Shogun Machine Learning - IRC Logs
--- Log opened Thu Feb 02 00:00:07 2017
-!- suhas2go [uid201652@gateway/web/] has quit [Quit: Connection closed for inactivity]00:17
@sukeyPull Request #3608 "Move the solver related code to new file"  synchronized by MikeLing -
@sukeyPull Request #3608 "Clean up KNN" -
-!- lisitsyn [~lisitsyn@] has quit [Ping timeout: 264 seconds]07:14
-!- lisitsyn [~lisitsyn@] has joined #shogun07:14
-!- shogitter [] has quit [Ping timeout: 264 seconds]07:15
-!- ShellcatZero [] has quit [Ping timeout: 264 seconds]07:15
-!- shogitter [] has joined #shogun07:15
-!- nagdev [dce39546@gateway/web/freenode/ip.] has quit [Ping timeout: 260 seconds]07:18
-!- ShellcatZero [] has joined #shogun07:29
-!- abhinav_ [~abhinav@] has joined #shogun07:43
-!- abhinav_ [~abhinav@] has quit [Ping timeout: 240 seconds]07:53
-!- praisethemoon [~praisethe@] has joined #shogun09:03
-!- suhas2go [uid201652@gateway/web/] has joined #shogun10:16
@sukeyPull Request #3608 "Clean up KNN"  synchronized by MikeLing -
@sukeyPull Request #3608 "Clean up KNN"  synchronized by MikeLing -
-!- mikeling [uid89706@gateway/web/] has joined #shogun11:53
mikelinghey, why "./tests/unit/shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_*" doesn't work for me even after I run make shogun-unit-test12:41
mikelingok the unit-test works, but I have no idea what't the meaning of the error my got13:11
mikelinghi lisitsyn, could you give me some help on I have no idea about this error13:11
mikelingand how could I debug my code when the ctest told me unit-KNN has SegFault? I use lldb on OS X13:13
-!- nagdev [dce39546@gateway/web/freenode/ip.] has joined #shogun14:08
@wikingmikeling, still around?14:34
mikelingwiking: yep14:34
@wikingok so14:34
@wikinghow do you run your cmake on your machine?14:34
@wikingyou dont need DBUILD_DASHBOARD_REPORTS14:35
@wikingnor DDEBUG=ON14:35
@wikingyou should have14:35
@wikingimo it's on by default by just in case14:35
mikelingwiking: ok, I will try it right now. Thank you for your help!:)14:37
@wikingand note that some some part of knn14:37
mikelingdo you mean the error I got about knn unittest? could you tell me more specifically about that?14:40
@wikingok so i was wondering if the falconn part might be using openmp14:41
@wikingthat's why14:41
@wikingyou might not be able to reproduce it on osx with clang compiler by defalt14:41
@wikingbut i dont see any #pragma in the falconn code14:41
@wikingso that's not it14:41
@wikingso you should be able to reproduce14:41
@wikingbtw if you run14:42
@wikingeverything is passing you?14:42
@wikingbecause for me it seems there are some serious errors in everyjob14:42
@wikingon travis14:42
mikelingwiking: yeah, if I just use ctest and everything works for me14:44
@wikingyou might be missing some libraries?14:44
@wikinghence some code is just not being compiled?14:45
@wikingi'm not so sure14:45
@wikingbut in worst case14:45
@wikingyou could fire up a docker image :)14:45
@wikingand check it yourself14:45
@wikingimo there's a quite good docker implementation on osx already14:45
mikelingwiking: oh, so let me use docker image and see what i can got. I'm not familiar with docker so it may every consuming before I got things right ;)14:49
@wikingif you follow .travis.yml14:51
@wikingthen you'll actually get what travis does14:51
@wikingif you figure that out15:01
@wikingcould you write a couple of lines of markdown document15:01
@wikinghow to do it15:01
@wikingi mean it's there15:01
@wikingjust it'd be a good doc for everybody to share15:02
@wikinghow to reproduce travis errors locally on your machine ;)15:02
* wiking is being super lazy :P15:02
nagdevwiking: I read up on all the work that has been done on the usual suspects project previously and I was thinking about working with KNN or neural nets. Do you know if anyone is already working on it?15:05
@wikingnagdev, mikeling is working on KNN atm15:12
@wikingi dont recall anybody working on nns15:12
mikelingwiking: nagdev I'm working on this issue actually :)
mikelingwiking: sure, I will write all the things down if I figure out all the things. :)15:15
mikelingsorry for the late reply, I just miss the message somehow15:15
nagdevmikeling: so can I just start working on it with you or will I need some background?15:16
nagdevI know KNNs in general15:16
-!- abhinav [~abhinav@] has joined #shogun15:21
-!- abhinav [~abhinav@] has quit [Client Quit]15:21
mikelingnagdev: yeah sure, actually I had commit a PR for that issue(, but the SegFault stop me verify if all my code works15:22
mikelingit will be great appreciate if you can check that code and give me some feedback15:27
nagdevmikeling: Cool. Give me some time to read all the existing work and catch up. I'll start giving my inputs as soon as I'm done with that15:27
