--- Log opened Mon Jul 02 00:00:17 2012 | ||
blackburn | sonney2k: please do something with R static on buildbot | 00:06 |
---|---|---|
-!- blackburn [~blackburn@81.28.187.0] has quit [Quit: Leaving.] | 01:27 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 06:40 | |
n4nd0 | wiking: around? | 08:37 |
-!- uricamic [~uricamic@2001:718:2:1634:7c3b:f22a:ec37:4d] has joined #shogun | 08:58 | |
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has joined #shogun | 09:55 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 252 seconds] | 10:02 | |
cheng | Hi all, I spent some time mucking around trying to compile shogun on OSX, and was just wondering whether the problems I'm having are unique to my box. | 10:07 |
cheng | Basically, I have blas/lapack issues. | 10:07 |
cheng | OSX has plenty of different options, and in the end, I had to add --libs=/usr/lib to the configure to make it detect OSX veclib. | 10:08 |
cheng | Lots of other options had crazy errors due to gfortran versions, etc | 10:09 |
cheng | Just thought I'd share here to see whether anyone has any experience with this. It seems that /usr/lib should be obvious to add to the default configure. | 10:10 |
cheng | One more question: how do I get the SVM outputs for binary classification? | 10:31 |
cheng | get_labels returns -1,+1, and I'm not sure how get_confidence works. | 10:32 |
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has quit [Quit: Leaving.] | 10:32 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 11:11 | |
-!- blackburn [~blackburn@81.28.187.0] has joined #shogun | 11:33 | |
CIA-18 | shogun: Sergey Lisitsyn master * r7765f01 / src/shogun/lib/slep/slep_tree_mt_lr.cpp : Fixed tree guided multitask logistic regression - http://git.io/ffIYDA | 11:39 |
CIA-18 | shogun: Sergey Lisitsyn master * r84b7803 / (2 files): Merge branch 'slep' of git://github.com/lisitsyn/shogun - http://git.io/QQh87Q | 11:39 |
CIA-18 | shogun: Sergey Lisitsyn master * r82caba4 / src/shogun/lib/slep/slep_tree_mt_lsr.cpp : Fixes for multitask tree guided least squares - http://git.io/zc_MrQ | 11:39 |
blackburn | n4nd0: congrats :D | 11:44 |
n4nd0 | blackburn: thank you :) | 11:44 |
n4nd0 | kind of unbelievable, 4-0! | 11:44 |
blackburn | well they became pretty upset after 2-0 | 11:46 |
n4nd0 | Italy? | 11:47 |
@sonney2k | n4nd0, yeah congrats from my side too. the only game where spain played nicely | 11:47 |
@sonney2k | though | 11:47 |
@sonney2k | well deserved, very nice goals | 11:48 |
n4nd0 | sonney2k: c'mon ... against Ireland it was good too, similar to yesterday's one to tell the truth | 11:48 |
n4nd0 | the extra time against Portugal was good as well | 11:48 |
@sonney2k | true I forgot about that one | 11:48 |
@sonney2k | no portugal did actually deserve to win | 11:48 |
@sonney2k | but I really expected just a 1:0 from spain | 11:49 |
n4nd0 | :O I don't agree with you there, sorry | 11:49 |
@sonney2k | and then some spanish wall | 11:49 |
n4nd0 | yeah, they actually played, that was nice | 11:49 |
blackburn | sonney2k: seen cheng /usr/lib issue? | 11:50 |
@sonney2k | blackburn, I have no clue what goes wrong with R on the buildbot | 11:51 |
@sonney2k | I mean it obviously installs ?! | 11:51 |
blackburn | just like I have no clue what goes wrong with octave modular :D | 11:51 |
@sonney2k | blackburn, we need some mac system as buildbot... | 11:51 |
@sonney2k | blackburn, you mean with sparse typemaps? | 11:51 |
blackburn | yeah it reproduces only w/o valgrind/gdb | 11:52 |
blackburn | ah no | 11:52 |
blackburn | it reproduces with gdb | 11:52 |
@sonney2k | blackburn, btw I can no longer visit github.com since the lates chrome upgrade | 11:52 |
@sonney2k | always hangs | 11:52 |
blackburn | when was it? | 11:52 |
@sonney2k | the 20.x upgrade | 11:52 |
@sonney2k | I have no idea what goes wrong | 11:53 |
blackburn | 20.0 here | 11:53 |
blackburn | works smoothly | 11:53 |
@sonney2k | great | 11:53 |
@sonney2k | not here | 11:53 |
@sonney2k | blackburn, nvidia inside? | 11:53 |
blackburn | no, some intel | 11:53 |
blackburn | not discrete | 11:53 |
-!- Netsplit *.net <-> *.split quits: zxtx | 11:55 | |
-!- Netsplit over, joins: zxtx | 11:56 | |
@sonney2k | no idea - can only tell it sucks :/ | 12:01 |
blackburn | sonney2k: linux world :D | 12:01 |
@sonney2k | all other browsers work of course | 12:02 |
blackburn | just some hardware issue linux is full of | 12:03 |
@sonney2k | windoze too but please buy dedicated w* hardware so things are well tested. | 12:06 |
@sonney2k | n4nd0, btw do things go well now? | 12:08 |
blackburn | win is pretty nice with hardware because of $ | 12:08 |
@sonney2k | did you have a session with uricamic? | 12:08 |
n4nd0 | sonney2k: yes, I am working on the hm-svm | 12:08 |
n4nd0 | I have been discussing with him but nothing special | 12:08 |
n4nd0 | vojtech said he would send me a document with the unconstrained formulation of my problem so I could compare the results of my method | 12:09 |
@sonney2k | k | 12:09 |
n4nd0 | but I guess he has not managed to find time yet | 12:09 |
n4nd0 | I should talk to Nico soon | 12:13 |
n4nd0 | I hope he has some time today | 12:13 |
@sonney2k | uricamic, could you please ping vojta about this issue? | 12:16 |
n4nd0 | sonney2k: btw, Nico told me that I'd be able to ask Georg Zeller while he's on holidays, do you know if he's normally available? | 12:23 |
@sonney2k | n4nd0, just email him | 12:24 |
@sonney2k | the sooner the better though | 12:25 |
@sonney2k | I would even do it this week just to get him to know | 12:25 |
@sonney2k | I know him personally | 12:25 |
@sonney2k | he is the other of the matlab hm svm toolbox | 12:26 |
@sonney2k | and knows all the algorithmics but not so much about shogun | 12:26 |
n4nd0 | ok | 12:27 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Remote host closed the connection] | 12:27 | |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 12:28 | |
@sonney2k | n4nd0, and like nico he is a nice guy so no worries | 12:28 |
n4nd0 | good :) | 12:28 |
@sonney2k | n4nd0, I am too but that is not so obvious :P | 12:28 |
n4nd0 | hahaha | 12:29 |
blackburn | you are not! | 12:29 |
