Open in new window / Try shogun cloud
--- Log opened Sat Sep 06 00:00:30 2014
-!- PirosB3 [] has joined #shogun01:49
-!- PirosB3 [] has quit [Quit: PirosB3]03:33
shogun-buildbotbuild #901 of nightly_default is complete: Failure [failed notebooks]  Build details are at  blamelist: Fernando Iglesias <>, khalednasr <>03:55
-!- PirosB3 [] has joined #shogun12:24
-!- pickle27 [] has joined #shogun15:23
-!- pickle27 [] has quit [Remote host closed the connection]15:28
-!- abinash_panda [~abinash@] has joined #shogun19:53
-!- kislay [~Abhijeet@] has joined #shogun19:57
-!- jiaolong [58a0e807@gateway/web/freenode/ip.] has joined #shogun20:09
-!- lambday [7aa77165@gateway/web/freenode/ip.] has joined #shogun20:19
-!- HeikoS [] has joined #shogun20:39
-!- mode/#shogun [+o HeikoS] by ChanServ20:39
kislayhi HeikoS20:44
-!- yorkerlin [18d4da3a@gateway/web/freenode/ip.] has joined #shogun20:53
-!- sonne|osx [] has joined #shogun20:56
@HeikoSkislay: hey there!20:58
kislayHeikoS, Kevin won't be able to make it.20:58
@HeikoSkislay: ok20:58
@HeikoSkislay: well you are here and can represent him20:59
kislayyeah sure20:59
@HeikoSok let me check who is here20:59
@HeikoSor who is missing20:59
-!- khalednasr [~k.nasr92@] has joined #shogun21:00
@HeikoSabinash_panda, jiaolong, khalednasr, lambday, yorkerlin hi21:00
lambdayhi HeikoS21:00
@HeikoSkhalednasr: theo coming?21:00
khalednasrHeikoS, dunno21:00
@HeikoSkhalednasr: probably not, I met him and now know how unreliable his emails are ;)21:00
@HeikoSlisitsyn, wiking hi#21:01
-!- pickle27 [~AndChat79@] has joined #shogun21:03
pickle27Hey guys sorry I can't stay long, but congrats everyone on gsoc!21:04
-!- thoralf [~thoralf@] has joined #shogun21:05
-!- mode/#shogun [+o thoralf] by ChanServ21:05
@thoralfHello :)21:05
kislaypickle27, yeah :)21:05
kislaythoralf, Hi T!21:05
@thoralfHey K!21:06
-!- HeikoS [] has quit [Ping timeout: 245 seconds]21:06
@thoralfHey Sergey21:07
@thoralflisitsyn: How's your new job?21:08
-!- Saurabh7 [~guestie_@] has joined #shogun21:08
@lisitsynthoralf: pretty good!21:08
@lisitsynthoralf: though taking too much time :D21:08
@thoralflisitsyn: I can imagine (same here ;))21:08
-!- kislay__ [~Abhijeet@] has joined #shogun21:10
@lisitsynthoralf: so heiko should be there to lead this meeting thing? :D21:11
-!- kislay [~Abhijeet@] has quit [Ping timeout: 276 seconds]21:12
lambdaylisitsyn: you've got a new job?21:14
@lisitsynlambday: yes21:15
lambdaylisitsyn: which company?21:15
@lisitsynlambday: yandex21:15
@lisitsynlambday: not only job but a new city :D21:15
lambdaylisitsyn: you're not in samara anymore21:15
@lisitsynlambday: now moscow21:16
@lisitsynlambday: it was a good time to move on ;)21:16
lambdaylisitsyn: lol yeah getting old, eh?21:16
@lisitsynhaha yeah21:16
@lisitsynso it is better to move before getting way too old21:16
lambdaylisitsyn: you're still way younger and I'm jealous :D21:17
@lisitsynlambday: if you don't know about yandex in simple words it's just a russian google :D21:17
lambdaylisitsyn: really!! frickin' awesome!21:17
pickle27Nice dude that sounds awesome21:17
lambdaylisitsyn: google people called me for a job offer but I still don't have my frickin' degree! :(21:17
@lisitsynlambday: wow cool, is it like a real offer?21:18
lambdaylisitsyn: not offer - I've gotta appear for an interview21:18
@lisitsynI went through this interviw hell :D21:18
lambdaylisitsyn: sucks!21:18
lambdaylisitsyn: will try for Amazon! they already called for 6-7 times.. I'm hating it at Oracle21:19
@lisitsynlambday: ahh btw how is oracle?21:19
lambdaylisitsyn: company is good - my project - sucks!21:19
lambdaydon't want to work on such an application layer..21:19
@lisitsynlambday: what do you do?21:20
lambdaymostly j2ee and cloud stuf21:20
-!- Heiko_web [05439144@gateway/web/freenode/ip.] has joined #shogun21:20
lambdayhi Heiko_web21:20
Heiko_webI had some connectivity issues21:20
Heiko_webvery sorry21:20
Heiko_webMy client basically told me nobody was here :(21:20
Heiko_webWho is around?21:20
lambdayI am21:21
Heiko_webwas just writing an angry email, haha21:21
