Open in new window / Try shogun cloud
--- Log opened Fri Feb 20 00:00:27 2015
-!- nszceta [~nszceta@] has joined #shogun00:20
-!- anishshah_ [uid50585@gateway/web/] has quit [Quit: Connection closed for inactivity]00:40
-!- kingsley_ [] has joined #shogun01:02
-!- pickle27 [] has joined #shogun01:03
kingsley_For what it's worth, I ran the AI (Artificial Intelligence) Forum in Silicon Valley in the 1990s.01:04
kingsley_It has been fun to watch machine learning advance over the years.01:04
kingsley_Today I'd like to write a bash shell script that accepts comma separated variables on its standard input, and reports on standard output a.) a formula that predicts one of the input columns as a function of the others, and optionally b.) some measure(s) of how accurate the prediction is..01:07
kingsley_I'm interested in using Shogun's command line interface within the shell script.01:08
kingsley_So, I'm about to learn more about Shogun by watching the YouTube video named "Shogun Toolbox Workshop 2014: Introduction by Heiko Strathmann (1/6) ".01:09
kingsley_I hope that Shogun's command line interface allows it to01:11
kingsley_1.) Split categorical variables in multiple binary variables,01:11
kingsley_2.) normalize input data so all input variables have similar means and standard deviations,01:12
-!- shogun-notifier- [] has joined #shogun01:13
shogun-notifier-shogun-web: Kevin Hughes :gsoc1015 * f148611 / static/css/styles.css,templates/home.html:
shogun-notifier-shogun-web: add gsoc card to the home page01:13
kingsley_3.) select the input variables (ie: features) that are best at predicting,01:13
kingsley_4.) try several regression algorithms, each validated against several folds of the training data, and01:14
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 244 seconds]01:15
kingsley_5.) report the model that worked best, plus maybe01:15
kingsley_6.) some standard measures of how accurate its predictions are, like residuals, area under the curve, or standard error.01:16
kingsley_I also hope Shogun can report the best model and coefficients in plain text as a mathematical formula like...01:17
kingsley_variable1 = 1.23 * variable2 + 4.56 * variable3 + 7.89 * variable401:18
kingsley_so it can easily be migrated to other applications.01:18
* kingsley_ will be back later....01:19
-!- PirosB3 [] has quit [Quit: PirosB3]02:42
nszcetagood questions for @wiking02:44
-!- nszceta [~nszceta@] has quit [Quit: My Mac has gone to sleep. ZZZzzz...]03:01
shogun-buildbotbuild #850 of nightly_none is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>, Wu Lin <>03:04
-!- shogun-notifier- [] has quit [Quit: transmission timeout]04:13
shogun-buildbotbuild #968 of nightly_default is complete: Failure [failed notebooks]  Build details are at  blamelist: Heiko Strathmann <>, Wu Lin <>04:14
-!- pickle27 [] has quit [Remote host closed the connection]05:11
-!- prakhar2511 [~prakhar25@] has joined #shogun05:21
-!- kshitijk [~kshitijk@] has joined #shogun05:41
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 252 seconds]05:56
-!- lambday [c40f106a@gateway/web/freenode/ip.] has joined #shogun06:26
-!- mode/#shogun [+o lambday] by ChanServ06:26
-!- sanuj [~androirc@] has joined #shogun07:13
-!- sanuj [~androirc@] has quit [Read error: Connection reset by peer]07:23
-!- shogun-notifier- [] has joined #shogun07:24
shogun-notifier-shogun: Kunal Arora :develop * b5e81f2 / / (2 files):
shogun-notifier-shogun: added native support in VectorSum.h07:24
shogun-notifier-shogun: Kunal Arora :develop * 2175615 / src/shogun/mathematics/linalg/internal/implementation/VectorSum.h:
shogun-notifier-shogun: used std::accumulate and added REQUIRE07:24
shogun-notifier-shogun: Soumyajit De :develop * 17b4490 / / (2 files):
shogun-notifier-shogun: Merge pull request #2708 from curiousguy13/native07:24
shogun-notifier-shogun: added native implementation for sum of vector in linalg07:24
-!- travis-ci [] has joined #shogun07:26
travis-ciit's Soumyajit De's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []07:26
-!- nszceta [~nszceta@] has joined #shogun07:26
-!- sanuj [~androirc@] has joined #shogun07:27
-!- sanuj [~androirc@] has quit [Client Quit]07:31
shogun-buildbotbuild #2653 of bsd1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Kunal Arora <>, Soumyajit De <>07:31
shogun-buildbotbuild #985 of FCRH - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Kunal Arora <>, Soumyajit De <>07:37
shogun-buildbotbuild #698 of osx1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Kunal Arora <>07:38
shogun-buildbotbuild #699 of osx1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Soumyajit De <>07:40
-!- besser82 [~besser82@fedora/besser82] has joined #shogun07:53
-!- mode/#shogun [+o besser82] by ChanServ07:53
-!- nszceta [~nszceta@] has quit [Quit: Textual IRC Client:]08:03
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 246 seconds]08:46
kingsley_Hmmm... now that I see that the command line's static interface is evidently just for simple experiments.09:08
@lisitsynkingsley_: hey09:09
kingsley_lisitsyn: Hello09:09
@lisitsynkingsley_: yeah I'd not recommend static for you09:09
@lisitsynit would be better to use python09:09
-!- vortex_ape [~vortex_ap@] has joined #shogun09:09
