Open in new window / Try shogun cloud
--- Log opened Thu Jan 29 00:00:56 2015
-!- samkit [~samkit@] has quit [Ping timeout: 265 seconds]00:25
-!- exploreshaifali [~shaifali@] has quit [Quit: Leaving]00:29
-!- txomon|home [~txomon@unaffiliated/txomon] has quit [Ping timeout: 256 seconds]01:30
-!- pickle27 [] has joined #shogun03:06
-!- PirosB3 [] has quit [Quit: PirosB3]03:59
-!- pickle27 [] has quit [Remote host closed the connection]04:38
-!- curiousguy13 [~curiousgu@] has quit [Ping timeout: 252 seconds]05:33
-!- curiousguy13 [~curiousgu@] has joined #shogun05:45
-!- samkit [~samkit@] has joined #shogun06:16
-!- lambday [c40f1065@gateway/web/freenode/ip.] has joined #shogun06:25
-!- samkit [~samkit@] has quit [Read error: Connection reset by peer]07:10
-!- besser82 [~besser82@2a02:8108:8800:3d00:f2de:f1ff:fe89:42d4] has joined #shogun08:28
-!- besser82 [~besser82@2a02:8108:8800:3d00:f2de:f1ff:fe89:42d4] has quit [Changing host]08:28
-!- besser82 [~besser82@fedora/besser82] has joined #shogun08:28
-!- mode/#shogun [+o besser82] by ChanServ08:28
-!- HeikoS [] has joined #shogun10:07
-!- mode/#shogun [+o HeikoS] by ChanServ10:07
-!- HeikoS [] has quit [Quit: Leaving.]10:22
-!- PirosB3 [] has joined #shogun10:22
-!- HeikoS [] has joined #shogun10:39
-!- mode/#shogun [+o HeikoS] by ChanServ10:39
-!- curiousguy13 [~curiousgu@] has quit [Ping timeout: 256 seconds]11:33
-!- curiousguy13 [~curiousgu@] has joined #shogun11:45
@lisitsynHeikoS: wiking: anyone knows how to fix the build?11:53
@lisitsyntravis one11:53
@lisitsynits stopping 3 of my PRs :D11:53
lambdayand one of mine :/12:02
@lisitsynlambday: oh hey12:06
lambdaylisitsyn: yo12:07
@lisitsynlambday: lets make something cool12:07
lambdaylisitsyn: ah finally, you're back! :D12:07
@lisitsyntry to12:07
lambdaylisitsyn: I'm free now a days - getting fat! could use some brain work :P12:07
lambdaylisitsyn: let's finalize things on aer12:08
@lisitsynlambday: what about porting some ideas into shogun?12:08
@lisitsynone by one12:08
@lisitsynsay get/set by string/tag12:08
@lisitsynthen load library et12:08
lambdaylisitsyn: I think we better start experimenting on a separate branch12:08
@lisitsynhmm yeah probably12:09
* lambday afk12:09
-!- PirosB3 [] has quit [Quit: PirosB3]12:15
-!- dhfromkorea [] has joined #shogun12:21
lambdaylisitsyn: will be back in an hour.. let's discuss then12:22
lambdaylisitsyn: and create issues, whatever plans you're having..12:23
@lisitsynlambday: I am a bit slow, the workload is quite heavy last days12:24
@lisitsynjust do whatever comes into my mind12:25
-!- lambday [c40f1065@gateway/web/freenode/ip.] has quit [Ping timeout: 246 seconds]12:27
-!- txomon|home [~txomon@unaffiliated/txomon] has joined #shogun12:31
@besser82wiking, yo?12:32
@besser82wiking, can you confirm python_modular is not segfaulting anymore with SWIG 3.0.4?12:33
-!- dhfromkorea [] has quit [Remote host closed the connection]12:35
-!- anishshah_ [uid50585@gateway/web/] has joined #shogun12:38
-!- anishshah_ is now known as anishshah12:38
-!- dhfromkorea [] has joined #shogun12:44
-!- dhfromko_ [] has joined #shogun12:48
-!- txomon|home [~txomon@unaffiliated/txomon] has quit [Ping timeout: 276 seconds]12:52
-!- dhfromkorea [] has quit [Ping timeout: 240 seconds]12:52
-!- HeikoS [] has quit [Quit: Leaving.]14:05
-!- exploreshaifali [~shaifali@] has joined #shogun14:28