blackburn | :( | 12:29 |
blackburn | you are the dark lord and depression comes around you | 12:29 |
@sonney2k | blackburn, I know I am bad because I am pressing you to fix the build errors you introduced :P | 12:37 |
blackburn | meh :( | 12:42 |
blackburn | bike time | 12:42 |
blackburn | :D | 12:42 |
-!- blackburn [~blackburn@81.28.187.0] has left #shogun [] | 12:42 | |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Remote host closed the connection] | 12:51 | |
uricamic | hi, I notified Vojta about that, he promised he will send the document during afternoon | 12:51 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 12:51 | |
uricamic | sonney2k, n4nd0 ^ | 12:51 |
n4nd0 | good | 12:51 |
n4nd0 | thank you! | 12:51 |
@sonney2k | uricamic, thanks! | 12:51 |
uricamic | no problem, I just returned from the lunch, sorry for delay | 12:52 |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has joined #shogun | 13:19 | |
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has joined #shogun | 13:21 | |
-!- heiko [~heiko@host86-182-167-223.range86-182.btcentralplus.com] has joined #shogun | 13:23 | |
-!- blackburn [50ea71fe@gateway/web/freenode/ip.80.234.113.254] has joined #shogun | 13:26 | |
CIA-18 | shogun: Soeren Sonnenburg master * r8fb2bb1 / src/configure : fix typo in configure HAVE_ARPACK -> HAVE_EIGEN3 - http://git.io/aI9Waw | 13:27 |
@sonney2k | cheng, we really have problems getting access to some mac hardware | 13:28 |
@sonney2k | cheng, more permanently that is | 13:28 |
@sonney2k | we simply need one to have a buildbot reliably testing that arch | 13:28 |
cheng | sonney2k: I agree. | 13:29 |
cheng | The general problem with OSX is that one usually has about 4 versions of BLAS/LAPACK. | 13:30 |
cheng | So a buildbot doesn't really solve that problem. | 13:30 |
@sonney2k | well at least we could detect the correct one by default and make sure it stays that way | 13:31 |
cheng | For example, I have the Xcode version, the enthought python version, the macports version and one more version which I cannot quite figure out why it is there, perhaps gfortran install? | 13:31 |
shogun-buildbot | build #21 of deb1 - libshogun started, including [8fb2bb18892999a00adbd5fafd61da52f075eb0c] | 13:31 |
cheng | At least on my mac, veclib was detected by adding --libs=/usr/lib | 13:32 |
shogun-buildbot | build #21 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/21 | 13:32 |
shogun-buildbot | build #29 of deb2 - static_interfaces started, including [8fb2bb18892999a00adbd5fafd61da52f075eb0c] | 13:32 |
@sonney2k | shogun-buildbot, that was fast | 13:32 |
cheng | sonney2k: how do I get svm scores from BinaryLabels? | 13:33 |
@sonney2k | cheng, jsut get the confidences | 13:34 |
@sonney2k | these are the original outputs of the SVM | 13:34 |
cheng | blackburn: To answer your question via email, adding --libs=/usr/lib makes configure detect veclib correctly and compiles happily. | 13:34 |
cheng | sonney2k: BinaryLabels.get_labels() gives +1, -1. | 13:34 |
@sonney2k | cheng, get_confidences() | 13:35 |
shogun-buildbot | build #29 of deb2 - static_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/29 blamelist: Soeren Sonnenburg <sonne@debian.org> | 13:35 |
shogun-buildbot | build #28 of deb3 - modular_interfaces started, including [8fb2bb18892999a00adbd5fafd61da52f075eb0c] | 13:35 |
* sonney2k *argh* | 13:35 | |
cheng | sonney2k: BinaryLabels.get_confidences seems to want an int_32 as an argument. | 13:35 |
@sonney2k | my backup machine died | 13:35 |
@sonney2k | damn thunderstorms | 13:35 |
blackburn | cheng: I see | 13:35 |
@sonney2k | blackburn, isn't it get_confidences? | 13:36 |
blackburn | sonney2k: get_confidences should have no parameter at all | 13:36 |
blackburn | sonney2k: or what do you ask? | 13:36 |
@sonney2k | blackburn, yes exactly | 13:36 |
@sonney2k | cheng, don't mix it up with get_confidence(int) | 13:37 |
@sonney2k | get_confidences() ! | 13:37 |
blackburn | whoops I copypasted ARPACK badly | 13:37 |
blackburn | thanks for fixing | 13:37 |
@sonney2k | blackburn, heiko did you get an email from the buildbot | 13:37 |
@sonney2k | ? | 13:37 |
@sonney2k | (just now that is!) | 13:37 |
cheng | blackburn, sonney2k: thanks! Misread the s | 13:38 |
blackburn | let me check | 13:38 |
cheng | blackburn: I just tested it. It works. | 13:39 |
@sonney2k | blackburn, so? | 13:39 |
@sonney2k | should be your lisit* gmail address | 13:39 |
@sonney2k | gtg... nap time is over | 13:39 |
blackburn | sonney2k: syn | 13:39 |
cheng | sonney2k: later. | 13:40 |
blackburn | sonney2k: yes I got it | 13:40 |
blackburn | new folder at gmail heh | 13:41 |
cheng | blackburn: while I was mucking around with lapack linking, I noted that the GP code need lapack somewhere, but somehow tries to compile it without lapack anyway. | 13:45 |
blackburn | oh ok I'll let puffin know about it | 13:45 |
cheng | I can try to reproduce that if it helps. Not sure how much it is an OSX think. | 13:45 |
cheng | OSX thing. | 13:45 |
blackburn | ah nevermind it is just a #ifdef | 13:46 |
blackburn | cheng: ah I recall I have jmlr paper :D do you think it makes sense to try again after all that time passed? I've never managed to improve before but could start a little later | 13:47 |
cheng | blackburn: please resubmit soon. :-) | 13:48 |
blackburn | how soon? | 13:48 |
blackburn | cheng: you know I wanted to change some code stuff and it would take some time | 13:49 |
blackburn | anyway I feel pretty unoptimistic about that :D probably will end up with arxiv | 13:50 |
cheng | Of course, the sooner the better, but take your time to do a good job. | 13:51 |
cheng | Why unoptimistic? | 13:51 |
cheng | pessimistic? | 13:51 |
blackburn | hah | 13:51 |
blackburn | yeah | 13:51 |
blackburn | well it looks for me just like some crap - I see no valuable part here | 13:52 |
cheng | I see. Well, if it takes longer, the only danger is that you will get new reviewers, who may ask for more stuff. | 13:53 |
blackburn | ah I see | 13:57 |
blackburn | then I will try to add more algos here and provide interface | 13:57 |