kingsley_Maybe the modular interface to python would be better for the possibly ambitious goals I described 8 hours ago.09:10
@lisitsynkingsley_: yes09:10
@lisitsynkingsley_: well most of things are easy to do with shogun09:11
@lisitsynbut would take some learning curve indeed..09:12
@lisitsynby 1) you mean one-hot encoding?09:12
kingsley_lisitsyn: If I ask you nicely, will you please tell me if you happen to know of any sample python code that shows how to call shogun to do the stuff I wrote about 8 hours ago?09:13
-!- prakhar2511 [~prakhar25@] has joined #shogun09:13
@lisitsynkingsley_: well I can write that if you have some time to wait09:13
* kingsley_ looks up "one-hot encoding"...09:14
@lisitsynkingsley_: it is just a mapping of category to a vector of zeros but one at position associated with this category09:14
@lisitsynlike red=1,0,009:14
@lisitsynand green=0,1,009:14
kingsley_lisitsyn: Yes, I'd like to automate that mapping.09:15
-!- kshitijk [~kshitijk@] has joined #shogun09:15
-!- exploreshaifali [~shaifali@] has joined #shogun09:15
kingsley_And yes, that was the first goal I listed 8 hours ago.09:16
@lisitsynkingsley_: ok then what if I write up this thing a bit later today and send it to you?09:16
kingsley_lisitsyn: Well, well, well!09:17
kingsley_That's a very nice offer.09:17
kingsley_Thank you very much.09:17
kingsley_I'd be happy to see whatever you think would be appropriate.09:17
@lisitsynkingsley_: I am just curious what is the best way to do that within current stuff in shogun09:18
@lisitsynso it would be a good way to 'test' api ;)09:18
@lisitsynsounds like usual case09:18
@lisitsynkingsley_: but I have to warn you I'd probably get some time only this evening :)09:20
@lisitsynno idea what's your timezone though09:20
kingsley_I think it would be nearly impossible for you to have less experience with shogun than I, but it seems to me that so called "one-hot encoding" might be either a stand along python function, or be and option to either reading input data like a .csv file, or an option to feature selection or dimensional reduction code.09:21
kingsley_s/and option/an option/09:21
@lisitsynkingsley_: no surely not python function but yeah it would be something easy to do09:23
@lisitsynkingsley_: feature selection is probably good with PCA and lasso09:23
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 245 seconds]09:24
kingsley_It also seems to me that the hard part is reliably knowing when a variable contains categorical data that can safely be split into multiple binary variables. Maybe the presence of alphabetical characters like [a-zA-Z] would be a start, as would checking of purely numeric data has a common distribution. I guess another option would be to just let the user specify which input variables should be recoded.09:25
@lisitsynkingsley_: actually what you describe could be done with quite native R09:25
@lisitsynprobably with even better support for that09:26
@lisitsynI have to admit they do really support categorical stuff better :)09:26
kingsley_Maybe you have forgotten more than I know, but my research into R's data mining packages like caret and rminer suggested to me that the best models they found would remain encoded in R objects, rather than reported as text formulas as I described 8 hours ago, soon after 6.).09:28
@lisitsynkingsley_: hmm should be possible to extract the stuff still09:30
@lisitsynkingsley_: we don't have such functions like reporting in a formula as well anyway09:30
kingsley_But I suppose if there's an R function for just "one-hot encoding", that might be useful.09:30
@lisitsynbut that sounds like a cool idea09:31
@lisitsynkingsley_: it works directly on categorical features - I don't remember how09:31
kingsley_Can shogun try to fit input data with several models, like multiple linear regression, support vector machines and multi-layer perceptrons, and report how well each worked?09:32
kingsley_Or is that the user's job, say, in a python wrapper?09:33
@lisitsynkingsley_: yeah unfortunately it is user job to try all of them09:33
@lisitsynand then evaluate each09:33
@lisitsynevaluation is here though09:33
@lisitsynyou can compute 'answers' and check its auROC or so09:34
@lisitsynbut no high-level thing like you describe (but I'd agree we should have it)09:34
kingsley_So when the user finds a the best model, what format does shogun report that model and its coefficients in?09:35
@lisitsynkingsley_: you can get the weight vector from SVM, get coefficients from MLP and so on09:35
@lisitsynbut no pretty format way again09:35
@lisitsynwell if your models are big it would not be really readable :)09:36
kingsley_lisitsyn: Everything you wrote makes sense, and I am grateful for your time and attention.09:37
@lisitsynkingsley_: you're welcome09:38
@lisitsynkingsley_: I am putting some tasks on github so that we can fix that09:38
kingsley_I will only humbly add that formatting the best model and coefficients as a text formula would have a subtle advantage, even for big models.09:38
kingsley_The answer could be more easily migrated to other applications.09:39
@lisitsynkingsley_: yeah I am going to add a task so someone will add a way to do that09:39
kingsley_I said it before.09:41
kingsley_I'll say it again.09:41
@lisitsynkingsley_: can I refer your github profile if there is any?09:42