Heiko_webcould we get a ping from everyone?21:21
lambdayHeiko_web: lol :P21:21
kislay__I am here!21:21
Heiko_webjust found out over the channel logs21:21
-!- HeikoS [] has joined #shogun21:21
-!- mode/#shogun [+o HeikoS] by ChanServ21:21
@HeikoSah now things work again, weird pidgin21:21
@lisitsynI am here21:21
-!- Heiko_web [05439144@gateway/web/freenode/ip.] has quit [Client Quit]21:22
@thoralfHello H21:22
@HeikoSso lets see pickle27, lisitsyn, thoralf wiking hi21:22
@HeikoSkislay__: abinash_panda, jiaolong, khalednasr, Saurabh7, lambday, yorkerlin hi!21:22
kislay__hi HeikoS21:23
yorkerlinhi again21:23
@HeikoSvery sorry for the delay and my previous issues21:23
Saurabh7oh, heelllo21:23
@HeikoSI think the meeting can start21:23
@HeikoSSaurabh7: hi nice to see you21:23
@HeikoSyorkerlin: hi! emt will not make it21:23
@HeikoSlambday: dino probably also not21:23
@HeikoSbut thats fine21:23
@HeikoSok lets start21:23
@HeikoSWelcome to our GSoC 2014 celebration meeting.21:23
@HeikoSWe want to use this opportunity to thank everyone involved for their kick-ass work, we are extremely excited about the results this year and think that all project made significant contributions to Shogun.21:24
@HeikoSThe community of our little ML project is ever growing, and we really hope that a few of you guys will stay around afterwards -- you are MORE than welcome.21:24
@HeikoSAny form of contribution is welcome: extending your/other projects, fixing bugs in your/other projects, writing more examples, blog posts, release help.21:24
@HeikoSAgenda for today21:24
@HeikoS-Project summaries (please every student/mentor pair generate some 3-5 sentences that summaries the project, including links to notebooks/examples/etc21:24
@HeikoS-Blog posts about GSoC21:24
@HeikoS-What next?21:24
@HeikoSanyone has some other points?21:24
@HeikoSanyone not asleep? :D21:25
@thoralfThanks you for the introduction.21:26
lambdayI can go first21:26
@HeikoSok then, I suggest we start with the project summaries21:26
@HeikoSlambday: cool!21:26
lambday(have some text ready)21:26
lambdaymy project was about implementing a number of kernel hypothesis testing in shogun that would enable us to to perform a number of cutting edge kernel two-sample tests and kernel-independence tests to measure variable interactions in generic machine learning problems.21:26
@HeikoSthoralf: do you wanna go next with abinash_panda, jiaolong ?21:27
lambdayThe final target was to build a flexible framework using such measures for feature selection.21:27
lambdayIn course, we faced some difficulties while implementing the algorithms and dealt with the issues theoretically as well as implementation-wise.21:27
lambdayWhat we have presently in shogun is a framework for kernel hypothesis testing flexible enough to incorporate new measures as well as the feature selection framework that can readily use those measures as a criterion for selecting best features among all.21:27
lambdayA few things are yet to be completed - which includes refactoring the hypothesis framework to incorporate streaming features support for kernel independence test (something that we discovered later that we will need).21:27
lambdayAnother thing (wiking would like this) is that we'd want our feature selection algorithm to work with sparse features as well.21:27
@HeikoSlambday: haha, indeed wiking would like this ! :)21:28
lambdayI am still working on the notebook for feature selection to be finished and hoping to push this by tomorrow.21:28
-!- hushell [] has joined #shogun21:28
lambdayHeikoS: yes :P21:28
@HeikoSlambday: oh yeah that would be great21:28
@HeikoShushell: hi!21:28
lambdayIn future I will continue with this work and also planning to join Heiko and Dino in London for further work (which includes leaving my job which I hate currently).21:28
hushellHeikoS: hey!!21:28
* lambday done21:28
@HeikoSlambday: oh yeah we should talk about this a bit soon :)21:28
lambdayHeikoS: will do a PM21:28
@thoralfNext one.  Large-Scale Multilabel Project with abinash_panda.21:29
@HeikoSthoralf:  go for it!21:29