cheng | If I remember correctly, the problem was separating from shogun, and not how many algos you have. | 13:58 |
blackburn | sure | 13:59 |
blackburn | but I personally think it lacks some useful algos | 13:59 |
blackburn | t-SNE in particular, sammon mapping may be | 14:00 |
cheng | Your choice, I am just the messenger between reviewers and authors. :-) | 14:01 |
blackburn | thank you | 14:02 |
shogun-buildbot | build #28 of deb3 - modular_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/28 blamelist: Soeren Sonnenburg <sonne@debian.org> | 14:03 |
heiko | sonney2k, blackburn, the buildbot now sends emails to me :) | 14:08 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Operation timed out] | 14:15 | |
-!- cheng [~cheng@115-64-111-17.tpgi.com.au] has quit [Quit: Leaving.] | 14:16 | |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun | 14:54 | |
blackburn | heiko: yeah spambuildbot now | 14:55 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 15:21 | |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun | 15:28 | |
n4nd0 | hey nicococo | 15:31 |
nicococo | hola n4nd0.. how are you | 15:32 |
nicococo | ? | 15:32 |
n4nd0 | I am good, what about you? | 15:32 |
nicococo | well, alright, i think you have 4:0 reasons to feel better than me :) | 15:33 |
n4nd0 | haha :D yeah I think so | 15:34 |
n4nd0 | it was a good day yesterday definetely | 15:34 |
nicococo | i can imagine.. | 15:35 |
nicococo | okay, whats going on with the hmsvm? | 15:35 |
nicococo | (sorry, don't have that much time :( ) | 15:35 |
n4nd0 | no problem, I will try to be brief | 15:35 |
n4nd0 | I implemented the joint feature computation | 15:35 |
n4nd0 | and know I am planning how to do the argmax | 15:35 |
n4nd0 | I have almost all the pieces clear for it ... almost :) | 15:36 |
nicococo | fine | 15:36 |
n4nd0 | ok, so what about these vectors p and q? | 15:37 |
n4nd0 | do they need to be estimated using training data? | 15:37 |
n4nd0 | or are they defined in init_opt method | 15:37 |
n4nd0 | ? | 15:37 |
nicococo | usually no. in our implementation definitely not. | 15:38 |
n4nd0 | no to which question :P? | 15:38 |
nicococo | aehm.. they don't belong to the optimization problem of the sosvm, hence they are not defined in the init_op method | 15:38 |
n4nd0 | ok | 15:39 |
nicococo | 2nd question | 15:39 |
nicococo | are you confused by the p and q.. or is it clear? | 15:39 |
n4nd0 | I don't know how to get them | 15:40 |
nicococo | pre-defined | 15:40 |
nicococo | think of this: you now in which state you start and in which state you end | 15:40 |
n4nd0 | why? | 15:41 |
nicococo | and if you don't now.. then add 2 artificial states just for the start and end of each sequence | 15:41 |
n4nd0 | aham | 15:41 |
nicococo | it has some advantages. and p is mandatory since there is no p(q_1|q_0) defined | 15:42 |
nicococo | did you read the document? | 15:42 |
n4nd0 | when you say add 2 artificial states | 15:42 |
n4nd0 | do you mean to add two elements to the label sequence | 15:42 |
n4nd0 | but for example, if we have a two-state model | 15:42 |
nicococo | that would be one possibility | 15:43 |
n4nd0 | these artificial states would be either -1 or 1 | 15:43 |
n4nd0 | or another new state like 0 or so? | 15:43 |
nicococo | if you have a 2-state model you need two additional states (liek 3 and 4) | 15:43 |
n4nd0 | ok | 15:44 |
nicococo | but for now, p and q are just needed in the viterbi | 15:44 |
n4nd0 | yes | 15:44 |
n4nd0 | does that make a difference? | 15:44 |
nicococo | difference to what? | 15:45 |
n4nd0 | hehe like if to say that they are just needed in the Viterbi makes something different | 15:45 |
n4nd0 | ok, forget about it :) | 15:45 |
nicococo | :) | 15:46 |
n4nd0 | I think I am not very sure how to handle them, but I can think about it more | 15:46 |
nicococo | i would suggest to think about it less :) | 15:47 |
n4nd0 | haha ok | 15:47 |
nicococo | if you look at the definition of the transition matrix | 15:47 |
n4nd0 | yes | 15:47 |
nicococo | you have a transition from states y_i-1 to y_i right? | 15:48 |
n4nd0 | yes | 15:48 |
nicococo | when you start at the first position of the sequence (i=1) then you have a transition from y_0 to y_1 .. but there is no y_0 hence you need a transition solely depending on the y_1 | 15:49 |
nicococo | ...and thats p | 15:49 |
n4nd0 | ok, that's all right | 15:50 |
n4nd0 | it's like the pi vector using the conventional notation for hmm I thinks | 15:50 |
n4nd0 | think* | 15:50 |
nicococo | yes, indeed | 15:50 |
n4nd0 | but in our framework | 15:51 |
nicococo | did you read the FAQ? | 15:51 |
n4nd0 | yeah :) | 15:51 |
nicococo | is it alright or something missing/unclear? | 15:51 |
n4nd0 | in our framework, where should we define p and q? in the HMMSVMModel? | 15:51 |
n4nd0 | I think that it makes sense there | 15:51 |
nicococo | i agree | 15:51 |
n4nd0 | yeah, I have a question about the FAQ too | 15:52 |
n4nd0 | it is about the third question | 15:52 |
nicococo | okay | 15:52 |
n4nd0 | so when I said in the question | 15:52 |
n4nd0 | "Then, the | 15:52 |
n4nd0 | first observation sequence, x0 , is never taken into account | 15:52 |
n4nd0 | " | 15:52 |
n4nd0 | that is wrong | 15:53 |
n4nd0 | isn't it? | 15:53 |
nicococo | since we start with i=1 there is no x_0 | 15:53 |
nicococo | the sequences always go from i=1 ... T | 15:53 |
n4nd0 | I said, note that I am talking about 0-index ;) | 15:53 |
n4nd0 | so it's like the sequences go from i=0 ... T-1 | 15:54 |
n4nd0 | but I think I got the point anyway | 15:54 |
n4nd0 | I have to take into account all the features of x | 15:54 |
nicococo | aeh. oaky.. sure you can convert them | 15:54 |
n4nd0 | you know ... C people are crazy and start counting the things at zero ... don't blame me :) | 15:55 |
nicococo | it's okay. i think i can handle that :) | 15:55 |
n4nd0 | I think that the rest is ok | 15:56 |
nicococo | very good! | 15:56 |
n4nd0 | I am going to do my best to have the Viterbi ready by tomorrow | 15:56 |
nicococo | so, please try everything out such that if anything is unclear we can talk about it as soon as | 15:56 |
nicococo | possible | 15:56 |
n4nd0 | sure | 15:57 |