kingsley_Sorry, I'm not on github.09:42
* kingsley_ wonders if python-sklearn has any useful code.09:43
@lisitsynok no problem09:43
@lisitsynkingsley_: oh they did09:43
@lisitsynsorry do09:43
@lisitsynit is pretty easy I admit09:43
@lisitsyn ok that's the issue for human readable stuff09:44
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 250 seconds]09:45
kingsley_2728 looks good to me.09:46
kingsley_Can shogun normalize input data so all input variables have similar means and standard deviations?09:48
@lambdaylisitsyn: good for feature selection stuffs as well09:48
@lambdayalthough not always the features will be having meaningful names..09:49
@lambdaywe can default the names to feat0....featN then :D09:49
@lisitsynlambday: yeah they usually don't have names09:50
@lisitsynkingsley_: yes09:50
@lambdaylisitsyn: and you're thinking of get_feature_by_name() as well?09:50
kingsley_lambday: Yes, maybe if the user wants features to have meaningful names, he or she can add a row of meaningful variable names (ie: a header) to the top of an input .csv file.09:51
@lisitsynlambday: yes probably09:51
@lambdaylisitsyn: would be cool if we can read those names in data.names file which is there in most of the cases09:51
@lisitsynlambday: probably we should have a CFeature class09:52
@lambdaynice idea09:52
@lisitsynlambday: and we should merge labels and features :)09:52
@lisitsynit is one entity09:52
@lambdaylisitsyn: sometimes that's really useful :D09:53
@lambdaylisitsyn: just one matrix stores it all.. the last column is for labels09:53
@lambdayfeatures will then just be a block...09:54
@lisitsynlambday: in statistics they really can predict any column09:54
@lambdaylisitsyn: yeah... just change the target column will do the trick09:54
-!- kshitijk [~kshitijk@] has joined #shogun09:54
kingsley_If I might humbly add another reason to output text models like ...09:57
kingsley_variable1 = 1.23 * variable2 + 4.56 * variable3 + 7.89 * variable409:57
@lambdaykingsley_: you want to *print* this linear model09:57
kingsley_it would be that symbolic math programs like "mathomatic" can manipulate them symbolically, which includes factoring, differentiation, and solving.09:58
@lisitsynx = sigm(0, max(0,2*y+z) + max(0,6*z-y) + max(0,9*z-14y))09:58
kingsley_lambday: I happened to notice that you wrote ... "kingsley_: you want to *print* this linear model". I'm not sure I understand you. Please elaborate.09:59
@lambdaykingsley_: I assumed that you're talking about having the model printed in terms of features in order to establish the importance of feature names :)10:00
kingsley_lambday: Yes, you assumed right.10:01
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 264 seconds]10:02
kingsley_But to be fair, I can imagine including feature names in models having several benefits:10:05
kingsley_1.) As you say, easily recognizing which features are important.10:06
kingsley_2.) As Sergey wrote in 2728, being able to export the model to a spread sheet.10:07
kingsley_3.) Exporting the model to mathomatic, Latex, shell scripts, C, etc...10:08
kingsley_4.) Maybe more easily revealing hidden structure in the data.10:09
-!- prakhar2511 [~prakhar25@] has joined #shogun10:09
@lisitsynthat's pretty cool10:09
@lisitsynkingsley_: we definitely will add that10:09
@lisitsynat least for linear classifiers10:10
kingsley_lisitsyn: Thank you for your kind words and willingness.10:10
@lisitsynthanks for the idea ;)10:11
kingsley_You're welcome.10:11
kingsley_lisitsyn: New topic:10:12
kingsley_Are you multi-lingual?10:12
@lisitsynkingsley_: multi-lingual like? :)10:12
@lisitsynenglish is not my native if you mean it10:13
kingsley_Yes, I was wondering if I was chatting with some smart enough to understand more than one human language AND technology AND who is very polite.10:14
kingsley_These are all fine qualities.10:14
@lisitsynhaha :) kingsley_ same for lambday10:14
kingsley_I like smart people.10:15
@lisitsynyeah they are better :)10:15
@lisitsynkingsley_: shogun is developed by guys from germany(sonney2k), russia (me), india (lambday) and a few more countries10:17
* kingsley_ looks at scikit-learn's documentation to see if it has some code that may format models...10:17
@lisitsynafaik they don't have any kind of formatting10:17
@lisitsynat least one point we can take advantage :)10:18
kingsley_OK, /me looks at mlpy's documentation to see if it has some code that may format models...10:19
@lisitsynI think pandas has something on that10:20
@lisitsynlambday: haha take a look my vim got crazy and I tried to compile 'Pools.Dequeue(&&&&&&&&removedPool);'10:21
@lisitsynthat would be C++205110:21
-!- shogun-notifier- [] has quit [Quit: transmission timeout]10:24
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 264 seconds]10:27
-!- vortex_ape [~vortex_ap@] has joined #shogun10:28
shogun-buildbotbuild #538 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: Kunal Arora <>, Soumyajit De <>10:28
-!- artem_ [938ecf38@gateway/web/freenode/ip.] has joined #shogun10:40
-!- hargup_ [uid10222@gateway/web/] has joined #shogun10:41
-!- artem_ [938ecf38@gateway/web/freenode/ip.] has quit [Client Quit]10:42
@lambdaylisitsyn: what were you trying to do!!!10:43
@lisitsynlambday: it should have been just one&10:43
@lambdaylisitsyn: lol... terminator vim10:44