nagdevwiking: couple more things. Firstly, the PCA implementation is already significantly faster than other toolboxes. Why is PCA still in the list of algorithms to be improved? Are there any further changes that need to be made in it?15:30
nagdevSecond, some of the algorithms listed in project list were already taken up in prev year's GSoC. Are they still not complete?15:31
mikelingnagdev: oh great! Thank you :)15:34
mikelingnagdev: BTW, are you going to apply GSoC this year? Which project you are interested in? if I may ask :)15:35
@wikingnagdev, mmm there's alway room for improvement :)15:38
@wikingi do not know actually in case of each and every model what needs to be improved15:38
@wikingbut rest assured none of them is prefect :)15:38
@wikingjust pick anything that is to your liking and nobody else if working on (you can figure this out from the open prs)15:39
@wikingor just do something that you find interesting15:39
@wikingand you reckon it would be a good feature to add15:39
@wikingand it's easy to add to shogun15:39
@wikingyou do not need to necessarily stick to that list15:39
@wikingit's just giving u some ideas15:39
@wikinglike for example we do not have a preprocessor15:39
@wikingthat does one-hot encoding for a categorical variable15:40
nagdevwiking: Sure. I think I'll start working on KNN only and figure out if there is something else that I can work on as well15:48
@wikingsince mikeling is alrready working on knn15:48
@wikingi would really suggest you to pick something else15:48
@wikingotherwise it's inevitable that you'll end up doing double work15:49
@wikingand as we would like you to take something on your own to see how you handle things ..... dont get me wrong collaboration is very good but as well getting deep into some parts of the code alone is important15:50
nagdevyeah i understand15:50
nagdevis it ok if I just lurk around to see how he's working because I'm fairly new to open source and don't know much about how to go around doing things. That's why I wanted to work with someone to start off15:51
@wikingnagdev, mmm the point of entrance tasks are actually this15:52
@wikingthey are small15:52
nagdevI'll just see how he's working (general direction and coding style)15:52
@wikingfairly easy to get you started15:52
@wikingand it's quite autonomous15:52
@wikingjust pick anything from the entrance tasks15:53
@wikingtry to fix it15:53
@wikingand send in a pr15:53
@wikingas soon as you send in a pr we can give you a hand15:53
@wikingtell you what is ok and what is wrong15:53
@wikingusually the first pr takes about 3-4-6 iterations15:53
@wikingtill it gets actually merged15:53
@wikingthere's really no better way to get started with this is to try and fail15:54
nagdevok sure. I'll try out some tasks.15:54
nagdevmikeling: I'll only apply if shogun gets selected because afaik it is the only ML related project in the organization list. I'm mostly interested in projects which involve core ML implementation and improvement.15:57
nagdevSo from the available list, usual suspects, real life problem solving and shogun extension are my preferred options15:58
nagdevUsual suspects seems to be the most likely in that.15:58
nagdevIn that too, I would prefer to work on PCA, KNN or neural nets as I'm working on some papers to improve the algorithms and it'd be great to translate that into actual, functioning code16:01
nagdevwiking: the entrance task doesn't mean that I have to commit to the project right?16:06
@wikingwhat do you mean?16:07
nagdevIf i try to work on a bug on a particular algo (say GP), I can still switch to some other project (like NN) later right?16:09
@wikingthen entrance is just about16:10
@wikingseeing how can you contribute/code/interract etc16:10
nagdevso I was looking at and to start with. There has been some work in these already. Is it still fine if I start with one of these?16:14
-!- abhinav_ [~abhinav@] has joined #shogun16:44
-!- abhinav_ [~abhinav@] has quit [Remote host closed the connection]17:24
-!- Jyo__ [0e8b9b18@gateway/web/freenode/ip.] has joined #shogun18:03
-!- Jyo__ [0e8b9b18@gateway/web/freenode/ip.] has quit [Client Quit]18:03
-!- praisethemoon_ [~praisethe@] has joined #shogun18:16
-!- praisethemoon [~praisethe@] has quit [Ping timeout: 255 seconds]18:17
-!- mikeling [uid89706@gateway/web/] has quit [Quit: Connection closed for inactivity]18:49
-!- nagdev [dce39546@gateway/web/freenode/ip.] has quit [Quit: Page closed]19:35
shogitter(etotientz) are there no newcomer issues to solve at
shogitter(etotientz) I m interested in contributing here21:21
shogitter(etotientz) maybe relevant to gsoc this year21:21
--- Log closed Fri Feb 03 00:00:09 2017