n4nd0 | I can mail you once I have pushed the Viterbi part | 15:57 |
nicococo | that would be nice. | 15:57 |
n4nd0 | and if you can take a look to the HM-SVM Model it would be great | 15:57 |
nicococo | okay, i can do that. do you think we should talk tomorrow again? | 15:58 |
n4nd0 | maybe we can plan it by mail better? | 15:58 |
n4nd0 | depending on how my progress goes and if I get lot of doubts and stuff | 15:58 |
nicococo | alright. so, if question, hints, or whatever please mail me and we can add everything to the document | 15:59 |
n4nd0 | cool | 15:59 |
n4nd0 | ok, thank you very much then! | 15:59 |
n4nd0 | good luck with your work ;) | 15:59 |
nicococo | so, very good work, thank you for that.. and see you soon | 15:59 |
nicococo | thanks :) | 15:59 |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun [] | 16:00 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has left #shogun [] | 16:01 | |
-!- blackburn [50ea71fe@gateway/web/freenode/ip.80.234.113.254] has quit [Quit: Page closed] | 16:18 | |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Quit: Page closed] | 17:14 | |
-!- heiko1 [~heiko@host86-177-172-27.range86-177.btcentralplus.com] has joined #shogun | 17:16 | |
-!- heiko [~heiko@host86-182-167-223.range86-182.btcentralplus.com] has quit [Ping timeout: 264 seconds] | 17:18 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 17:27 | |
puffin444 | Hey heiko | 17:30 |
heiko1 | hey puffin444 | 17:38 |
puffin444 | what do you think about the model selection checksum? | 17:38 |
heiko1 | I actually just started reading | 17:39 |
heiko1 | Ill get back to you when I understoof it :) | 17:39 |
puffin444 | ok | 17:40 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 17:43 | |
n4nd0 | sonney2k: hey there | 17:44 |
heiko1 | puffin444 this get_char_description | 17:49 |
heiko1 | what do you need it for? the hashing? | 17:49 |
puffin444 | Yes. I use it to make a char array that the MD5 function can hash. Do you think its too complicated? | 17:50 |
heiko1 | it looks a bit strange, why all these type distinctions and redundant code? | 17:50 |
puffin444 | I store the value of the parameter in the description; the value may need more space depending on the type. | 17:51 |
heiko1 | oh btw there is a method to get the size of a TParameter's primitive type | 17:51 |
puffin444 | That would help alot. | 17:51 |
heiko1 | its a bit hidden: have a look at TSGDataType in DataType.h | 17:53 |
heiko1 | every TParameter instance has a TSGDataType | 17:53 |
heiko1 | and this you can ask these things | 17:53 |
heiko1 | but I still dont really get these loops | 17:53 |
puffin444 | Yes that would simplify things. | 17:54 |
heiko1 | big_name contains a char description of the data of the parameter? | 17:54 |
puffin444 | big_name contains a description of the name and values of the parameter as well as all of its children. | 17:55 |
puffin444 | The char description is recursively built. In order to store a value, I cast it as an unsigned char and store it that way. That is what the loops are for. | 17:56 |
heiko1 | where children are respect to what? | 17:56 |
puffin444 | Children that are SGObjects might have modelselection parameters themselves. I make sure that they are part of the description. | 17:56 |
heiko1 | ok, i see ... | 17:57 |
heiko1 | continuing reading .... | 17:57 |
heiko1 | why only model-selection parameters? | 17:57 |
heiko1 | because you only want to know when these changed, right? | 17:58 |
puffin444 | Yes. I suppose I could check all parameters with m_parameters, yes? | 17:58 |
heiko1 | but are other parameters not als important? | 17:58 |
heiko1 | for what do you need that hashing? | 17:58 |
heiko1 | for model-selection? | 17:58 |
heiko1 | or for just knowing when to recompute stuff? | 17:59 |
puffin444 | Knowing when to recompute stuff. sonney2k suggested it so I tried it out. | 17:59 |
heiko1 | ok, yeah I mean its a nice idea | 17:59 |
heiko1 | but I would include all parameters | 18:00 |
puffin444 | will do. | 18:00 |
heiko1 | since these might change during training or whatsoever | 18:00 |
heiko1 | still this description code is strange ;) | 18:00 |
puffin444 | Yes I did fear it was a little over-complicated. | 18:00 |
heiko1 | I will go thorugh it a bit more and see if its possible to simplify | 18:03 |
puffin444 | I also don't know if it's really appropriate for the parameter itself to include SGObject children in its description, but I don't think this checksum would be useful if I didn't look at all of parameters. | 18:03 |
heiko1 | no youre right, a parameter has changed if the sub-parameter has changed | 18:04 |
heiko1 | imagine a svm with a kernel with width | 18:04 |
heiko1 | when the width has changed, the kernel has changed and an evtl kernel matrix in svm has to be recomputed | 18:04 |
heiko1 | puffin444 I think this has to be done differently, see comments in pull request | 18:24 |
puffin444 | Absolutely | 18:25 |
puffin444 | I have been reading your comments. | 18:25 |
heiko1 | (but I dont know how currently, complex thing) | 18:25 |
heiko1 | I also just added some thoughts on the general approach | 18:25 |
puffin444 | heiko, perhaps murmur would be better? | 18:31 |
puffin444 | If I interpret this correctly, I can call CHash::IncrementalMurmurHash incrementally | 18:32 |
heiko1 | yeah could do that | 18:33 |
heiko1 | you would need to extend it a bit | 18:33 |
heiko1 | so that you can specify an address and a number of bytes | 18:33 |
puffin444 | sounds like a simple thing to do. | 18:34 |
heiko1 | yeah, the more complicated thing is not to loose the overview :) | 18:34 |
heiko1 | have a look at the serialisation code first | 18:34 |
heiko1 | thats pretty complicated | 18:34 |
-!- blackburn [~blackburn@81.28.187.0] has joined #shogun | 18:35 | |
heiko1 | many cases | 18:35 |
puffin444 | Doesn't serialization lead to the same problem? I'm doing the copying again, instead this time its for something to be written on disc. | 18:36 |
heiko1 | yes | 18:37 |
heiko1 | you can write to memory | 18:37 |
heiko1 | ybut i meant the way the parameters are visited | 18:37 |
