Open in new window / Try shogun cloud
--- Log opened Mon Aug 06 00:00:17 2012
@sonney2kfor linear svms yes for kernel svms dot() won't change00:00
@sonney2kohh it does - you also need to use bit shifts00:01
blackburnbut it is probably the same00:01
blackburnit would make sense if I had no memory but it fits00:01
@sonney2kdepends - less mem req. may mean better cache usage00:04
@sonney2kand you will have much less mem accesses too00:04
blackburnsonney2k: unfortunately it stopped on 90% :(00:04
-!- n4nd0 [] has joined #shogun00:06
n4nd0hi blackburn00:06
n4nd0let me ask you00:07
n4nd0my compile log looks pretty ugly00:07
n4nd0full of warnings related to Eigen00:07
n4nd0something local I guess00:07
n4nd0lot of things like this one, give me a sec00:07
blackburnwe all have it :)00:07
n4nd0doesn't it look pretty bad?00:09
blackburnno not really00:10
n4nd0mmm ok00:10
blackburnI wanted to fix ithat but forgot00:10
n4nd0at least pretty long00:10
n4nd0I got a bit concerned because I run some examples but didn't get anything on screen00:11
blackburnthe key is to make it -isystem instead of -I00:11
n4nd0something fishy with SG_PRINT I guess00:11
n4nd0I thought it could be related00:11
blackburnno should not00:11
n4nd0can you please tell me if your SG_PRINT works fine even with those warnings around?00:12
blackburnyes, it works00:12
n4nd0ok, thank you00:12
blackburnn4nd0: you could get warnings off for now00:13
blackburnjust open .config00:13
blackburnfind -I of eigen00:13
blackburnand make it -isystem instead00:13
n4nd0ok, I will do that after my current compilation00:13
blackburnI can't stop looking at that cat00:21
@sonney2kblackburn, it has your eyes :D00:23
n4nd0good night guys01:19
-!- n4nd0 [] has quit [Quit: leaving]01:35
-!- gsomix [~gsomix@] has quit [Ping timeout: 272 seconds]01:43
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]03:29
shogun-buildbot_build #51 of nightly_default is complete: Success [build successful]  Build details are at
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Quit: Page closed]05:14
-!- gsomix [~gsomix@] has joined #shogun07:24
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]08:34
-!- pluskid [~pluskid@] has joined #shogun10:55
-!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has joined #shogun11:05
-!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has left #shogun []11:07
-!- uricamic [~uricamic@2001:718:2:1634:c92:f823:ac7e:c99f] has joined #shogun11:42
-!- blackburn [~blackburn@] has joined #shogun12:26
-!- pluskid [~pluskid@] has quit [Ping timeout: 246 seconds]12:27
-!- pluskid [~pluskid@] has joined #shogun12:44
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]13:52
-!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has joined #shogun16:13
-!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has left #shogun []16:23
-!- Netsplit *.net <-> *.split quits: pluskid16:53
-!- uricamic [~uricamic@2001:718:2:1634:c92:f823:ac7e:c99f] has quit [Quit: Leaving.]16:55
-!- Netsplit over, joins: pluskid16:55
-!- pluskid [~pluskid@] has quit [Ping timeout: 240 seconds]17:08
-!- pluskid [~pluskid@] has joined #shogun17:08
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has joined #shogun17:27
-!- heiko [] has joined #shogun17:34
-!- pluskid [~pluskid@] has quit [Quit: Leaving]17:41
heikowiking, around?18:00
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has quit [Quit: Page closed]18:08
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has joined #shogun18:53
-!- KMcQuisten_ [d8338942@gateway/web/freenode/ip.] has joined #shogun18:55
KMcQuisten_silly webchat18:55
KMcQuisten_why do you torment me so?18:55
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:57
-!- blackburn [~blackburn@] has joined #shogun19:09
-!- n4nd0 [] has joined #shogun19:14
KMcQuisten_Hey there blackburn19:15
blackburnKMcQuisten_: hey I just saw your mail19:15
blackburnno, gsomix was a little away19:16
blackburnKMcQuisten_: actually all method calls increase reference count with director19:24
blackburnso even some debug stuff makes it increase with no limits :D19:24
KMcQuisten_That's a bit of a problem, isn't it?19:25
blackburnKMcQuisten_: yes and in your example something makes features to be deleted :)19:25
KMcQuisten_Very odd.19:25
blackburnkind of shootin' leg19:25
KMcQuisten_Yeah I hope gsomix can figure it out soon.  I have a whole host of custom kernels that I want to use the directorkernel framework for.19:26
blackburnKMcQuisten_: okay I reached him so we will try to figure it out19:30
KMcQuisten_Awesome.  Thanks so much, guys.19:31
blackburnKMcQuisten_: try to call __disown__() on each created director kernel19:44
blackburnbefore using19:44
blackburnI have to reconfigure shogun meanwhile19:44
KMcQuisten_WHen should I call it?19:46
KMcQuisten_Before I pass it to the machine with set_kernel()?  OR after?19:49
blackburnKMcQuisten_: after creation19:50
blackburnwhen you construct combined kernel I think19:50
KMcQuisten_Is it a method call from the kernel object, or CombinedKernel.__disown__(directorkernel)?19:51
KMcQuisten_or is disown a global function19:51
blackburnKMcQuisten_: kernel object19:55
