Open in new window / Try shogun cloud
--- Log opened Mon Oct 10 00:00:26 2016
-!- zxtx_ [] has joined #shogun10:55
-!- Netsplit *.net <-> *.split quits: sukey, shogitter, zxtx10:56
-!- HeikoS [] has joined #shogun12:11
-!- mode/#shogun [+o HeikoS] by ChanServ12:11
@HeikoSwiking: around?12:12
@HeikoScfjhallgren: jojo12:13
@wikingHeikoS: yes12:14
@HeikoSwiking: airbnb, see my meail12:28
@HeikoSwiking: should book today or so12:28
@HeikoSdid you see the listings I sent back a while ago?12:28
@wikingi saw12:29
@HeikoSlisitsyn: hi12:33
@HeikoSlisitsyn: all good?12:34
@HeikoSall good here12:35
@HeikoSlisitsyn: are busy are you?12:36
@HeikoSlisitsyn: would love to have this cross-language integration testing going12:36
lisitsynHeikoS: a bit tired of being that kind of corporate slave :D12:36
@HeikoSfor that, we need this guy here in the modular interfaces12:37
lisitsynohh it sounds very boring12:37
@HeikoSlisitsyn: it is :)12:38
@HeikoSbut it would be increase test coverage a lot12:39
@HeikoSwhich is  even more boring :D12:39
lisitsynIm dying12:39
lisitsynhow can you put your spare time12:39
lisitsynon such a  b o r i n g thing12:39
@HeikoSits the last step of the meta examples12:39
@HeikoSthen its done and I dont need to touch it again12:39
@HeikoStesting is solved in shogun12:40
@HeikoSso it is in anticipation of not spending time with testing anymore12:40
@HeikoSlisitsyn: well I try again tomorrow12:40
@HeikoSyou seem to change mood ;)12:40
@HeikoSlisitsyn: even more fun in the admin of the foundation12:41
@HeikoSbecause we did a formal mistake in the application12:41
@HeikoSso need to send again12:41
@HeikoSnow THAT is meeeg12:41
lisitsynoh fuck12:42
lisitsynI would kill myself12:43
lisitsynit is the definition of boresome12:43
@HeikoSlisitsyn: so what excites you these days?12:45
@HeikoSused to be different!12:45
lisitsyndid I say I am fancy of doing *integration testing*? :P12:45
lisitsynHeikoS: I would be happy to spend some time finishing get/set thingy12:46
lisitsynits kind of game changer12:47
@HeikoShey yeah12:51
@HeikoSthat would be absolutely amazing12:51
@HeikoSlisitsyn: finish it!12:51
@HeikoSget rid of all the shit12:51
@HeikoSmake the plugins works12:51
@HeikoSlisitsyn: so whats next step?12:51
@HeikoSlisitsyn: all get/set patches merged?12:51
@HeikoSwhats needed?12:51
@HeikoSlisitsyn: you know wiking and me will spend quite some time hacking shogun in two weeks12:52
@HeikoScan we help you?12:52
lisitsynprobably not sure12:52
lisitsynI'd have to check the status12:52
lisitsyndon't remember it by heart12:52
@HeikoSremember how the project ended?12:53
@HeikoSI think the parameter stuff is in right?12:54
@HeikoSit needs to be applied shogun wide now12:55
lisitsynkind of12:56
lisitsynthere is a branch12:56
lisitsynnot merged yet12:56
lisitsynif it is not merged it is lost12:56
@HeikoSwhats needed to merge it?12:56
lisitsynI am not even sure it is mergeable now12:56
lisitsynit is!12:58
lisitsynthat's good12:58
-!- HeikoS [] has quit [Ping timeout: 260 seconds]13:25
CaBahave the examples been adopted after the GaussianKernel paramter change from 'width' to 'log_width'?14:13
CaBagaussian_kernel_width.build_values(-math.log(2.0), 0.0, R_EXP, 1.0, 2.0)14:13
CaBais this example actually a useful one?14:13
-!- HeikoS [] has joined #shogun14:48
-!- mode/#shogun [+o HeikoS] by ChanServ14:48
@HeikoSlisitsyn: merged it?14:49
lisitsynnono not yet14:49