heiko1 | because its exectly the same problem: print/hash values of all parameters (including sub-parameters, vectors/matrices etc) | 18:38 |
puffin444 | Yes. I can't find the class that does this. What's its name? | 18:39 |
heiko1 | SGObject starts the process | 18:39 |
heiko1 | save_serializable or so | 18:39 |
puffin444 | Oh now I see. | 18:39 |
heiko1 | I meant just for inspiration | 18:40 |
@sonney2k | n4nd0, yes? | 18:40 |
heiko1 | you dont need to go into everything, just for getting a feeling for all the cases | 18:40 |
@sonney2k | blackburn, well I always received these emails -guess why I spend so much time fixing issues others create? | 18:40 |
CIA-18 | shogun: Heiko Strathmann master * re5b0dbe / (2 files): removed some of the asserts which might be too strong - http://git.io/nMRneQ | 18:40 |
CIA-18 | shogun: Heiko Strathmann master * rfea2a1e / examples/undocumented/libshogun/Makefile : added statistics examples again (hopefully work now) - http://git.io/FVOQ1Q | 18:40 |
CIA-18 | shogun: Heiko Strathmann master * rf72cc6d / (3 files): Merge pull request #616 from karlnapf/master - http://git.io/b4I9lQ | 18:40 |
blackburn | what? | 18:40 |
puffin444 | Yes this looks like a much better approach. | 18:41 |
shogun-buildbot | build #22 of deb1 - libshogun started, including [e5b0dbeead7748f512a081d326cbb7e388c2b774] | 18:41 |
heiko1 | puffin444, if you then iterate over all these parameters and incrementally hash | 18:41 |
heiko1 | then everything is in place and nice | 18:41 |
heiko1 | also I would moive the stuff to TParameter as I wrote | 18:42 |
shogun-buildbot | build #22 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/22 | 18:42 |
shogun-buildbot | build #23 of deb1 - libshogun started, including [fea2a1e8af7bc6d513d3550705ba5bfdc184ae11, f72cc6d6bb058427d603f016ab9898c2ecf66f07] | 18:42 |
heiko1 | to be able to hash single parameters if needed from another place | 18:42 |
puffin444 | will do. | 18:42 |
blackburn | sonney2k: what is this lament about? :D | 18:42 |
heiko1 | and then a method hash_all_parameters or so for SGObject | 18:42 |
shogun-buildbot | build #23 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/23 | 18:42 |
shogun-buildbot | build #30 of deb2 - static_interfaces started, including [e5b0dbeead7748f512a081d326cbb7e388c2b774] | 18:42 |
puffin444 | I am also working on integrating vector parameters into model selection | 18:42 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Remote host closed the connection] | 18:43 | |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 18:43 | |
heiko1 | wow, another large thing :) | 18:43 |
heiko1 | you will be a framework guru when gsoc is over :D | 18:43 |
shogun-buildbot | build #31 of deb2 - static_interfaces is complete: Retry [retry exception slave lost] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/31 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 18:43 |
shogun-buildbot | build #32 of deb2 - static_interfaces started, including [e5b0dbeead7748f512a081d326cbb7e388c2b774, fea2a1e8af7bc6d513d3550705ba5bfdc184ae11, f72cc6d6bb058427d603f016ab9898c2ecf66f07] | 18:43 |
n4nd0 | sonney2k: can you open this http://83.49.221.169:8080/home? | 18:44 |
heiko1 | argh | 18:44 |
n4nd0 | it is a very first draft though :D | 18:44 |
blackburn | hah | 18:44 |
blackburn | I have seen that template somewhere | 18:44 |
@sonney2k | n4nd0, heh - only problem is that this is the style I use for my homepage - so I would rather have used that other one... | 18:44 |
heiko1 | sonney2k, what is that? | 18:45 |
heiko1 | remoteFailed: [Failure instance: Traceback (failure with no frames): <class 'twisted.internet.error.ConnectionLost'>: Connection to the other side was lost in a non-clean fashion. ] | 18:45 |
@sonney2k | ...I pointed too :/ | 18:45 |
@sonney2k | heiko1, I am still messing with the bb | 18:45 |
heiko1 | sonney2k, ah ok puh :) | 18:45 |
@sonney2k | heiko1, trying to figure out why email notifications for dependent builds don't work | 18:45 |
heiko1 | oh man | 18:46 |
heiko1 | sounds annoying | 18:46 |
@sonney2k | heiko1, my only luck is that people on #buildbot are *extremely* helpful | 18:46 |
@sonney2k | like here I would say | 18:46 |
heiko1 | ah thats good | 18:46 |
heiko1 | power of open-source :) | 18:46 |
shogun-buildbot | build #32 of deb2 - static_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/32 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 18:47 |
shogun-buildbot | build #29 of deb3 - modular_interfaces started, including [e5b0dbeead7748f512a081d326cbb7e388c2b774, fea2a1e8af7bc6d513d3550705ba5bfdc184ae11, f72cc6d6bb058427d603f016ab9898c2ecf66f07] | 18:47 |
-!- uricamic [~uricamic@2001:718:2:1634:7c3b:f22a:ec37:4d] has quit [Quit: Leaving.] | 18:47 | |
puffin444 | heh lol | 18:48 |
@sonney2k | heiko1, blackburn btw I increased verbosity of the bb to maximum | 18:48 |
@sonney2k | basically telling us when it builds what | 18:48 |
@sonney2k | when it finishes with error/success | 18:48 |
@sonney2k | etc | 18:48 |
heiko1 | sonney2k, its nice to see when it start/ends | 18:48 |
blackburn | sonney2k: the only missing thing is bettyboo-style chatting | 18:48 |
@sonney2k | hopefully | 18:48 |
@sonney2k | heiko1, I unfortunatelly killed the other buildslaves at TU and the admin is on vacations | 18:49 |
blackburn | killed admin and buildslaves are on vacations? | 18:49 |
@sonney2k | actually 'killed' windows did some security upgrades and did restart | 18:49 |
heiko1 | lol :) | 18:50 |
heiko1 | ah the joys of windows | 18:50 |
@sonney2k | so now it is gone | 18:50 |
heiko1 | man, hopefully people wont kill you :D | 18:50 |
@sonney2k | heiko1, blackburn opinions on http://83.49.221.169:8080/home? | 18:50 |
@sonney2k | what n4nd0 posted... | 18:51 |
blackburn | sonney2k: I thought it is your homepage first | 18:51 |
blackburn | :D | 18:51 |
@sonney2k | I personally would choose a different style for obvious reasons | 18:51 |
@sonney2k | it is maybe not such a bright idea to connect shogun too much with myself | 18:51 |
n4nd0 | yeah ... sure | 18:51 |
@sonney2k | I dont' want this project to end if I loose interest or so | 18:52 |
