabh7yeah have some time20:02
@HeikoSSaurabh7: so lets discuss the first week20:02
@HeikoSand update the doc20:02
Saurabh7I have prepared a draft of first task20:02
@HeikoSSaurabh7: nice!20:02
Saurabh7HeikoS, yes i will update the detail part tomorrow first thing20:03
@HeikoSSaurabh7: could you give me a link?20:03
Saurabh7HeikoS,  should i send in a pr about the first one?20:03
@HeikoSSaurabh7: no need for a PR, just a link for me to see20:03
Saurabh7nbviewer?20:03
Saurabh7ok20:03
@HeikoSSaurabh7: but maybe even put the current state into a PR then I can start reviewing20:03
Saurabh7sec20:03
@HeikoSSaurabh7: but wait20:03
@HeikoSSaurabh7: lets also discuss a bit what to go in there, what is your plan for this notebookss20:04
@HeikoS?20:04
@HeikoSbut maybe its good to go from notebook, so lets see the link! :)20:04
Saurabh7HeikoS, ah20:05
Saurabh7here it is http://nbviewer.ipython.org/gist/Saurabh7/5557715ec5ad323ffafe#20:05
@HeikoSSaurabh7: this already looks very nice20:07
Saurabh7HeikoS,  oh ! good..20:08
Saurabh7any suggestions?20:08
@HeikoSSaurabh7: I like this direction a lot20:08
@HeikoSSaurabh7: providing some very simple example of things and then pointing to the other notebooks20:08
Saurabh7HeikoS, yes thats what i planned20:09
@HeikoSSaurabh7: so this is like basic supervised classification20:09
Saurabh7HeikoS, we can group similar nbs together further20:09
@HeikoSSaurabh7: I will try to come up with some feedback soon, but for now this looks good so you can keep on working in this direction20:10
Saurabh7HeikoS, yes its the simplest i could get hold of20:10
@HeikoSSaurabh7: yes I agree20:10
@HeikoSSaurabh7: great work! :)20:10
Saurabh7so instead of say 'supervised'  pointing to wiki , we can point it to the bunch20:10
@HeikoSlike the reference to the web demo20:10
@HeikoSSaurabh7: yeah maybe, thats also good for the website20:11
@HeikoSlike it scikit20:11
@HeikoSSaurabh7: BTW could you condense a list of Shogun's features on the fly20:11
@HeikoSmaybe put them in the document20:11
@HeikoSSaurabh7: since we desperately need this up to date :)20:11
@HeikoSSaurabh7: ok this looks good,20:12
Saurabh7HeikoS, yes because the nbs are pretty complicated to browse through20:12
Saurabh7HeikoS, yup i am trying to set up a diagram20:12
@HeikoSSaurabh7: I guess you could elaborate a bit on other cases: other features, regression ... etc, but that might come later20:12
@HeikoSSaurabh7: yes, great idea! really like it20:12
@HeikoSSaurabh7: You should add some section on the interfaces of features20:13
@HeikoSSaurabh7: like number of vectors, dimension, etc20:13
@HeikoSAnd maybe also some things on other feature types20:13
Saurabh7HeikoS,  ok nice, will try to look into that20:13
@HeikoSthis notebook can grow a bit, there could also be an example for regression and maybe even a preprocessor20:14
@HeikoSsince those are very fundamental20:14
Saurabh7yes20:14
Saurabh7i was afraid to make it too complicated20:14
@HeikoSlike one simple example each?20:14
@HeikoSyes I know what you mean20:14
@HeikoSbut if the thing just grows a bit thats fine20:15
Saurabh7ok i will try to come up with something about regression20:15
@HeikoSnice20:15
@HeikoSI mean the focus is just on the setting of the method, not so much about details, so you can just use a particular algorithm under the CMachine interface and then point to another notebook for method details,20:16
@HeikoSlike you already did20:16
@HeikoSmean remove, unit variance preprocessor might be good20:16
@HeikoSor whitening20:16
@HeikoSsince those are std things20:16
@HeikoSSaurabh7: ok, let me know how it goes20:17
Saurabh7yup seems good20:17
@HeikoSSaurabh7: btw you should include nbviewer links in your weekly updates to the list, people will love them :)20:17
Saurabh7i will try to set up a diagram and example by tomorrow20:17
@HeikoScool20:17