@HeikoSso how can I help?14:49
lisitsynHeikoS: unlikely I get time before this night14:50
@HeikoSthats why I asked how can I help?14:50
lisitsynHeikoS: nah just wait :)14:52
@HeikoSok :)14:52
@HeikoSI want the game to be changed :)14:52
@HeikoSCaBa: big changes incoming ;)14:53
lisitsynlet me just pass biyearly performance review wuhahhah14:53
@HeikoSand more payraises ;)14:54
CaBaHeikoS: what's happening?14:54
lisitsynyes that's why14:54
@HeikoSCaBa: lisitsyn is getting a payraise ;)14:55
@HeikoSCaBa: and some GSoC fruits to be harvested14:55
CaBaHeikoS: what's affected?14:57
CaBaby the way, the parallel xval is for sure functional? i'm performing a rather large grid search and see no parallelization in effect there...14:58
@HeikoSCaBa: shoguns core architecture can be modernized then14:58
@HeikoSCaBa: I think it works14:59
@HeikoSyou might have to set no of threads manually as detection might not work14:59
CaBaHeikoS: how are the available threads distributed among the x-val and the machine?14:59
@HeikoSCaBa: at the moment, we use a very crude approach15:00
CaBai think detection works because the SVM does use more than one thread15:00
@HeikoSbut it is effective15:00
CaBahmmm... my process is pretty much at ~250 % cpu permanently...15:00
@HeikoScan you print the number of threads using the above?15:00
CaBaon a 24 thread system15:00
@HeikoSI mean how many shogun has set15:01
CaBawith the corresponding getter you mean?15:01
@HeikoSfrom Pythin15:01
@HeikoSfor me returns 115:02
@HeikoSeven though I have 415:02
@HeikoSso I need to15:02
@HeikoSIn [6]: sg.GaussianKernel().get_global_parallel().get_num_threads()15:02
@HeikoSOut[6]: 115:02
@HeikoSIn [7]: sg.GaussianKernel().get_global_parallel().set_num_threads(4)15:02
@HeikoSIn [8]: sg.GaussianKernel().get_global_parallel().get_num_threads()15:02
@HeikoSOut[8]: 415:02
@HeikoSIn [9]:15:02
@HeikoSand after the xvalidation uses 4 threads15:02
@HeikoSCaBa: let me know15:03
@HeikoSCaBa: btw this is xvalidation that is parallelised15:04
@HeikoSnot the grid search (atm)15:04
@HeikoSCaBa: we could work on that if you want15:04
CaBabut parallelization is over folds, not runs, right?15:04
@HeikoSCaBa: check the code, ill send a link15:05
@HeikoSso we could offer multiple strategies to parallelise15:06
@HeikoSover runs15:06
@HeikoSover folds15:06
@HeikoSover parameter sweeps15:06
@HeikoSCaBa: so over folds only really helps if each fold takes some time15:07
CaBaHeikoS: number of threads reported is 2415:07
@HeikoSbut then helps a lot15:07
@HeikoShow many folds are you doing?15:07
@HeikoShow many runs?15:07
@HeikoShow many points does your grid have?15:07
@HeikoS24 folds is unlikely I guess ;)15:07
@HeikoSso we could add parallelisation over runs as well?15:08
@HeikoSor do it in the grid search15:08
@HeikoSwhatever suits you15:08
@HeikoSyou willing to send a patch?15:08
@HeikoSI can help15:08
CaBaHeikoS: during model selection i use 1 run, 10 folds15:08
@HeikoSbut the code in the link should give a good idea of how it works15:08
@HeikoSok, so you cant fully exploit your setupo15:09
CaBaso the folds won't run in parallel?15:09
@HeikoSthey do15:09
@HeikoScheck the link15:09
@HeikoSalso tested etc15:10
@HeikoSoh and only if machine is not locked15:10
@HeikoScan you check whether yours is locked?15:10
CaBaHeikoS: CCrossValidation() is initialized with autolock=true15:12
CaBaHeikoS: but i dont manually lock the machine anywhere15:12