-!- dhfromko_ [] has quit [Remote host closed the connection]14:54
-!- Floatingman [] has quit [Remote host closed the connection]15:10
@sonney2kbesser82, you are too fast for me :)15:21
@besser82sonney2k, what?15:23
@besser82sonney2k, what did i do?  :P15:24
@sonney2k<besser82> sonne|work, Fedora-builds are running for F20+ :D15:32
@sonney2k^ besser82 I guess that is related to shogun 4.015:33
@besser82sonney2k, yes :D15:33
-!- exploreshaifali [~shaifali@] has quit [Ping timeout: 245 seconds]15:33
-!- ChanServ changed the topic of #shogun to: Shogun Machine Learning Toolbox | SHOGUN 4.0.0 Kose No Maro just released | Devs are usually around between 10am-10pm UTC | When asking questions please be patient or use the mailing list. We always answer but it might take a while. | Channel logs
@besser82sonney2k, currently doing test-builds and pimping the spec-file, to get stuff going into CentOS / RHEL /Scientific Linux {6,7}15:33
-!- HeikoS [] has joined #shogun16:08
-!- mode/#shogun [+o HeikoS] by ChanServ16:08
@sonney2kbesser82, I am far away debian wise for that...16:48
@sonney2kI am only hoping that the debian release is imminent...16:48
@besser82sonney2k, who uses debian anyways?  :P16:50
@sonney2kbesser82, :P16:50
@lisitsynhey guys16:51
@lisitsynsonney2k: about a-z, answered your mail :)16:51
@besser82sonney2k, I think there will be packages going out to FC and EL dists at beginning of next week16:51
@lisitsynthey are all like tokugawa ywhatever16:51
-!- sonne|osx [] has joined #shogun17:12
@HeikoSlisitsyn:  hey17:20
@HeikoSlisitsyn:  could you check this here
@HeikoSabout making CKernel::compute virtual17:20
@HeikoSI wonder whether and how bad that is17:20
@lisitsynlet me check17:21
@HeikoSinline float64_t kernel(int32_t idx_a, int32_t idx_b)17:23
@HeikoSvirtual float64_t kernel(int32_t idx_a, int32_t idx_b)17:23
@lisitsynHeikoS: compute is already virtual?17:23
@HeikoSits "kernel"17:23
@lisitsynHeikoS: performance wise it doesn't matter17:24
@lisitsynwe already call normalize and compute which are virtual17:24
@lisitsynbut the idea was to always check indices17:24
@HeikoSthought so, but its another one17:24
@HeikoSI see17:24
@HeikoSso as long as he does this its fine17:24
@lisitsynHeikoS: what's the issue?17:25
@lisitsynI mean why it has to be virtual?17:25
@HeikoShe want to change the thing17:26
@HeikoSbecause we need a kernel that can compute derivative wrt to some subset of data (inducig points)17:26
@HeikoSGP stuff17:26
@lisitsynHeikoS: its actually good to have kernel non-virtual17:27
@lisitsynyou know it is a good practice to have public methods non virtual17:27
@lisitsynjust protected ones17:27
-!- Floatingman [] has joined #shogun17:28
@HeikoScan you comment on this in the thread?17:28
-!- curiousguy13 [~curiousgu@] has quit [Ping timeout: 255 seconds]17:33
@lisitsynHeikoS: I glanced over thread17:34
@lisitsynstill not really get what's the issue with making it virtual17:34
@HeikoSI just wanted to know *whether* there is an issue17:37
-!- sonne|osx [] has quit [Quit: sonne|osx]17:37
@lisitsynI see17:37
@lisitsynno I mean17:37
@lisitsynwhy it is needed17:37
@lisitsynHeikoS: whats fitc anyways?17:37
@lisitsynI am looking and having no clue what's going on :D17:37
@HeikoSit is a sparse approximation to the likelihood in Gaussian process17:40
@HeikoSits a good read17:40
@HeikoSand thats what wu wants to do17:41