@thoralfThe project aimed on large-scale multilabel classification in SHOGUN.  We succeeded in proving that we can learn the multilabel problem and implemented different approaches.21:30
@HeikoS(the others can prepare text already)21:30
@thoralfWe actually didn't succedd in the "large-scale" part, but we already have some ideas ready how to improve perfomance of structured-output framework.21:30
-!- sonne|osx [] has quit [Quit: sonne|osx]21:30
@HeikoSthoralf: cool, details?21:30
@HeikoSSaurabh7: can you be the next one? after thoralf is done?21:31
Saurabh7HeikoS, okie21:31
@thoralfOverall complexity was quite high, so I'd say it's great for SHOGUN that we can advertize multilabel classification as a new feature.21:31
@thoralfDetails of how to improve SO framework would exceed this meeting.  I guess I'm done here.  ;)21:32
@HeikoSthoralf: ok21:32
@HeikoSSaurabh7: want to be next?21:32
@HeikoSyorkerlin: can you also already prepare something?21:32
Saurabh7hi guys21:32
Saurabh7So as a part of my project I wrote detailed notebooks about various topics like SVMs,  regression models (, introduction to ml with shogun (, evaluation , etc. I have tried to organize them properly so as to benefit new users.21:32
Saurabh7Also  I made various changes and improvements to the web demos adding features like model selection, usage of databases and more flexibility and coverage of shoguns capabilities e.g in GPs.21:32
@HeikoSSaurabh7: the new web demos with GP are quite cool btw everyone,21:33
Saurabh7done :)21:33
@HeikoSSaurabh7: wiking needs to help deploy this, but you can already clone the repo and play with it locally21:33
@HeikoSSaurabh7: thanks! :)21:33
Saurabh7eah we should update those on the website i guess21:33
@HeikoSSaurabh7: you should help us pushing this (otherwise it might take a while, its your work so you have a good motivation :)21:34
@HeikoSyorkerlin: want to be next?21:34
yorkerlinstill writing on it21:34
-!- sonne|osx [] has joined #shogun21:34
@HeikoSjiaolong: ?21:35
@thoralfApproximate inference in SO framework (hushell and Jiaolong).  ipython notebooks:21:35
@thoralfI can try a quick summary on this one as well.21:35
@HeikoSthoralf:  great !21:36
@HeikoSok while people are writing, here is one for parijat and fernando21:36
@thoralfI think this project was successful as well.  The notebooks show different applications that Jiaolong made possible with our SO learning framework.21:36
hushellThanks thoralf :)21:36
@HeikoSParijat: "The summary of the work done in Fundamental ML algorithms in the second half of GSoC is as follows :1. Implement KD-Trees and Ball trees for fast NN search.2. Implement kernel density estimation using single tree as well as dual tree approach.3. Implement generic EM and generic mixture model classes. The CGaussian class needs to be refactored and other classes like that for multinomial distribution need to be implemented so that this whol21:37
@HeikoSWith orientations and seminars everyday, I am finding much less time to spend on Shogun these days. But as things settle down a bit, I should be able to contribute more often. "21:37
jiaolongThere is anoter one for scene segmentation:
@HeikoSeveryone check out parijats stuff, very nice!21:37
@thoralfjiaolong: Can you name the three (?) approximations you implemented?21:37
@HeikoSjiaolong, thoralf, these notebooks should be merged asap btw21:37
jiaolongHeikoS: yes, I am just coming back from two confernece, will finish them soon21:38
@HeikoSjiaolong: great!21:38
hushellWe still have several PRs left, will be done shortly21:38
@HeikoShushell: great, yeah that should be done soon, before we forget21:38
@HeikoSbut lets talk about that later,21:39
jiaolongHeikoS: this summer we just finish two inference algs: graph cuts, max-product LP relaxation21:39
jiaolongHeikoS: yeah21:39
jiaolongI would like to stay in shougun to improve current algorithms and also implements new algorithms.21:39
jiaolongAs I am also doing working on domain adaptation, I would like to contribute to domain adaptation projects in shogun as well.21:39
hushellHeikoS: I'll keep them in mind :)21:39