n4nd0 | the guy who is working on it will appear around soon and ask which one you like | 18:52 |
heiko1 | sonney2k, I like the tabs | 18:52 |
heiko1 | design - dont know | 18:52 |
heiko1 | I dont like the archive links | 18:52 |
blackburn | sonney2k: hahah what do you expect if you loose interest then | 18:53 |
heiko1 | they are messy and people are probably not interested in that | 18:53 |
heiko1 | actually I would remove this sidebar | 18:53 |
heiko1 | I like the categories | 18:54 |
heiko1 | What would be nice would be a feature list, which is a bit nicer to overview that the main text | 18:54 |
heiko1 | gotta catch train, see you later | 18:55 |
@sonney2k | blackburn, there are enough people nowadays that know their way around in shogun | 18:57 |
blackburn | we need stalin here :D | 18:58 |
blackburn | sonney2k: any stalindates? | 18:58 |
@sonney2k | blackburn, stalin is the national hero of some other $COUNTRY | 18:58 |
blackburn | national what? | 18:58 |
blackburn | :D | 18:58 |
blackburn | sonney2k: he can be hero only of some really braindamaged people | 18:59 |
-!- heiko1 [~heiko@host86-177-172-27.range86-177.btcentralplus.com] has quit [Ping timeout: 264 seconds] | 19:00 | |
blackburn | sonney2k: ok in other words we have no leader but you - better? | 19:00 |
@sonney2k | n4nd0, did you see heikos feedback? | 19:01 |
@sonney2k | n4nd0, ok cool | 19:01 |
n4nd0 | sonney2k: I'll tell my friend to read it in the logs | 19:02 |
-!- bern4rd [~bernard@169.Red-83-49-221.dynamicIP.rima-tde.net] has joined #shogun | 19:12 | |
@sonney2k | n4nd0, I had this template in mind http://www.nuviotemplates.com/template/gr8-01/ | 19:12 |
@sonney2k | hi bern4rd | 19:12 |
bern4rd | hi | 19:12 |
@sonney2k | inferring from the 4 in you name you are a friend of n4nd0 :D | 19:12 |
bern4rd | yes :) | 19:13 |
blackburn | haha | 19:13 |
blackburn | really? | 19:13 |
@sonney2k | bern4rd, so we had some brief discussion - basically we have a webtemplate clash (the one you used is also the one I use on my homepage and shogun should not be too connect to me) | 19:14 |
@sonney2k | so I would prefer http://www.nuviotemplates.com/template/gr8-01/ | 19:14 |
bern4rd | i am downloading it right now | 19:14 |
@sonney2k | bern4rd, is the page already django based or did you 'just' modify the template btw? | 19:15 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 264 seconds] | 19:15 | |
@sonney2k | bern4rd, I just now realize that one can already click through things | 19:15 |
bern4rd | the template is based in django | 19:15 |
@sonney2k | so you did all the website work already ! | 19:16 |
shogun-buildbot | build #29 of deb3 - modular_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/29 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 19:16 |
bern4rd | the basic stuff yes | 19:16 |
bern4rd | but i suppose you want to add more things | 19:17 |
bern4rd | maybe a blog to allow your developers comment their work | 19:17 |
bern4rd | right now the webpage is only showing content so is not too difficult | 19:18 |
@sonney2k | bern4rd, but that is all not so important compared to the redesign | 19:18 |
@sonney2k | bern4rd, so the admin users can edit this content right? | 19:18 |
bern4rd | yes yes I know, i will use the template gr8-01 | 19:18 |
@sonney2k | bern4rd, no what I meant is going from the old page to something new | 19:19 |
@sonney2k | not the cry* style to gr8 change | 19:19 |
@sonney2k | bern4rd, I am impressed already how much work you invested... | 19:19 |
bern4rd | thanks :) | 19:20 |
bern4rd | as we set the deadline on wednesday can I continue working on it? | 19:21 |
@sonney2k | bern4rd, sure sure | 19:21 |
@sonney2k | bern4rd, just come here for updates and discussions | 19:21 |
bern4rd | sure | 19:22 |
@sonney2k | I would say you convert to gr8-01 first | 19:22 |
@sonney2k | then we finalize tabs | 19:22 |
@sonney2k | and then we can actually already put it live | 19:23 |
bern4rd | nice | 19:23 |
@sonney2k | blackburn, did you for build failure #29 receive an email? | 19:24 |
blackburn | sonney2k: I received only one | 19:25 |
blackburn | no idea about # | 19:25 |
@sonney2k | blackburn, this one <shogun-buildbot> -build #29 of deb3 - modular_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/29 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 19:26 |
@sonney2k | ? | 19:26 |
@sonney2k | bern4rd, thanks for you work | 19:26 |
blackburn | Build Reason: downstream | 19:26 |
blackburn | Build Source Stamp: [branch master] 8fb2bb18892999a00adbd5fafd61da52f075eb0c | 19:26 |
blackburn | Blamelist: Soeren Sonnenburg <sonne@debian.org> | 19:26 |
blackburn | sonney2k: just check 'To' | 19:26 |
bern4rd | you're welcome | 19:26 |
bern4rd | I go to work. see you | 19:27 |
-!- bern4rd [~bernard@169.Red-83-49-221.dynamicIP.rima-tde.net] has quit [Quit: Lost terminal] | 19:27 | |
blackburn | sonney2k: argh | 19:28 |
blackburn | spam catched it | 19:28 |
blackburn | I did | 19:28 |
blackburn | sonney2k: no idea what is wrong with octave modular | 19:36 |
blackburn | sonney2k: PARALLEL | 19:44 |
@sonney2k | what is run in parallel? | 19:46 |
@sonney2k | blackburn, with directors or w/o? | 19:47 |
blackburn | sonney2k: get_feature_vector | 19:47 |
blackburn | operator= running in parallel does BOOM | 19:47 |
blackburn | sonney2k: I have a few possible solutions in mind | 19:47 |
blackburn | disable parallel for sparse, synchronize at sparsefeatures or synchronize ref in sgrefdata | 19:48 |
blackburn | sonney2k: your choice? | 19:48 |
@sonney2k | blackburn, it should crash not only in octave | 19:50 |
@sonney2k | we indeed have a couple of algos that are running in parallel for sparse stuff | 19:50 |
blackburn | sonney2k: does distance_sparseeuclidean_modular.py work on your machine? | 19:51 |
@sonney2k | no idea | 19:51 |
blackburn | I am pretty sure it fails in octave for this reason | 19:52 |
blackburn | once I disable threads it works | 19:52 |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has quit [Ping timeout: 245 seconds] | 19:54 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 19:55 | |