@HeikoSlisitsyn: thanks for the feedback17:42
@lisitsynHeikoS: okay anyway - normalizer::normalize and kernel::compute are virtual17:43
@lisitsynit would matter if no virtual functions involved17:43
-!- curiousguy13 [~curiousgu@] has joined #shogun17:45
-!- exploreshaifali [~shaifali@] has joined #shogun17:48
-!- sonne|osx [] has joined #shogun18:10
-!- sonne|osx [] has quit [Read error: Connection reset by peer]18:12
-!- sonne|osx [] has joined #shogun18:13
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 276 seconds]18:24
@lisitsynHeikoS: how do GP people handle categorial features?18:36
@HeikoSlisitsyn: with kernel functions on categorical variables18:37
@HeikoScalled diffusion kernels18:37
@lisitsynanything else apart from just matches?18:37
@HeikoSno one can define kernels induced by categorical18:37
-!- sonne|osx [] has quit [Quit: sonne|osx]18:37
@HeikoSyou kind of get a gaussian kernel analogue even18:37
@lisitsynHeikoS: is it better than just dot product = number of matched features?18:38
@lisitsynHeikoS: I am trying to beat people at kaggle that's why asking :D18:38
@lisitsynsounds like something smart18:39
@lisitsynHeikoS: what should be done to have it in shogun?18:39
@HeikoScheck page 7 on left bottom18:40
@HeikoSthe counting is a limiting case of this kernel18:40
@HeikoSits like a gaussian kernel using a hamming distance rather than a euclidean18:40
@lisitsynHeikoS: any chance to learn that on 40M examples?18:41
@HeikoSbut possible18:42
@HeikoSI will offer a GSoC project on that in fact18:42
@HeikoSlow rank expansion of kernel (say incomplete cholesky) and then a conjugate gradient for the linear solves18:42
@lisitsynif it makes me win this kaggle I take it :D18:42
@lisitsynHeikoS: will this approx still be better than linear?18:43
@lisitsynHeikoS: I want to take over this thing about REST interface18:44
@lisitsynI think that's the way to win18:44
@lisitsynno mature framework has REST18:44
@HeikoSthe approx is basically doing linear in an approximate feature space18:44
-!- lambday [6a338604@gateway/web/freenode/ip.] has joined #shogun18:45
@lisitsynlambday: I just told HeikoS I am now interested in REST18:45
@lisitsynwanna have some fun with that?18:45
lambdaylisitsyn: REST? as in REST API?18:45
lambdaylisitsyn: I've been working on this only @job18:46
lambdayjersey :P18:46
@lisitsynwe absolutely should have a shogun server18:46
lambdayI don't think anyone ever did that for a ML library :P18:46
@lisitsynah and another thing18:46
@lisitsynwe should have something like MOE/spearmint18:46
@lisitsynthat's the way to go18:47
@lisitsynI am 900% sure18:47
lambdaywhat are these?18:47
@lisitsynmodel selection18:47
@lisitsynvia Heikoian stuff18:47
@HeikoSsergey is that really feasible ;)18:47
lambdayokay.. HeikoS is the guy for that :D18:47
@lisitsynI mean bayesian18:47
@lisitsynHeikoS: at least we can interface18:47
@HeikoSlisitsyn: no need to be bayesian if you have 40M points on a linear regression model18:47
@HeikoSonly if oyu have 40m parameters too18:48
@lisitsynHeikoS: yes18:48
@lisitsynHeikoS: but my linear model sucks18:48
@HeikoSI see18:48
@HeikoSlambday: hey man how are things? :)18:49
@HeikoSlisitsyn: well then18:49
@HeikoSmake a better thing :)18:49
lambdayHeikoS: lazy18:49
@lisitsynoh wanna join and consult me? :)18:49
@lisitsynfair share18:49
@HeikoSgood feautres and well tuned parameters do most18:49