@HeikoSjiaolong: you are super welcome to do so. lambday just joined the core-dev team as you know :)21:39
@HeikoSjiaolong: great!, in particular your research paper code might be useful21:40
@HeikoShushell: nice :)21:40
@HeikoSok cool all done?21:40
@HeikoSkislay__: ?21:40
kislay__yeah ready.21:40
@HeikoSkislay__: go!21:40
kislay__The project aimed at making a lucid interfacing between Shogun and OpenCV21:41
kislay__The first half consisted of integrating OpenCV's Mat datastructure with Shogun's SGMatrix. Did some changes in the FisherLDA class for implementing fisherfaces face rec. directly.21:41
kislay__The second half consisted mainly of showing cool computer vision applications using Shogun and OpenCV through notebooks. It comprised of Automatic Number Plate Recognition, Scene classification using SIFT, Active Appearance Model along with markdown documentations showcasing the C++ integration. This half was more fun as it consisted of mainly implementations.21:41
kislay__I realized that there's humongous possibility of doing things in Shogun related to ML and CV. I have a thesis this year and was really thinking of doing something that not only helps Shogun but also spruce up my thesis. I haven't decided the topic yet so will love to get your inputs for this.21:41
-!- sonne|osx [] has quit [Quit: sonne|osx]21:42
kislay__done i think.21:42
@HeikoSkislay__: great!21:42
@HeikoSkislay__: what about your open PRs?21:42
@HeikoSkislay__: you sent an email on thise21:42
@HeikoSwe should try to get them merged soon, you should push everyone to do this :)21:42
kislay__yeah. I have to take care of some of them.21:42
@HeikoSok good!21:42
@HeikoSso next one: yorkerlin21:43
yorkerlinThe GSoC project I implemented several variational methods for GPC. The variational methods include: approximating the posterior covariance matrix exactly (for accuracy) assuming artificial structure of the posterior covariance matrix (for speed) However, there methods still NOT scalable to big data. I still working on refining the notebook but will attempt to finish it this week.21:43
yorkerlinThe next step is to implement large scale methods via stochastic variational approach and variable selection approach. (already read some papers and will implement them this year) The final goal is to extend to latent GP models and hierarchical latent models (eg, correlated topic models).21:43
@HeikoSyorkerlin: yeah the notebook updates are important, please push them21:44
@HeikoSyorkerlin: awesome, looking super forward to you to implement those things!21:44
@HeikoSyorkerlin: we will talk to emt about all that21:44
@HeikoSI might submit a nips workshop on this21:44
@HeikoSwe will see21:44
@HeikoSok, did we forget any project?21:44
@HeikoSlambday, lisitsyn ?21:45
khalednasrHeikoS, I21:45
@HeikoSkhalednasr: ah yes21:45
@HeikoSkhalednasr: go for it :)21:45
khalednasrI'll go :)21:45
khalednasrThe goal of the project was to implement the basic deep learning algorithms. The following was implemented: Standard neural networks, onvolutional nets, Restricted Boltzmann Machines, Deep Belief Networks, Deep autoencoders and Convolutional autoencoders.21:45
khalednasrAs documentation, I wrote notebooks and c++ examples that cover all the algorithms in the neuralnets module.21:45
khalednasrI also did some work refactoring the linalg library and adding most of the methods needed by the neuralnets module to it.21:45
khalednasrWhat remains to be done is modifying the neuralnets module to make it use the linalg library, so that it'll support running neural nets on the GPU. I'll try to find time to work on that sometime soon.21:45
@HeikoSkhalednasr: do the things already work again linalg?21:45
@HeikoSkhalednasr: that would be *amazing*21:46
@lisitsynyeah that's the crucial thing21:46
@HeikoSkhalednasr:  in fact, if we get lambday's parallel/distributed framework running, we can parallelise things on GPU clusters21:46
@HeikoSkhalednasr: (you need some minor refactoring for that only)21:46
khalednasrHeikoS, yeah that'd be good21:46