@lambdaykingsley_: where are you from?10:44
kingsley_I'm near the Pacific Ocean and the border of the U.S. and Canada.10:46
kingsley_My web site is
@lambdaykingsley_: oh.. near Seattle/Vancouver?10:47
kingsley_Where are you?10:47
@lambdayI'm from India.. at Bangalore presently.. was in Mumbai until a few weeks back10:48
@lisitsynkingsley_: you've got some interesting content on your website :)10:50
kingsley_lisitsyn: Thank you.10:50
@lambdaykingsley_: I like the diversity of topics you have there :)10:50
kingsley_If you happen to have the time, and are so inclined, I'd like to know what interested you.10:51
@lambdaykingsley_: sergey here has a knack for many other topics has well :)10:51
@lisitsynkingsley_: yeah main thing is diversity10:51
kingsley_OK, fair enough.10:51
@lisitsynkingsley_: and apparently anti war is a good thing I support for sure10:52
@lambdaykingsley_: me and lisitsyn were discussing something about philosophy the other day\10:52
kingsley_lisitsyn: Yes, I agree.10:52
@lisitsyneven though I am in quite aggressive country :)10:52
@lambdaylisitsyn: but we can't thank you guys enough for vodka :D10:53
kingsley_I admire how much respect, peace and prosperity the Swiss have enjoyed since they adopted a neutral foreign policy 500 years ago.10:54
@lisitsynkingsley_: yeah I'd better be swiss10:55
@lisitsynwould not have been ashamed :)10:55
@lambdayI might be out of my ways here, but isn't that where all the black money goes because of that policy?10:57
@lisitsynhah that's true10:57
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 245 seconds]10:57
@lambdaylisitsyn: Indian politicians love swiss banks!10:57
@lisitsynlambday: %s/Indian/Russian10:58
@lisitsynit is general policy10:58
@lisitsynin 'developing' countries10:58
kingsley_lisitsyn: With your intellect, and if you haven't done so already, maybe you could learn a Swiss language like German or French, and get a tech job there.10:58
@lisitsynyou get money you put it to swiss bank10:59
@lisitsynkingsley_: yeah probably at some point I will emmigrate10:59
-!- shogun-notifier- [] has joined #shogun11:00
shogun-notifier-shogun-web: Kevin Hughes :master * f148611 / static/css/styles.css,templates/home.html:
shogun-notifier-shogun-web: add gsoc card to the home page11:00
shogun-notifier-shogun-web: Fernando Iglesias :master * e04d9c6 / static/css/styles.css,templates/home.html:
shogun-notifier-shogun-web: Merge pull request #65 from shogun-toolbox/gsoc101511:00
shogun-notifier-shogun-web: add gsoc card to the home page11:00
kingsley_New topic:11:01
kingsley_SymPy can pretty print its output.11:01
@lisitsynah yeah sure11:01
@lisitsynsympy is for that11:01
kingsley_I wonder if it might have some useful code for 2728.11:01
@lisitsynkingsley_: hardly as our main code is in C++11:02
@lisitsynbut maybe we can integrate somehow11:02
kingsley_OK, that makes sense.11:03
kingsley_lambday and especially lisitsyn: It has been a pleasure chatting with you.11:05
@lisitsynalright :)11:05
@lambdaykingsley_: likewise :) thanks for sharing your ideas..11:06
kingsley_Thanks again for your informed and very polite thoughts, especially in English.11:06
kingsley_At the risk of appearing rude, I need to go.11:06
@lisitsynkingsley_: hah I guess it is quite late already in your timezon11:06
kingsley_However, I'll be thinking about your advice.11:06
@lisitsynsee you11:06
@lambdaysee you :)11:06
@lambdaylisitsyn: you ever used weblogic server?11:07
@lisitsynlambday: yeah probably11:08
@lisitsynbut not really configured it11:08
@lisitsynit worked already :)11:08
@lambdaylisitsyn: haha :D.. I'm getting MaxPermGen error.. it doesn't respect global setups!11:09
-!- Utsav [73f91219@gateway/web/freenode/ip.] has joined #shogun11:13
-!- Utsav [73f91219@gateway/web/freenode/ip.] has quit [Client Quit]11:14
-!- Utsav [73f91219@gateway/web/freenode/ip.] has joined #shogun11:16
UtsavHi, I am a Cmputer Science undergrad from India. I wanted to know if Shogun is participating in this year's Google Summer of Code ?11:16
@lambdayUtsav: hi! we have applied as an org this year as well :)11:17
Utsavlambday : I have a couple of ideas in my mind. I just had a look at the previous year's projects11:18
@lambdaylisitsyn: did you happen to have a look at the linalg ideas page?11:18
@lisitsynlambday: not yet11:18
@lisitsynlambday: oh11:18
@lisitsynI forgot two things11:18
@lisitsynznc and mentor11:18
@lisitsynare you already a mentor?11:18
@lambdayUtsav: that's great! :) I'd suggest that you say hi to the mailing list and share your ideas :)11:18
@lambdaylisitsyn: nah11:19
@lambdaylisitsyn: I sent you a mail :D11:19
UtsavI am sorry if I go wrong in communicating. I am using IRC for the first time so I am not too sure about the IRC etiquettes !11:19
@lisitsynoh haha that's very polite day11:20
@lambdayUtsav: don't worry :)11:20
@lambdaylisitsyn: indeed :)11:20
@lambdayUtsav: I think it is a good idea to subscribe to our mailing list11:20
@lambdayUtsav: we'd be more than happy to hear your ideas11:21
Utsavlamday: Thanks ! Do I need  to use the mailing list ( to broadcast ideas .. ?11:21
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 264 seconds]11:21
@lambdayUtsav: yeah.. the more people hear your ideas the better :)11:21
@lisitsynUtsav: there is no real etiquette apart from not tell us something really rude :)11:21