lambdayHeikoS: not too much work to do at job18:49
@HeikoSlisitsyn: haha18:49
@lisitsyncode is ready just tell me what to do :D18:49
@HeikoSlisitsyn: ICML deadline18:49
@HeikoSlisitsyn: maybe, when does that end?18:49
@lisitsyn9 feb18:49
@HeikoSlambday: sounds good18:49
@HeikoSlots of time for shogun18:49
@HeikoSlisitsyn: too early18:50
@HeikoSicml is 6th18:50
lambdayHeikoS: we should get back to the big tests work - once you get free18:50
@HeikoSlisitsyn: try this:
@HeikoSlambday: yeah totally, currently working on submission of the above18:50
@HeikoSthen another thing18:50
@HeikoSand then got time for big testing18:50
@lisitsynHeikoS: yeah so I will read all these smart things till feb 9 of 201618:50
@HeikoSdino is also busy but should be better within a month  or so18:50
lambdayHeikoS: alrighty.. I'll work on the prototype for the implementation I was working earlier18:51
@HeikoSlambday: cool!18:51
@lisitsynokay actually I am like driven by kaggle things18:51
@lisitsynit is a way to improve shogun also18:51
@HeikoSlambday: btw I like the linalg progress18:51
@HeikoSlisitsyn: thats great18:51
@HeikoSI mean there is so much to do18:51
lambdaylisitsyn: you participated?18:51
@HeikoSlisitsyn: I though about having a GSoC project on building a kaggle pipeline18:51
@HeikoSwhat do you think?18:51
@HeikoSfor supervised prediction18:52
@lisitsynlambday: trying to get to the top of avazu ctr :D18:52
lambdaylisitsyn: what's your current position :D18:52
@lisitsynHeikoS: this needs real apriori experience18:52
@lisitsynlambday: 160th18:52
lambdaylisitsyn: not too bad! :D18:52
lambdayHeikoS: linalg has many more things to be included inside..18:53
@HeikoSbtw i recently did a kaggle18:53
@HeikoSgot 9th :)18:53
lambdayHeikoS: good that with the new patch we can accommodate native implementations18:53
* lambday bows18:53
@HeikoSlambday: cool, well done thats important if people have nothing18:53
@lisitsynHeikoS: hmm what are your papers about if you weren't the first :D18:54
lambdayHeikoS: which will hardly be the case :P but still good to have - since we can get rid of all linalg operations from SGMatrix/SGVector18:54
@HeikoSlambday: exactly18:54
@HeikoSlisitsyn: my papers are about buzz words, haha :D18:55
lambdayis any of the paid courses @stanford center of professional development worth the $$?18:56
lambdayI saw a course for some professional certification in large scale ML for $19k18:56
lambdaywho does these courses!!18:56
@HeikoSlambday: Chinese people18:57
@HeikoSthe ML MSc costs like 25k for chinese18:57
@HeikoShere in london18:57
@HeikoSand they are the largest group18:57
lambdayHeikoS: man you must absolutely love Bayes :P new paper is also about that :D18:59
@HeikoSlambday: sure :)18:59
@HeikoSits trendy18:59
lambdayHeikoS: btw is this a good time to merge a new patch, or shall we wait for a few days19:01
@HeikoSlambday: currently also working on infinite exponential family density estimates for hamiltonian dynamic MCMC :)19:01
@HeikoSlambday: why?19:01
lambdayHeikoS: no idea what that means.. know the individual words but don't know what it means when you put 'em all together :D19:02
lambdayHeikoS: regarding release..19:02
@HeikoSlambday: I see19:05
@HeikoSlambday: but we released right?19:05
@HeikoSlambday: so now put stuff in and use it to clean up shoguns code :)19:05
@HeikoSthats very good for a clean up release19:05