@HeikoSkhalednasr: but very exciting, I think we should aim to reproduce a large benchmark result, I can donate some time on a GPU cluster21:47
@HeikoSkhalednasr: so if you are interested, we can make Shogun be able to do that21:47
@HeikoSkhalednasr: but linalg is first21:47
@HeikoSkhalednasr: can you also post your notebook links=?21:47
@HeikoSok, so I guess thats it, any comments?21:48
@HeikoSif not, the second part of the meeting is quick: just me talking about a few things, please stop me and get a discussion going, if you feel like it21:48
@HeikoSSo second part of agenda:21:48
@HeikoS-Blog posts about GSoC21:48
@HeikoSWe would like to encourage everyone of you to blog about GSoC.21:48
@HeikoSThis is a great way for us to get publicity.21:48
@HeikoSIt helps us doing the GSoC again next year: It helps us getting in, it makes more students apply, etc.21:48
@HeikoSWe would be EXTREMELY thankful if you could produce something *this* week.21:48
@HeikoS(or next)21:49
@HeikoSIt does not have to be super polished - just write about your experience, and present some of your work.21:49
@HeikoSsocial media would also be fine21:49
@HeikoSeven if just a few sentences21:49
@HeikoSWe would like to collect all posts and reference from a big one that we are writing.21:49
@HeikoSThe plan is to get it featured at
@HeikoSSee also the github issue
@HeikoSlisitsyn, thoralf, sonney2k, wiking, abinash_panda, hushell, khalednasr, kislay__, lambday, pickle27, Saurabh7, yorkerlin ?21:50
yorkerlincool. will working on it next week since the notebook job.21:50
@HeikoSyorkerlin: as said, it doesnt need to be super advanced21:50
@HeikoSyorkerlin: just some content on Shogun and GSoC21:51
@HeikoSOk then next point! (wiking would be helpful here)21:51
@HeikoSRELEASE. Shogun 3.421:51
@thoralfHeikoS: Fine with writing a blog post here.21:51
@thoralfHeikoS: One per project or a big one?21:51
@HeikoSthoralf: cool, please share the link once done!21:51
@HeikoSthoralf: both (coordinate the big one with me)21:51
@thoralfHeikoS: Current status is "I like the idea" ;)21:51
@HeikoSthoralf: Ill share a doc for the big one, but a small one on the projects would also be cool21:51
@HeikoSthoralf: greatly appreciated!21:52
@HeikoSWe aim for a release ASAP to put all your great work to the world (really important, and cool 8-) ).21:52
@HeikoSFor that, we need your help.21:52
@HeikoSFirst, make sure your project is in a release-able state (people will judge you on the quality of your work, so better clean again ;) ).21:52
hushellalso on personal webpages, facebook, twitter, google+ will be nice ;)21:52
@HeikoShushell: yes !21:52
@HeikoSSecond, we got a list of things to solve for the release that are general minor (and major) issues all around Shogun.21:52
@HeikoSFeel free to help us hacking on any of those.21:52
@HeikoSHere is a list
@HeikoSAlso this means: FEATURE FREEZE until we release.21:53
@HeikoSPlease do not send patches that extend Shogun before release, but rather fix some open problem. We count on you :)21:53
@HeikoSlisitsyn, thoralf, sonney2k, wiking, abinash_panda, hushell, khalednasr, kislay__, lambday, pickle27, Saurabh7, yorkerlin we need all the hands we can for a release, any comments on that?21:53
@HeikoSif not, lets go to the last point of the meeting21:54
@thoralfHeikoS: Please let's start with a milestone on github.21:54
lambdayHeikoS: my first goal is to finish the notebook - then will write a NEWS about linalg which will be useful for the release21:54
hushellwhat's the timeline for 3.4?21:54
@HeikoShushell: it would be good this month21:54
@HeikoSthoralf: that link is a milestone21:54
lambdayHeikoS: also, was discussing something with lisitsyn about something - will discuss with that issue with you21:54
@HeikoSthoralf: feel free to extend (there is an old clean-up release milestone)21:54
@thoralfHeikoS: :)21:54
@HeikoSthoralf: but only feasible stuff21:55
@HeikoSwe tend to put too much21:55
@HeikoSthis release is really GSoC feature, so just a few things that should work21:55
hushellHeikoS: ok, the left behind PRs have to be hurried up21:55
@HeikoSmainly binary packages, warnings, severe bugs, build, etc21:55