Utsavlambday: I will make sure that never happens ! :) Are you a mentor this time around ?11:22
@lisitsynboth of us are11:22
-!- exploreshaifali [~shaifali@] has quit [Quit: Leaving]11:22
UtsavGreat ! Can I enquire about a couple of ideas I have in my mind ?11:23
Utsavright here I mean ..11:23
@lambdayUtsav: absolutely11:24
UtsavI checked out Parijat's project that he did for Shogun - involved coding up data mining algorithms like ID3 and Random Forests ..11:24
UtsavHow about doing it the Fuzzy way ?11:25
@lisitsynlike more smooth predictions?11:25
@lambdaylisitsyn: Fernando is doing fundamental ML 2 right?11:25
@lisitsynlambday: yes11:25
UtsavFuzzy Logic gives a more intuitive way of thinking .. much more understandable to humans and better error tolerance11:26
UtsavThe decision at each node in a decision tree would involve fuzzy logic and the splitting attribute would also be selected based on fuzzy information gain11:27
UtsavI was just giving an example how it could work out11:27
@lisitsynsure, are there any well established algorithms?11:27
UtsavA lot of research papers have been and are being published in this field11:27
@lambday^ I was going to ask11:28
@lisitsynUtsav: actually Heiko planned to propose a project on bayesian trees11:28
UtsavFuzzy Decision Trees do return a lot of results when googled ... and I was studying a few lately11:28
@lisitsynthat's something similar but in bayesian fashion11:28
UtsavMine involves a good use of Fuzzy Logic ... with a variety of implementations of different member ship functions that fuzzy logic has to offer11:29
UtsavSo  I was wondering if it sounds good enough to be contemplated upon ..11:30
@lisitsynUtsav: sounds good but you'd have to talk to these guys:
@lisitsynthey have something on mind already11:30
UtsavI just see Kalman Filter there ... I have already done a project on Kalman Filters :D11:31
UtsavNothing too advanced though11:31
UtsavBut helped me understand how it works11:31
@lisitsynyeah they are not on fuzzy but probabilistic11:32
@lisitsynI can't really compare them but probabilistic sounds like more theoretically based11:32
Utsavindeed ... I don't see fuzzy logic involved11:32
@lisitsynwhile fuzzy is more about some heuristics11:32
@lambdayUtsav: could you briefly explain how fuzzy works for decision trees?11:33
Utsavsure ...11:34
@lambdayUtsav: or point me to some paper that you've read on this topic11:34
Utsavmost certainly ... I was going to suggest that11:35
Utsavone sec ..11:35
UtsavI guess this one would give a good explaination11:36
@lisitsyn29 citations ;)11:37
@lisitsynin google scholar it is 199 though11:38
UtsavIs it good ? (Not much experience with publishing papers ! )11:38
@lisitsyn199 is pretty popular yeah11:39
@lisitsynsvm is cited by 16597 papers though11:39
Utsavthat is a huge number11:39
@lisitsynand CART is cited by 26911 papers11:40
UtsavAlso ... an extension of fuzzy decision trees would be fuzzy random forests11:40
@lisitsynyeah quite natural extension11:41
@lambdayUtsav: I would strongly suggest that you share your ideas to a mail to fernando and heiko, cc-ing the mailing list11:42
UtsavAnd this paper gave explicit membership values to variables ... which could be replaced with different types of membership functions11:42
@lambdayorganize your thoughts and see if you can accommodate these ideas alongwith fundamental ML 2 project11:42
@lambdaycan't access the pdf.. :(11:42
@lambdaygotta ask my insti friends11:42
UtsavSo you feel that this idea has some scope to be taken ahead ( a small chance maybe) ?11:43
UtsavI would definitely try to organise it11:43
@lambdayUtsav: yeah but fernando and heiko are the guys to talk for that :)11:44
UtsavLooking forward to it .. :)11:44
UtsavAnd, my programming experience is not very advanced - sort of intermediate ... is that ok ?11:45
UtsavI am willing to put in an effort to learn as much as I can though :)11:45
@lambdayUtsav: that's good! we expect some fluency with C++ though.. some design patterns.. but as long as you can pick things up quickly on your way its cool11:46
@lambdayalso, it would be worth to get familiar with shogun codebase as well11:47
UtsavHow do I do that ?11:47
@lambdayUtsav: best way is to start with make/install/run shogun.. run the examples.. next is to aim at issues on github..11:48
Utsavok .. will try to do that ! Thanks a lot11:49
@lambdayonce you think you have solved some bug/enhancement, send a pull request11:49
@lambdayyw :)11:50
-!- Utsav [73f91219@gateway/web/freenode/ip.] has left #shogun []11:54
-!- AS [938ecf38@gateway/web/freenode/ip.] has joined #shogun12:00
AShi all, I'm trying to install & run Shogun. Cmake, make, make install -- all seemingly finish with no errors. However, many of the tests fail (especially, python_modular-related but not all of them) and simple python scripts that use Shogun object are segfault-ing. Anyone has ideas?12:06
ASlogs: cmake , make , make install , ctest
@lambdayAS:  I'm afraid you haven't cloned shogun data12:13
ASYou mean I should run ctest from the data directory?12:14
@lambdayAS: no that should be run from the build directory only12:15
@lambdayAS: just check if you have the data in shogun/data12:16
-!- thoralf [~thoralf@] has joined #shogun12:18