@HeikoScan you set it to false and try?15:15
@HeikoSdo you use a kernel machine there?15:16
@HeikoSI see15:16
@HeikoStry disabling the autolock15:16
@HeikoSjust for debugging15:16
@HeikoSthen cpu should go up15:16
@HeikoSand then I can add parallel xvalidation for precomputed kernel matrices15:17
@HeikoSshouldnt be too hard15:17
@HeikoSand its good to have a use case from your side15:17
CaBaHeikoS: indeed now i seem to have a higher CPU usage15:20
@HeikoSCaBa: ok!15:21
CaBaHeikoS: ~660% instead of ~230%15:21
@HeikoSthat's it then15:21
@HeikoSironically, it might slower after all15:21
@HeikoSsince kernel matrix is not precomputed now15:21
@HeikoSdo you notice?15:21
@HeikoSbut we can add that feature back in15:21
@HeikoSlet me know15:22
@HeikoSgotta run now, but back later15:22
CaBacan't say, the runs take me half a day15:22
CaBai'm really quite frustrated with that to be honest :(15:22
CaBaok, cool cu later15:22
-!- HeikoS [] has quit [Ping timeout: 240 seconds]15:48
-!- HeikoS [] has joined #shogun15:57
-!- mode/#shogun [+o HeikoS] by ChanServ15:57
-!- mizari [] has joined #shogun16:20
CaBais there an interface to get the kernel matrix out of a machine?17:01
-!- HeikoS1 [] has joined #shogun17:36
-!- HeikoS [] has quit [Ping timeout: 250 seconds]17:38
Saurabh7CaBa: ckernel->get_kernel_matrix i think ..18:18
HeikoS1Caba yes18:37
HeikoS1thats it ^18:37
Saurabh7HeikoS1: hey !18:37
Saurabh7how are you18:37
CaBaHeikoS1: is that a minor change to make the parallel xval work on locked machines?18:38
Saurabh7ah yes that was left unfinished18:38
Saurabh7I can attempt to do something this week :)18:39
-!- HeikoS1 [] has quit [Ping timeout: 260 seconds]18:41
Saurabh7I had it working for locked iirc, but it was hackish, so HeikoS made me remove it :p18:45
-!- HeikoS [] has joined #shogun18:58
-!- mode/#shogun [+o HeikoS] by ChanServ18:58
-!- HeikoS [] has quit [Ping timeout: 260 seconds]19:02
-!- c4goldsw [8a33f9de@gateway/web/cgi-irc/] has joined #shogun19:57
c4goldswHey wiking, you around?19:57
-!- c4goldsw [8a33f9de@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]20:09
-!- mizari [] has quit [Quit: Leaving]20:18
-!- GandalfTheWizard [~Emma@] has joined #shogun20:20
-!- GandalfTheWizard [~Emma@] has quit [Max SendQ exceeded]20:20
-!- GandalfTheWizard [~Emma@] has joined #shogun20:21
-!- HeikoS [] has joined #shogun20:24
-!- mode/#shogun [+o HeikoS] by ChanServ20:24
-!- HeikoS [] has quit [Client Quit]20:25
-!- HeikoS [] has joined #shogun20:25
-!- mode/#shogun [+o HeikoS] by ChanServ20:25
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has joined #shogun20:37
c4goldswHey HeikoS20:45
-!- HeikoS [] has quit [Quit: Leaving.]20:47
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]21:05
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has joined #shogun21:15
-!- GandalfTheWizard [~Emma@] has quit [Ping timeout: 260 seconds]21:32
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]21:40
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has joined #shogun21:41
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]22:19
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has joined #shogun22:21
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]22:35
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has joined #shogun22:36
-!- c4goldsw [8a33fa12@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]23:10
--- Log closed Tue Oct 11 00:00:27 2016