@HeikoShushell: yeah exactly21:55
@HeikoSall PRs should be closed21:55
@HeikoShushell: I mean merged21:56
@HeikoSlambday: ok21:56
@HeikoSlambday: release related21:56
@thoralfHeikoS: Binary package for OSX shouldn't be part of the milestone.  This issue may explode. ;)21:56
hushellHeikoS: got it21:56
@HeikoSthoralf: yep, I agree, feel free to remove it21:56
lambdayHeikoS: not exactly ... but future shogun goal related21:56
@HeikoSlambday: ok21:56
@HeikoSso really main focus is the release now21:57
@HeikoSthe world shall know about the great stuff we did21:57
@HeikoSlisitsyn, thoralf, sonney2k, wiking, abinash_panda, hushell, khalednasr, kislay__, lambday, pickle27, Saurabh7, yorkerlin there is for example the task to remove all warnings from the buildbot21:57
@HeikoSvery easy, but lot of work21:57
@HeikoSmostly documenting functions, fixing math, etc21:57
@HeikoSstatic warnings another one21:57
@HeikoSok if no comments, then the last part of the agenda:21:58
@HeikoSWHATS NEXT? (after release)21:58
@HeikoSStammtisch: We would love to get the Stammtisch going again.21:58
@HeikoSThis is every Monday evening, and everyone is welcome to talk about Shogun, and other things.21:58
@HeikoSI will send emails about this. It would be great if you guys would hang out and if we get a discussion going on a regular basis.21:58
@HeikoScomments on Stammtisch?21:58
@HeikoSIf not, the very last point21:59
@HeikoSFinally, GSoC brought a few things to light, that would be great to continue working on.21:59
@HeikoSIn particular, the linalg framework would be a great great improvement if being used globally in all Shogun.21:59
@HeikoSkhalednasr, lambday, lisitsyn you guys shoud lead that btw21:59
@HeikoSEveryone could help pushing this via adding new operations to the interface, and using them from algorithms.21:59
@HeikoSI will soon add github issues on this.21:59
@lisitsynyeah I'll try to join that process again21:59
@HeikoSlisitsyn: its not really about adding operations to the interface, and start porting our algorithms to that22:00
@HeikoSAny other thoughts ideas? What do you think are things that would be good to work on this winter (before next GSoC)22:00
@lisitsynwe need GWoC22:00
@thoralfHeikoS: Regarding what's next.22:00
@thoralf1. Improving usability for "new" users.  Providing installation scripts/binary packages for common platforms (Ubuntu/Debian, MacOSX, ?).22:00
@thoralfMaybe this can be done after GSoC release within another release.  I created a milestone 3.4.1 for this.  This is open for discussion and changes.22:00
@thoralf2. Making SHOGUN more scalable (multilabel project showed that we still don't scale well).22:00
@thoralf2a. If we don't know how far we scale try "bigger" examples, not only small toy data sets.  (Not neccessarily from tests)22:00
@thoralf2b. Investigate bordercases which don't scale and try to optimize SHOGUN on them.22:01
@HeikoSthoralf: great points, maybe also add this to the shogun hack agenda22:01
@HeikoSthoralf: great ideas,22:01
@HeikoSthoralf: I agree we really want to make shogun large scale22:01
@HeikoSthoralf: we should have github issues for all this, and link them to the wiki22:01
@thoralfHeikoS: Okay, regarding 1. we're on the same page.22:01
@HeikoSthoralf:  its here:
@thoralfRegarding 2. we first should create awareness in the core team to encourage people to benchmark and optimize.22:02
@HeikoSthoralf: very good point,22:02
@thoralf(Not stopping after successful toy examples.)22:02
@HeikoSgunnar and me talked a bit about that in fact22:02
@HeikoSthoralf: yeah. totally22:02
@HeikoSthoralf: and stop adding new methods, the red line for the next year should be improving existing things22:02
@HeikoSi.e. faster/stable/usable22:03
@HeikoSthoralf: would be great if you could document your thouhts, that makes it easier22:03
@HeikoSthoralf: maybe you have an idea for a kind of development guideline that ensures scalability ? (like we did with the unit tests back then) ?22:03
@HeikoSlambday: you also had a point on whats next?22:04