-!- mode/#shogun [+o thoralf] by ChanServ12:18
ASjust tried putting the data folder from the data tarball into shogun-install/data and into shogun-4.0.0/data -- the same result12:18
@thoralfHey lovely people :)12:19
@thoralfbesser82: Around?12:19
ASok, will try cloning from git (previously tried only downloading tar.gz)12:19
@lambdayAS: yeah git is better12:21
@thoralfHay lambday12:21
@lambdaythoralf: hey! long time12:21
@lambdaythoralf: how are you doing?12:22
@thoralflambday: I'm fine.  Just realized GSoC is getting hot now :)12:22
@lambdaythoralf: haha yeah! its that time our commits/month ratio overshoots12:24
ASlambday: well, the git version does not even compile12:25
ASshogun/tests/unit/mathematics/linalg/ In member function 'virtual void AddMatrix_native_backend_Test::TestBody()':12:25
ASshogun/tests/unit/mathematics/linalg/ error: 'add' is not a member of 'shogun::linalg'12:25
@lambdayAS: really! let me check12:26
@lambdayAS: src looks fine! maybe I am missing something. don't have a machine right now where I can experiment. Will be able to test only after an hour12:29
@lambdayAS: meanwhile, you can just link the shogun/data to the unzipped data directory you have..12:29
@lambdayfor shogun-4.0.012:29
ASlambday: yes, just did that -- same errors12:29
@lambdayAS: hang on. will get back soon!12:30
ASlambday: ok! just thoughts: ok, even if that solves the issue with tests, the python code I'm trying to follow (the FGM notebook: is still giving me segmentation fault once I try to create some shogun object (specifically on cell 9)12:33
@lambdayAS: ah I figured it out. Will send a patch as soon as I get my machine :)12:33
ASlambday: thanks! looking forward:)12:34
-!- vortex_ape [~vortex_ap@] has joined #shogun12:35
@lambdayAS: this one?? samples, labels = prepare_data(p_tr, l_tr, ftype_all, n_tr_samples)12:35
ASlambday: soory, the previous cell. Actually I traced it down to "fac_type_u = TableFactorType(0, cards_u, w_gt_u)". Once it's created, even exiting from the script results in segfault12:37
@lambdayAS: ah that's bad. maybe these things depend on some dependencies you don't have installed?12:40
ASlambday: That's what thought too, but cmake did not complain.12:41
ASlambday: only a strange line "-- The following REQUIRED packages have not been found:   * OpenCL"12:42
@lambdayAS: nah that's not it12:43
@lambdayAS: can you please try running this instead?12:44
ASlambday: hang on12:45
ASlambday: got the same line "Segmentation fault"12:49
ASlambday: could be because of some python version mismatch? the environment here is kind of a mixture of 2.6 and 2.712:50
@lambdayAS: multiple versions of python can be there in the same system.. shouldn't be a problem12:51
@lambdayAS: I assume that it must be some installation issue. cause I see buildbot happy here
@lambdayAS: please note that many of our methods currently requires Eigen3 library12:52
@lambdaywithout that, many methods will be unavailable12:52
@lambdayso Id' try installing eigen3-devel and then freshly make shogun12:52
ASok, will try. thank you for your help!12:52
@lambdayAS: yw12:53
-!- abinashm [uid66179@gateway/web/] has joined #shogun12:57
-!- vortex_ape [~vortex_ap@] has quit [Quit: Leaving]13:02
@thoralfSee you guys. :)13:03
-!- thoralf [~thoralf@] has quit [Quit: Konversation terminated!]13:03
* lambday brb13:07
@lisitsynlambday: would you like to move arpack wrapper to the linalg?13:09
@lisitsynas well as FEAST13:09
@lisitsynI'd like someone to write FEAST wrapper13:09
-!- lambday [c40f106a@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]13:11
-!- exploreshaifali [~shaifali@] has joined #shogun13:11
-!- abinashm [uid66179@gateway/web/] has left #shogun []13:18
-!- prakhar2511 [~prakhar25@] has joined #shogun13:43
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has joined #shogun13:58
-!- mode/#shogun [+o lambday] by ChanServ13:58
* lambday Re13:59
-!- shogun-notifier- [] has quit [Quit: transmission timeout]14:00
ASlambday: hi again. just to let you know -- i have tried with eigen3 v. 3.2.4 (both bundled and not bundled) and got the same segfault again :(14:13
-!- HeikoS [] has joined #shogun14:22
-!- mode/#shogun [+o HeikoS] by ChanServ14:22
-!- exploreshaifali [~shaifali@] has quit [Quit: Leaving]14:31
ASlambday: Ah yeah, all this still not working with the latest git version (btw, thanks for fixing!)14:35
@lambdayAS: hey! I still haven't sent the patch yet14:36
@lambdayAS: I'll ping you once I do14:36
@lambdayman when I try cmaking shogun without eigen3, a whole lot of demons come out! gotta fix quite a few places14:38
@lambdayHeikoS: there?14:39
@HeikoSlambday:  yes14:39
ASlambday: ok, that is strange:) anyway the "'add' is not a member of 'shogun::linalg'"-kind of errors are gone somehow:)14:40
@lambdayAS: yeah once you install eigen3 it will be gone14:41
-!- sourabh13 [ca4eafc7@gateway/web/freenode/ip.] has joined #shogun14:41
@lambdayHeikoS: LinearARDKernel and GaussianARDKernel is giving some errors in absence of linalg lib.. fixing14:42
@HeikoSlambday: ah yeah, my bad for merging that14:42
@HeikoSlambday: thanks14:42
@HeikoSlambday: could you reference wu in the patch? so that he know for the future?14:42
@lambdayHeikoS: we should have these tests in travis only14:42
@HeikoSlambday: what do you mena?14:42
@lambdayHeikoS: absolutely14:42