@sonney2k | blackburn, well this example should compute the distance matrix in parallel | 19:56 |
@sonney2k | though I see conflicts only on diagonal | 19:56 |
blackburn | sonney2k: so? | 19:57 |
blackburn | we need synchronization here | 19:57 |
@sonney2k | yes | 19:57 |
blackburn | where? | 19:57 |
@sonney2k | same overhead as in SGObject | 19:57 |
blackburn | I'll add it to sparsefeatures then | 19:58 |
@sonney2k | phtread_mutex_t is 40 bytes | 20:01 |
@sonney2k | spinlock is just int | 20:01 |
@sonney2k | so 4 bytes | 20:01 |
@sonney2k | so 4 more bytes are OK I would say | 20:02 |
blackburn | sonney2k: I'll add it to get_feature_vector then | 20:02 |
@sonney2k | blackburn, so do the same as in sgobject in sgreferenced data | 20:02 |
blackburn | not to sgrefdata | 20:02 |
blackburn | do you think so? | 20:02 |
blackburn | I don't mind any | 20:03 |
@sonney2k | yeah, we will run into other issues and 4 bytes per vector is something we have to live with | 20:03 |
@sonney2k | yes shogun won't be fastest this way when we have 3 dimensional data sets | 20:03 |
@sonney2k | but why would one need billions of examples to get accurate estimates for such low dim data anyways? | 20:04 |
@sonney2k | so we are now at 16 byte overhead | 20:04 |
@sonney2k | and if we are really unhappy with it we can of course change things later | 20:05 |
blackburn | sonney2k: argh what is behavior on =? | 20:07 |
blackburn | there should be pointer on lock I guess? | 20:07 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 20:09 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Client Quit] | 20:10 | |
@sonney2k | blackburn, ??? | 20:12 |
@sonney2k | blackburn, whenever the refcount stuff is changed we need the spinlock | 20:12 |
blackburn | sonney2k: just like reference counter it should be shared between vectors/matrices | 20:12 |
@sonney2k | ahh | 20:12 |
@sonney2k | kaboom | 20:12 |
blackburn | why so? | 20:13 |
@sonney2k | ok maybe it is not too bad | 20:13 |
@sonney2k | just create a struct with the refcount and spinlock | 20:14 |
@sonney2k | and alloc this in SGReferencedData | 20:14 |
blackburn | why struct? | 20:14 |
@sonney2k | so the former refcount ptr points there | 20:14 |
@sonney2k | blackburn, how could it work otherwise? | 20:14 |
@sonney2k | you need a lock for the refcount | 20:14 |
@sonney2k | and you need to share it among SGVectors | 20:15 |
blackburn | sonney2k: just track both separately | 20:15 |
@sonney2k | why that? | 20:15 |
@sonney2k | then you have 2 ptrs and 2 ints | 20:15 |
@sonney2k | with a struct you will only have 1 ptr and 2 ints | 20:15 |
blackburn | ahh | 20:15 |
@sonney2k | 8 bytes saved | 20:15 |
blackburn | ok makes sense | 20:15 |
@sonney2k | I hope there is no race condition accessing *ptr though | 20:16 |
blackburn | whoops | 20:16 |
blackburn | hah | 20:16 |
blackburn | well | 20:16 |
blackburn | lets see | 20:16 |
blackburn | I don't think there is | 20:16 |
blackburn | sonney2k: I actually have pointers to locks in LTSA/LLE/blahblah | 20:17 |
blackburn | works nice I believe | 20:17 |
@sonney2k | yeah lets hope so | 20:17 |
@sonney2k | actually it should be fine... only race condition would be when the struct is destroyed | 20:18 |
@sonney2k | but this can only happen when only one object is available | 20:18 |
@sonney2k | so all good | 20:18 |
@sonney2k | let me break the build just for the fun of it | 20:19 |
blackburn | sonney2k: ? | 20:19 |
blackburn | I am doing that now | 20:19 |
@sonney2k | blackburn, you continue fixing things | 20:20 |
@sonney2k | I am testing the bb | 20:20 |
blackburn | ah | 20:20 |
blackburn | I thought you want to break build with these changes | 20:20 |
CIA-18 | shogun: Soeren Sonnenburg master * rebfa8bf / src/configure : breaking configure (buildbot test) - http://git.io/Hxu-4g | 20:21 |
@sonney2k | blackburn, I am configuring the bb for the last several days so no | 20:21 |
@sonney2k | once I am done I focus on website | 20:21 |
shogun-buildbot | build #24 of deb1 - libshogun started, including [ebfa8bf667745650d24afd845ccb239d8d43ffd6] | 20:23 |
shogun-buildbot | build #24 of deb1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/24 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:23 |
@sonney2k | blackburn, hah perfect I got an email to my gmail and debian address :-)) | 20:25 |
@sonney2k | now lets make things a bit more complicated... | 20:25 |
blackburn | I didn't | 20:25 |
blackburn | damn gmail | 20:25 |
blackburn | hmm | 20:26 |
CIA-18 | shogun: Soeren Sonnenburg master * rcf9d5de / src/configure : Revert "breaking configure (buildbot test)" - http://git.io/fHBEOw | 20:26 |
blackburn | sonney2k: should I receive it? | 20:26 |
@sonney2k | blackburn, yes | 20:26 |
@sonney2k | and I've seen you got it | 20:26 |
@sonney2k | (the email had you in CC) | 20:27 |
blackburn | idiot gmail keeps putting it into spam | 20:27 |
blackburn | now in main account | 20:27 |
shogun-buildbot | build #25 of deb1 - libshogun started, including [cf9d5deaa537aad24b403053e88aede97d6fad13] | 20:28 |
@sonney2k | hmmhh no idea how to fix that :/ | 20:28 |
blackburn | sonney2k: fix whaT? | 20:28 |
@sonney2k | blackburn, that gmail puts it into spam | 20:28 |
shogun-buildbot | build #25 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/25 | 20:29 |
shogun-buildbot | build #33 of deb2 - static_interfaces started, including [cf9d5deaa537aad24b403053e88aede97d6fad13] | 20:29 |
blackburn | ah well I set up filter | 20:29 |
@sonney2k | haha success email arrived :D | 20:29 |
blackburn | forcing it not to go to spam | 20:29 |
@sonney2k | now lets see how it does about static | 20:29 |
blackburn | yeah | 20:29 |
blackburn | here | 20:29 |
@sonney2k | blackburn, btw ccache is on the bb | 20:29 |
blackburn | ohh that's nice | 20:29 |
@sonney2k | so that is why it took only 1 min | 20:30 |
@sonney2k | or less :D | 20:30 |
blackburn | sonney2k: I really would like to see interfaces compiling faster | 20:30 |
@sonney2k | me too | 20:30 |
@sonney2k | but I don't know how | 20:30 |
@sonney2k | blackburn, btw where is gsomix? | 20:31 |
blackburn | sonney2k: I don't know | 20:31 |
blackburn | sonney2k: ++(m_refcount->rc) - does this work? | 20:31 |