@HeikoSI guess we can close the meeting and then continue discussing22:04
@HeikoSIf there are no other points, we would like to heartly thank you all again for your great work. We love you all and hopefully see you around :)22:04
@HeikoSMeeting is closed.22:04
lambdayHeikoS: something that I thought of while trying to refactoring hypothesis framework22:04
lambdayHeikoS: in many many places in shogun, we need to rely on some combinatorial combinations of componets22:05
lambdayfor which we presently mostly rely on virtual calls22:05
lambdayI was thinking if we can do better than that22:05
@thoralfHeikoS: It's hard to automate benchmarking.  I think people should try to apply their stuff to bigger data and then use profiling to find "hotspots".22:06
lambdaythe alternative being relying on highly templated design22:06
@thoralfThen start optimizing the hotspots.22:06
lambdaythe only issue with this is that we might have difficulties while exposing to modular interaces22:06
@HeikoSthoralf: yeah, but I mean something like, you always have to include a benchmark or so, because people quickly forget those things22:06
lambdayfor that I was thinking that maybe we can use CSGObjects as wrappers mostly22:06
@HeikoSlambday: cool, so whats the gain there? speed?22:06
lambdayand send the most time consuming parts to c++ internals22:07
lambdayHeikoS: exactly22:07
@HeikoSlambday: in fact, that could be great as a motivation to clean shoguns internals up22:07
lambdayHeikoS: just have a look at this simple example -
@HeikoSlambday: would you mind documenting this a bit? like basic idea, example, what has to be done, etc?22:07
lambdayHeikoS: yeah22:07
@HeikoSlambday: and put it in the wiki, I mean, write a gitrhub issue and link it in the wiki for the hack22:07
lambdayHeikoS: I was already planning to write something on this on the refactoring hypothesis frameworks issue22:08
lambdayHeikoS: for example, it sucks that we can't use sparse features with gaussian kernels currently22:08
lambdaybecause using existing stuffs its tough to reimplement that22:08
-!- sonne|osx [] has joined #shogun22:08
lambdayI think we can do better22:08
@HeikoSlambday: so thats one order of magnitude?22:09
@HeikoSlambday: yeah22:09
@HeikoSlambday: wow, that actually would mean we could fe-factor a lot of the internals right?22:09
lambdayHeikoS: also, plugin framework that lisitsyn proposed is essential I think22:09
@HeikoSlambday: these things are the ones we should do within the next year, really get Shoguns c++ code modernized, and faster this way22:09
lambdayHeikoS: yeah just an idea though22:09
@lisitsynguys as for plugins framework22:09
@HeikoSlambday: also we can think about many existing problems22:09
@HeikoSas serialisation etc22:10
lambdayHeikoS: yeah22:10
@lisitsynI need some help with it I think22:10
@HeikoSlambday: can you write a little motivation for all this?22:10
@HeikoSso that everyone is aware?22:10
@HeikoSlambday: this is also super useful for grant applications (like for the hack22:10
@HeikoSlisitsyn: same there,22:10
lambdayso maybe we can spend until next GSoC more on these engineering issues than ML stuffs22:10
@HeikoSlambday: totally22:10
@lisitsynHeikoS: same with what? :)22:10
@HeikoSlisitsyn: I am willing to help22:10
@lisitsynok just a few first steps then22:11
@HeikoSlisitsyn: would be good to write down your ideas in a compressed way, and then link them in the wiki I put up22:11
lambdayHeikoS: lisitsyn will have to help in this :D22:11
@HeikoSlisitsyn: very useful for getting money for the hack (if we have discussions and concrete ideas)22:11
@lisitsynlets we three HeikoS lambday agree on naming22:11
@lisitsynthen I put it into shougn22:11
@lisitsynas shogun org22:11
@lisitsynat github22:11
@lisitsynwe will have some kind of separate project :D22:11
@lisitsynso if anyone wants some kind of plugin based swig library - just use our core thing22:12
@HeikoSlisitsyn: everyone wants this22:12
@HeikoSlisitsyn: you know, at the meeting in NY, everyone was jealous that we can do the interface thing22:12
@HeikoSlisitsyn: would be good to have a separate project, I already talked to s?ren about this last year22:13