@lambdayHeikoS: in travis these parts don't even compile.. so we won't get to know from travis if there are some errors14:43
@lambdayHeikoS: in buildbots I guess we have eigen3 everywhere14:43
@HeikoSI see14:43
@HeikoSso you want a build where eigen is disabled?14:43
@lambdayHeikoS: yeah - only the *REQUIRED* packages should be there14:43
@HeikoSI guess wiking is the one to ask14:43
@lambdayso that we know that our builds are sane even withtout the optional pkgs14:44
@HeikoSI agree we should have a check for that14:44
@HeikoSlambday: could you write a request to shogun-team and cc wiking?14:44
@lambdayHeikoS: absolutely14:44
@lambdaywill do that after the fix14:45
@HeikoSlambday: thanks for taking care!14:45
@lambdayHeikoS: I need to rename eigen3 and viennacl to something tmp in order to prevent cmake from finding them :D do you know a better way?14:45
@lambdaycmake should have something like -DDisableEigen3 -DDisableViennaCL14:46
@HeikoSlambday: I agree that should be a switch14:46
@HeikoSlambday: cant be too hard to add14:46
@HeikoSI think I added something like that recently14:46
@HeikoSfor a guy who wanted to disable something14:47
@HeikoSthat was found on his system14:47
@HeikoSah yeah14:47
@lambdayHeikoS: oh so what's the switch?14:47
@HeikoSlet me check14:47
@HeikoSsomething similar to that14:48
@lambdayHeikoS: great! that will be helpful for eigen3 and viennacl as well14:50
@lambdayHeikoS: I'll add those14:50
@lambdayHeikoS: thanks :)14:50
-!- hargup_ [uid10222@gateway/web/] has quit [Quit: Connection closed for inactivity]14:51
-!- attia42 [83f6c22c@gateway/web/freenode/ip.] has joined #shogun14:52
-!- sourabh13 [ca4eafc7@gateway/web/freenode/ip.] has quit [Quit: Page closed]14:52
-!- sourabh13 [ca4eafc7@gateway/web/freenode/ip.] has joined #shogun14:52
-!- sourabh13 [ca4eafc7@gateway/web/freenode/ip.] has quit [Client Quit]14:55
attia42hello, I have been trying LMNN and it doesn't seem to use multi threads while training. (it uses multi threads at first but then not, guessing it's only during  pairwise distances calculations). is it possible to make it use more threads? any references other than the  LMNN ipython tutorial ?14:57
@HeikoSattia42: I think the main method is single thread14:57
@HeikoSattia42: write to iglesias (Fernando, who wrote the code) to ask him on that, cc the mailing list14:58
@HeikoSattia42: all Shogun distance methods use multiple threads14:58
@HeikoSattia42: I am not sure whether its possible to parallelise LMNN easily even? Isnt the solver iterative?14:58
attia42@HeikoS iterative gradient based updates HeikoS15:01
attia42@HeikoS Thanks will try to contact him15:01
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 246 seconds]15:03
@HeikoSattia42: so can one even parallelise these gradient updates from within?15:04
-!- abinashm_ [cb6ef315@gateway/web/freenode/ip.] has joined #shogun15:08
attia42HeikoS: I suppose yes, there should be a way to parallelise it15:12
@HeikoSattia42: ok, best liase with fernando15:12
@lambdayAS: sent the fix for linalg things
@lambdayAS: you can pull the latest once I merge it (after travis is finished I'll merge)15:18
ASlamdbay: Thanks, will try it! Meanwhile I tried with ArchLinux packaged shogun on a different machine, again no complains about missing dependencies. But the same segmentation fault on :(15:20
@HeikoSlambday: hey15:33
@HeikoSas you are also a superb software engineer, and you know more and more about shogun, would you be interested in being a co-mentor with thoralf15:33
@HeikoSif so, add yourself and your ideas15:34
@HeikoSthoralf might be unavailable during the second half of GSoC, so we need someone to take over15:34
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]15:35
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has joined #shogun15:36
-!- mode/#shogun [+o lambday] by ChanServ15:36
@lambdayHeikoS: hey! sorry I got disconnected15:36
@lambdayHeikoS: checking your mail now15:36
@lambdayHeikoS: reading the project desc15:37
@HeikoSgoing to a talk now, so see you later15:37
@HeikoSlambday: have a look whether this is interesting for you15:37
@HeikoSjust though you could bring in all your cool ideas in there15:37
@lambdayHeikoS: sure :)15:37
@HeikoSwas really impressed by the linalg design15:37
@HeikoSso we need more of that :)15:37
@HeikoSsee you later15:37
@lambdayHeikoS: hehe :D alright see you :)15:37
@lambdaylisitsyn: there??15:37
-!- HeikoS [] has quit [Quit: Leaving.]15:37
@lambdaylisitsyn: you're not floating the plugin project this year?15:38
@lambdaylisitsyn: or we're saving that one for ourselves :D15:38
@lisitsynlambday: hmm15:38
@lisitsynlambday: I am a bit swamped so didn't have time to write it up15:38
@lisitsynprobably we can integrate it in some project15:39
@lisitsynthoralf's detox or other15:39
@lambdaylisitsyn: but that itself will be a huge one15:39
@lambdaydesign is the most imp there15:39
@lisitsynlambday: then may be doing it by ourselves is better15:40
@lambdaymost challenging part is to be cross-platform15:40
@lambdayas shogun should work fine on windows as well15:40
@lambdaylisitsyn: I've got a nice tutorial on dr. dobbs few days back.. will discuss once I read it.. maybe it contains some ideas on how to do that15:41
-!- kshitijk [~kshitijk@] has joined #shogun15:52