@HeikoSSaurabh7: and send a PR of this one already, together with the nbviwer link, I will put some comments,20:18
@HeikoSSaurabh7: Nice! Glad how this is going! Keep on the good work!20:18
Saurabh7ok will send it now20:18
Saurabh7but its late here so will have to address them tomorrow20:18
@HeikoSSaurabh7:  I will leave now then. Having a look later today.20:18
Saurabh7thanks!20:18
@HeikoSSaurabh7: yeah no problem20:18
Saurabh7ok np20:18
-!- sonne|osx [~sonne@89.204.139.93] has joined #shogun20:21
kislaySaurabh7, I too saw the nbviewer link! it's cool :)20:21
Saurabh7kislay, hey , thanks :)20:22
Saurabh7kislay,  hows ur project going?20:22
kislaySaurabh7, nice.20:23
Saurabh7kislay, great20:23
Saurabh7see you later .. pretty late now :)20:24
kislaySaurabh7, not much late.20:25
kislay;)20:25
kislaySaurabh7, bye20:25
Saurabh7kislay,  i am at home now so .. :p20:25
kislaySaurabh7, me too actually. but never mind.20:25
-!- HeikoS [~heiko@0542590b.skybroadband.com] has quit [Quit: Leaving.]20:28
-!- Saurabh7 [~Saurabh7@117.200.183.180] has quit [Quit: Leaving]20:33
-!- sonne|osx [~sonne@89.204.139.93] has quit [Quit: sonne|osx]20:38
-!- thoralf|home [~thoralf@91-65-137-195-dynip.superkabel.de] has joined #shogun20:48
thoralf|homeHey.20:48
@besser82thoralf|home, ho!  :D21:19
-!- pickle27f [~pickle27@216.191.231.218] has quit [Quit: This computer has gone to sleep]21:24
-!- pickle27f [~pickle27@216.191.231.218] has joined #shogun21:26
thoralf|homebesser82: Bo!21:37
shogun-buildbotbuild #312 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/312  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>21:46
@besser82thoralf|home, :D21:49
@besser82thoralf|home, ^^^  Heiko hat's kaputt gemacht  :-P21:49
thoralf|homeNee.  Have a look: https://github.com/shogun-toolbox/shogun/pull/223221:51
thoralf|homeThis will "fix" memcheck.21:51
thoralf|homeFinally.21:51
thoralf|homeSpent so many hours on debugging and fixing the whole mess-up.21:51
thoralf|homes/-up//21:52
@besser82thoralf|home, ic  ;)21:52
@besser82thoralf|home, to me the pr looks good  ;)21:52
@besser82thoralf|home, shall I merge?21:52
@besser82thoralf|home, so they can blame me for doing nasty stuff?21:53
thoralf|homeIf you like. ;)21:54
@besser82thoralf|home, sure  ;)21:55
@besser82thoralf|home, travis lloks good so lets-a-go!21:55
shogun-notifier-shogun: Thoralf Klein :develop * d61a691 / configs/valgrind.supp: https://github.com/shogun-toolbox/shogun/commit/d61a6914c92de2c7853833f48e7e80220f4eebf121:55
shogun-notifier-shogun: Removing last memory warnings on buildbot (could not reproduce them locally).21:55
shogun-notifier-shogun: Bj?rn Esser :develop * da1698f / configs/valgrind.supp: https://github.com/shogun-toolbox/shogun/commit/da1698f3fb6da96907fcbde3e41d2de9b9148cef21:55
shogun-notifier-shogun: Merge pull request #2232 from tklein23/more_valgrind_suppressions21:55
shogun-notifier-shogun:21:55
shogun-notifier-shogun: Removing last memory warnings on buildbot (could not reproduce them locally)21:55
@besser82thoralf|home, btw. what is the correct procedure to convert the mallet model and stuff to vectors?21:57
@besser82thoralf|home, so I can train a model from the findings with liblinear?22:01
thoralf|homebesser82: You can.22:01
thoralf|homebesser82: Create a feature vector, where the i-th component is the probability of the i-th topic.22:02
@besser82thoralf|home, let's get that german  ;)22:02
-!- travis-ci [~travis-ci@ec2-54-234-84-161.compute-1.amazonaws.com] has joined #shogun22:39
travis-ci[travis-ci] it's Bj?rn Esser'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/2572695222:39
-!- travis-ci [~travis-ci@ec2-54-234-84-161.compute-1.amazonaws.com] has left #shogun []22:39
-!- pickle27f [~pickle27@216.191.231.218] has quit [Quit: This computer has gone to sleep]23:14
-!- khalednasr [~k.nasr92@41.69.237.19] has quit [Quit: Leaving]23:51
--- Log closed Thu May 22 00:00:59 2014