@HeikoSas no new features added but internals cleaned up and possible speed gain19:06
@HeikoSso merge it!19:06
@HeikoSits awesome stuff19:06
lambdayHeikoS: alright.. so I'm merging the native implementation patch
@HeikoSlambday: all tested etc right?19:06
lambdayyep.. travis fails are not related19:06
@HeikoSlambday: the native implementations are important since it allows us to use the interface before we got backends19:06
@HeikoSlambday: so now a sweep over Shoguns code that makes use of those methods is needed19:07
@HeikoSlambday: put it as an entrance taks and let sutdents do it?19:07
lambdayHeikoS: that's what I was thinking :D19:07
lambdayHeikoS: but first I'd like to move the linalg code from SG* to linalg19:08
@HeikoSlambday: yeah very good stuff19:08
-!- shogun-notifier- [] has joined #shogun19:11
shogun-notifier-shogun: lambday :develop * 91ecdd2 / / (7 files):
shogun-notifier-shogun: Added Shogun's native dot impl to linalg under NATIVE backend19:11
shogun-notifier-shogun: Soumyajit De :develop * acdd979 / / (7 files):
shogun-notifier-shogun: Merge pull request #2672 from lambday/develop19:11
shogun-notifier-shogun: Added Shogun's native dot impl to linalg under NATIVE backend19:11
shogun-buildbotbuild #2634 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>, Soumyajit De <>19:18
shogun-buildbotbuild #966 of FCRH - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>, Soumyajit De <>19:21
-!- travis-ci [] has joined #shogun19:22
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 []19:22
shogun-buildbotbuild #201 of osx2 - modular_interfaces is complete: Failure [failed csharp modular]  Build details are at  blamelist: lambday <>19:30
shogun-buildbotbuild #202 of osx2 - modular_interfaces is complete: Failure [failed csharp modular]  Build details are at  blamelist: Soumyajit De <>19:34
-!- exploreshaifali [~shaifali@] has quit [Ping timeout: 245 seconds]19:34
-!- HeikoS [] has quit [Quit: Leaving.]19:43
-!- HeikoS [] has joined #shogun19:44
-!- mode/#shogun [+o HeikoS] by ChanServ19:44
@HeikoSgoing climbing now, see you guys19:44
-!- HeikoS [] has quit [Client Quit]19:48
-!- dhfromkorea [] has joined #shogun19:50
-!- dhfromkorea [] has quit [Remote host closed the connection]19:54
-!- nszceta [] has joined #shogun21:03
-!- nszceta [] has quit [Max SendQ exceeded]21:04
-!- nszceta [] has joined #shogun21:04
-!- nszceta [] has quit [Max SendQ exceeded]21:05
-!- nszceta [] has joined #shogun21:06
-!- nszceta [] has quit [Max SendQ exceeded]21:07
-!- nszceta [~nszceta@] has joined #shogun21:11
-!- nszceta [~nszceta@] has quit [Max SendQ exceeded]21:11
-!- exploreshaifali [~shaifali@] has joined #shogun21:15
-!- shogun-notifier- [] has quit [Quit: transmission timeout]22:11
shogun-buildbotbuild #520 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: lambday <>, Soumyajit De <>22:14
-!- lambday [6a338604@gateway/web/freenode/ip.] has quit [Quit: Page closed]23:23
-!- curiousguy13 [~curiousgu@] has quit [Ping timeout: 245 seconds]23:33
-!- txomon|home [~txomon@unaffiliated/txomon] has joined #shogun23:37
-!- curiousguy13 [~curiousgu@] has joined #shogun23:46
-!- PirosB3 [] has joined #shogun23:53
-!- dhfromkorea [] has joined #shogun23:55
-!- txomon|home [~txomon@unaffiliated/txomon] has quit [Ping timeout: 240 seconds]23:56
-!- dhfromkorea [] has quit [Ping timeout: 256 seconds]23:59
--- Log closed Fri Jan 30 00:00:57 2015