-!- abinashm_ [cb6ef315@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]15:58
-!- vortex_ape [~vortex_ap@] has joined #shogun16:03
-!- prakhar2511 [~prakhar25@] has joined #shogun16:05
-!- shogun-notifier- [] has joined #shogun16:15
shogun-notifier-shogun: lambday :develop * 1c79aec / src/shogun/ (9 files):
shogun-notifier-shogun: fix issue #272916:15
shogun-notifier-shogun: Soumyajit De :develop * 2e45f03 / src/shogun/ (9 files):
shogun-notifier-shogun: Merge pull request #2731 from lambday/develop16:15
shogun-notifier-shogun: Fix for issue #272916:15
@lambdayAS: merged. will check your issue soon!16:18
@lambdayAS: could you please try to make the git one now? [develop branch]16:19
shogun-buildbotbuild #2654 of bsd1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: lambday <>, Soumyajit De <>16:21
shogun-buildbotbuild #700 of osx1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: lambday <>, Soumyajit De <>16:23
shogun-buildbotbuild #986 of FCRH - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>, Soumyajit De <>16:26
* lambday afk16:30
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 245 seconds]17:02
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 246 seconds]17:18
-!- hargup_ [uid10222@gateway/web/] has joined #shogun17:22
ASlambday: Cool, thank you very much, now with Eigen3 the tests pass! The runs as well. Need test it more, but it looks great so far!17:24
-!- AS [938ecf38@gateway/web/freenode/ip.] has left #shogun []17:25
-!- prakhar_ [~prakhar25@] has joined #shogun17:28
-!- prakhar_ [~prakhar25@] has quit [Ping timeout: 255 seconds]17:40
-!- prakhar_ [~prakhar25@] has joined #shogun17:43
-!- prakhar_ [~prakhar25@] has quit [Ping timeout: 245 seconds]17:49
@lambdaylisitsyn: have you watched Rusev fighting?17:55
@lambdayRef. WWE17:56
@lisitsynlambday: nope17:56
@lambdaylisitsyn: in WWE world, he's the Russian hero!17:56
@lambdaytaking on everyone that dares to come in his way17:56
@lisitsynno idea17:56
@lambdayI love him!17:57
@lambdayah that didn't come out right!17:57
@lisitsynhaha he is bulgarian17:58
@lambdaylisitsyn: err I just noticed now!17:58
@lambdaylisitsyn: WWE is such a fake crap17:58
@lambdaylisitsyn: still at office?17:59
@lisitsynlambday: I am now drawing a new logo in gimp18:01
@lambdaylisitsyn: show me :D maybe my "artistic skills" can offer something :D18:01
@lambdaylogo for your own startup? :P18:01
-!- vortex_ape [~vortex_ap@] has joined #shogun18:04
@lambdaylisitsyn: we need your vote -
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 264 seconds]18:05
@lisitsynlambday: sorry leaving now18:07
@lisitsynlambday: take a look at org18:07
@lisitsynI put some better logo18:08
@lisitsynstill ugly though18:08
@lambdaylisitsyn: still the old one18:08
@lambdaylisitsyn: anyway go home man! it's Friday :D18:09
@lisitsynhmm no18:09
@lisitsynshould be new18:09
@lisitsynin gsoc18:09
@lambdaywhat changed?18:09
@lisitsynlets see if I get home faster than deadline comes18:10
@lisitsynso that I can update it once more18:10
@lambdaylisitsyn: can't find org page :(18:10
@lambdaylisitsyn: hehe.. run forrest run18:10
@lambdaylisitsyn: I can't see I guess.. can't find the link anywhere18:11
-!- HeikoS [] has joined #shogun18:25
-!- mode/#shogun [+o HeikoS] by ChanServ18:25
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]18:32
-!- prakhar_ [~prakhar25@] has joined #shogun18:40
-!- attia42 [83f6c22c@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]18:46
-!- prakhar_ [~prakhar25@] has quit [Ping timeout: 256 seconds]18:47
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has joined #shogun18:47
-!- mode/#shogun [+o lambday] by ChanServ18:47
-!- kshitijk [~kshitijk@] has joined #shogun18:49
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 252 seconds]19:03
-!- lambday [6a33f0a8@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]19:04
-!- HeikoS [] has quit [Ping timeout: 256 seconds]19:15
-!- shogun-notifier- [] has quit [Quit: transmission timeout]19:15
-!- mindBlasting [~Nitin@] has joined #shogun19:15
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 252 seconds]19:21
shogun-buildbotbuild #539 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: lambday <>, Soumyajit De <>19:22
-!- PirosB3 [] has joined #shogun19:23
-!- PirosB3 [] has quit [Quit: PirosB3]19:32
-!- sonne|osx [] has joined #shogun19:47
-!- kshitijk [~kshitijk@] has joined #shogun19:59
-!- prakhar_ [~prakhar25@] has joined #shogun20:18
-!- prakhar_ [~prakhar25@] has quit [Client Quit]20:19
-!- prakhar2511 [~prakhar25@] has joined #shogun20:20
-!- exploreshaifali [~shaifali@] has joined #shogun20:34
-!- prakhar2511 [~prakhar25@] has quit [Ping timeout: 240 seconds]20:51
-!- prakhar2511 [~prakhar25@] has joined #shogun20:53
-!- exploreshaifali [~shaifali@] has quit [Quit: Leaving]21:26
-!- hargup_ [uid10222@gateway/web/] has quit [Quit: Connection closed for inactivity]21:31
-!- sonne|osx [] has quit [Quit: sonne|osx]21:49
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 246 seconds]22:03
-!- vortex_ape [~vortex_ap@] has joined #shogun22:03
-!- vortex_ape [~vortex_ap@] has quit [Ping timeout: 256 seconds]22:09
-!- kshitijk [~kshitijk@] has quit [Ping timeout: 252 seconds]22:24
-!- mindBlasting [~Nitin@] has quit [Quit: Leaving]23:24
--- Log closed Sat Feb 21 00:00:29 2015