@HeikoSlisitsyn: please write down22:13
@HeikoSlisitsyn: all good for the hack22:13
@HeikoSlisitsyn: we can do all this in a week22:13
@lisitsynHeikoS: no I just propose to move it into github org already22:13
-!- Mathnerd314 [] has quit [Read error: Connection reset by peer]22:13
@lisitsynwiki is ok as well22:13
@lisitsynbut I want to push it a bit other way as well22:13
@HeikoSlisitsyn: just for internal synchronisation etc22:13
@lisitsynI need a name :D22:14
@lisitsynHeikoS: lambday:22:14
@HeikoSlisitsyn: I will think of something22:14
@lisitsyncurrently its aer :D22:14
@HeikoSlisitsyn:  applications of this are huge, stan people for example want this too22:14
lambdaywe gotta be the best in the field, men!!22:14
@HeikoSlisitsyn:  john langford told me: "Basically you have everything that every other project would like to have"22:14
lambdaylisitsyn: name... bad person to ask22:14
@HeikoSlambday: yes sir!22:14
-!- Saurabh7 [~guestie_@] has quit [Quit: Leaving]22:15
@lisitsynokay then it stays aer yet :D22:15
lambdaylisitsyn: aer++22:15
-!- Mathnerd314 [] has joined #shogun22:15
@HeikoSlisitsyn: please write a bit about the general idea in the wiki, cannot say that often enough :)22:15
@HeikoSaer ++ haha22:15
lambdaylatin for air, eh? :P22:16
-!- khalednasr [~k.nasr92@] has quit [Quit: Leaving]22:16
@HeikoSlambday: I will check your c++ listings22:16
lambdaylisitsyn: so we'll start a whole new branch of shogun, refactoring bits and pieces as we go along22:16
@thoralfHeikoS: I put my ideas to the hack page.22:16
@lisitsynok guys22:17
@lisitsynohh it is outdated :D22:17
lambdayHeikoS: just a small example that is - to demonstrate that we can do better than java style interfaces using c++ templates :D22:17
-!- kislay__ [~Abhijeet@] has quit [Quit: Leaving]22:17
@lisitsynok guys be back later22:19
@lisitsynI'll try to update you on this thing22:19
lambdayI am pretty psyched to get these stuffs working before the hackathon22:19
@HeikoSlisitsyn: cool!22:19
@HeikoSsee you :)22:19
@HeikoSlambday: awesome :)22:19
@HeikoSyeah me too22:19
lambdayand there we can integrate the whole thing and have a super cool release22:19
@HeikoSI am psyched about the hackathon22:19
@HeikoSjust need to get the money22:20
@HeikoSand for that we need documented discussions22:20
@HeikoSlambday: yeah agreed22:20
@HeikoSlambday: ok cool22:20
@HeikoSlambday: gotta go now, lets keep the stammtisch going, we should discuss those stuff at least once a week22:21
lambdayI'm gonna start this discussion by putting a few of the ideas for hypothesis testing framework22:21
lambdayHeikoS: absolutely22:21
@HeikoSlambday: sweet22:22
@HeikoSlambday: I will be around a bit more from now on22:22
lambdayif we have a working demo inside shogun that we can do this - then that's a good point to start future discussions22:22
@HeikoSI agree22:22
lambdayHeikoS: me too22:22
@HeikoSlambday: like the linalg stuff22:22
lambdayHeikoS: yeah22:22
lambdaythanks to khaled for the brilliant stuffs he added22:23
@HeikoSlambday: +122:23
@HeikoSlambday: I gotta go now (dinner)22:23
@HeikoSlambday: lets chat this week!22:23
@HeikoSI will be online!22:23
lambdayHeikoS: alright22:23
lambdayHeikoS: I'll try to chalk out a few things by then22:24
@HeikoSlambday: cool22:24
@HeikoSlambday: lets talk at stammtisch on monday eve latest22:24
lambdayHeikoS: alright.22:25
-!- thoralf [~thoralf@] has quit [Quit: Konversation terminated!]22:27
-!- PirosB3 [] has quit [Quit: PirosB3]22:30
-!- pickle27 [~AndChat79@] has quit [Quit: Bye]22:40
lambdaylisitsyn: please write something about the ideas on the plugin framework :P22:47
lambdaylisitsyn: wiking suggested that we can take inspiration from other existing libraries as well22:48
-!- HeikoS [] has quit [Quit: Leaving.]23:09
-!- yorkerlin [18d4da3a@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]23:38
-!- pickle27 [] has joined #shogun23:41
-!- PirosB3 [] has joined #shogun23:52
-!- HeikoS [] has joined #shogun23:53
-!- mode/#shogun [+o HeikoS] by ChanServ23:53
--- Log closed Sun Sep 07 00:00:31 2014