-!- heiko [] has quit [Quit: Leaving.]19:57
blackburnuh finally compiled with directors19:57
blackburnit goes pretty slow we need to speed up it somehow :D19:58
KMcQuisten_Ok, that fixed that segfault.  Now, what about other situations where I'm creating, using and destroying a bunch of director kernels, but not in a combinedkernel?  Do I needto call disown on any director kernel that gets passed to a regular shogun object?19:58
blackburnyeah example works here too19:59
blackburnKMcQuisten_: yeah each time you create anything director - disown it19:59
KMcQuisten_Disown it after i pass it to another object, yes?  If I disown it right away won't it be deleted?19:59
blackburnKMcQuisten_: well yeah, disown it before use in any shogun object20:02
blackburnI am unsure if that can be done automaticall20:02
KMcQuisten_I can do an isinstance check in my code for that20:03
blackburnI will check whether that can be done automatically20:03
-!- heiko [] has joined #shogun20:11
-!- heiko [] has quit [Client Quit]20:15
-!- heiko [] has joined #shogun20:15
-!- n4nd0 [] has quit [Quit: Ex-Chat]20:20
-!- KMcQuisten_ [d8338942@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:27
-!- heiko [] has left #shogun []20:28
@sonney2kblackburn, hmmhh would be better to discus the stuff KMcQuisten runs into on the mailinglist than in privat21:38
blackburnsonney2k: yeah21:43
blackburnsonney2k: did you know about that issue btw?21:43
@sonney2kblackburn, I don't even know now and no one else except you does :D21:50
@sonney2kblackburn, do we have a function in evaluation to get the confusion matrix?21:50
blackburnsonney2k: yes, multiclassaccuracy has one21:51
@sonney2kblackburn, because
@sonney2kI guess that one shoudl use it then?21:52
blackburnyes pluskid came with his own solution again21:52
@sonney2kok told him21:56
@sonney2kblackburn, do you know if n4nd0 fixes up matrix features to be more general / whether he introduces SGMatrixList as new data type to not have yet another type*, int, int issue later?21:57
@sonney2kblackburn, ahh and do you know where gsomix is hiding :D21:57
blackburnno idea21:57
blackburnabout first21:57
blackburngsomix is available in jabber21:57
@sonney2kit is pretty bad that we always get *huge* PRs :(21:58
@sonney2kblackburn, I don't know about you but really carefully doing code review for these big monsters is a pain so I only do it superficially...21:59
blackburnsonney2k: yeah it can be hard sometimes22:00
@sonney2kpluskids PR has ~1700 lines n4nd0s 2600...22:05
@sonney2kblackburn, ohh I see your eyes in github22:06
* sonney2k files hypnotized22:06
blackburnsonney2k: :D22:06
blackburnsonney2k: isn't that cat awesome?22:06
@sonney2kblackburn, yes22:08
@sonney2kblackburn, did you receive the github love email22:08
blackburnsonney2k: yeah22:08
-!- n4nd0 [] has joined #shogun22:22
blackburnwiking: what's that about?:)22:31
wikingthe mars thingy has landed todaybtw :)22:35
blackburnwiking: yeah pretty funny for me to compare that with inquisition taking place in russia22:36
-!- gsomix [~gsomix@] has joined #shogun22:36
gsomixgood evening22:37
blackburnwiking: btw I was playing around with freaks and they are pretty interesting22:38
blackburnwiking: I'd like to use opencv to compute features now22:38
wikingblackburn: eheheh told you22:38
blackburnrewriting such things is way too slow22:38
wikingopencv is good enough for that22:38
wikingnothing else though :)22:38
blackburnit is ugly actually but nothing better22:38
blackburnwiking: so are we going to use BoW?22:39
wikingwell since i have the code for it22:41
wikingwe could22:42
blackburnwiking: I tried freak on road signs with one keypoint at center of image22:42
wikinghow was it?22:43
blackburnnot the best descriptor for sure but ~90% of accuracy can be reached22:43
blackburnactually it would be better for a few keypoints I think22:44
blackburnwiking: what I do not understand is alex's comment22:44
blackburnI mean authors state it works..22:45
blackburnone reason to try freak is that it is terrible fast22:45
wikingwell can't we do some dense freak ?22:48
wikingor at least yeah we should do several keypoints for bow22:49
blackburnwiking: what is dense here?22:49
wikingwell good question22:51
blackburnwiking: you mean some grid of keypoints?22:52
blackburnwhy to do that? in these pics I thought we need a few keypoints22:52
blackburnwiking: what is your bow code based on?23:06
wikingblackburn: that's fully java+hadoop+mahout code23:10
blackburnwiking: how do you compute keypoints/descriptors?23:10
wikingblackburn: that's coming from the opencv code in my case23:10
blackburnwiking: do you think we can optimize it for binary descriptors?23:11
wikingafaik we could23:12
wikingbut we should just run  test asap23:12
blackburnwiking: what's actual deadline btw?23:12
-!- n4nd0 [] has quit [Ping timeout: 252 seconds]23:20
CIA-18shogun: Sergey Lisitsyn master * rfb93440 / (19 files in 2 dirs): Crossvalidation support for SLEP based multitask -
shogun-buildbot_build #161 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>23:41
--- Log closed Tue Aug 07 00:00:17 2012