@sonney2k | sure | 20:31 |
shogun-buildbot | build #33 of deb2 - static_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/33 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:32 |
shogun-buildbot | build #30 of deb3 - modular_interfaces started, including [cf9d5deaa537aad24b403053e88aede97d6fad13] | 20:32 |
-!- blackburn [~blackburn@81.28.187.0] has left #shogun [] | 20:32 | |
@sonney2k | yes it works :D | 20:33 |
-!- blackburn [~blackburn@81.28.187.0] has joined #shogun | 20:36 | |
@sonney2k | let me disassemble the backup machine | 20:37 |
* sonney2k prays everything will be OK once this is done | 20:37 | |
blackburn | +86C | 20:37 |
blackburn | uhh | 20:37 |
@sonney2k | heh | 20:37 |
blackburn | ultrabooks are really hot | 20:37 |
blackburn | modshogun_wrap.cxx: In function 'octave_value_list _wrap_refcount_t_lock_get(const octave_value_list&, int)': | 20:39 |
blackburn | modshogun_wrap.cxx:76885:124: error: invalid conversion from 'volatile void*' to 'void*' [-fpermissive] | 20:39 |
blackburn | sonney2k: ooop! | 20:39 |
blackburn | volatile void?? | 20:39 |
@sonney2k | well we don't need that function right? | 20:40 |
@sonney2k | so %ignore it | 20:40 |
@sonney2k | and the data type | 20:40 |
@sonney2k | but yes volatile seems necessary | 20:40 |
blackburn | sonney2k: I do not understand why it appears | 20:41 |
@sonney2k | blackburn, swig ... | 20:41 |
-!- gsomix [~gsomix@188.168.2.163] has joined #shogun | 20:41 | |
blackburn | ahh so it added new struct? | 20:41 |
gsomix | good evening | 20:41 |
gsomix | sonney2k, around? | 20:41 |
gsomix | sorry I was gone .__. | 20:42 |
@sonney2k | gsomix, hey there | 20:42 |
@sonney2k | blackburn, please just use %ignore for the struct and lock functions | 20:42 |
gsomix | sonney2k, I have some problems with directors. | 20:42 |
blackburn | sonney2k: isn't IGNORE_IN_CLASSLIST enough? | 20:42 |
@sonney2k | no need to have that in interfaces | 20:42 |
@sonney2k | blackburn, that is only for serialization | 20:42 |
blackburn | argh | 20:42 |
blackburn | ok | 20:43 |
@sonney2k | gsomix, I also have problems with directors | 20:43 |
gsomix | huh | 20:43 |
gsomix | my problems related to dotfeatures | 20:43 |
gsomix | *is | 20:44 |
@sonney2k | gsomix, don't stop there | 20:44 |
gsomix | mmm? but this is important | 20:44 |
@sonney2k | gsomix, == please continue | 20:45 |
gsomix | ah, ok | 20:45 |
@sonney2k | gsomix, I am waiting for a proper problem description | 20:45 |
@sonney2k | so far there is no way anyone can help | 20:45 |
blackburn | argh still fails with refs locking :( | 20:46 |
@sonney2k | blackburn, hmmh - but it works with nthreads = 1 | 20:49 |
@sonney2k | ?? | 20:49 |
blackburn | sonney2k: wait was typo here | 20:49 |
@sonney2k | heh | 20:49 |
blackburn | hmm | 20:49 |
blackburn | no | 20:49 |
blackburn | still | 20:49 |
@sonney2k | blackburn, with threads == 1 all good? | 20:49 |
blackburn | sonney2k: yeah.. | 20:50 |
@sonney2k | blackburn, can you do one test | 20:51 |
@sonney2k | increase refcount of these SGReferenced data stuff to say 1000 or so | 20:51 |
@sonney2k | if the problem still appears then it must be sth else | 20:52 |
blackburn | compiling | 20:53 |
gsomix | Ok, we have add_to_dense_sgvec(blabla, SGVector<float64_t>, blabla) method. If this method is called form C++ side, we get SGVector object on python side. Instead numpy array. | 20:54 |
gsomix | SGVector have `+=' and `+' operators, so: | 20:54 |
gsomix | SGVector+=numpy_array - works proper | 20:54 |
gsomix | SGVector+=SGVector - fail, because SGVector is not numpy array, as required by typemaps. | 20:54 |
gsomix | I tried to use directorin/out typemaps, but it's fail. It seems there is bug in swig2.0.4. Next I installed swig2.0.7 and got segfault in compilation. | 20:54 |
gsomix | And last. I tried to use directored features in svm - fail: | 20:54 |
gsomix | >> dsvm.train() | 20:54 |
gsomix | TypeError: not a numpy vector of appropriate type | 20:54 |
gsomix | I think it's related to method with SGVector<>. | 20:54 |
gsomix | sonney2k, that's all. so floody .___. | 20:54 |
gsomix | *related to methods | 20:55 |
blackburn | gsomix: what is svm you have been using? | 20:56 |
gsomix | blackburn, LibLinear | 20:58 |
blackburn | ok | 20:58 |
@sonney2k | gsomix, python buffer protocol certainly would workaround that | 21:00 |
shogun-buildbot | build #30 of deb3 - modular_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/30 blamelist: Soeren Sonnenburg <sonne@debian.org> | 21:01 |
shogun-buildbot | build #6 of nightly_none started, including [] | 21:01 |
@sonney2k | gsomix, btw as workaround one should be able to create a SGVector from numpy and so then add things | 21:02 |
gsomix | >> create a SGVector from numpy | on python side? | 21:02 |
gsomix | but typemaps do not allow such a method. | 21:03 |
@sonney2k | gsomix, v=SGVector(numpymatrix) | 21:05 |
@sonney2k | I think it is called RealVector(numpymatrix) | 21:05 |
@sonney2k | blackburn, yay! buildbot email notification works now :)) | 21:06 |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has joined #shogun | 21:06 | |
blackburn | sonney2k: bah.. that didn't help | 21:10 |
blackburn | okay so I have no idea then | 21:11 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 21:13 | |
shogun-buildbot | build #6 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/6 | 21:13 |
shogun-buildbot | build #9 of nightly_default started, including [] | 21:13 |
blackburn | sonney2k: when I added debug print it worked | 21:18 |
blackburn | :D | 21:18 |
blackburn | heisenbug | 21:18 |
@sonney2k | blackburn, debug prints are kind of sleep delays... | 21:27 |
blackburn | yeah I think I know what is wrong | 21:28 |
@sonney2k | ? | 21:28 |
blackburn | I didn't synchronized deletion | 21:28 |
blackburn | so one sets to 0 | 21:28 |
blackburn | and other sets to -1 | 21:28 |
blackburn | but both try to delete | 21:28 |
blackburn | my guess that is | 21:29 |
-!- alexlovesdata_ [c25faeba@gateway/web/freenode/ip.194.95.174.186] has quit [Quit: Page closed] | 21:29 | |
blackburn | yeah | 21:29 |
blackburn | works | 21:29 |
gsomix | sonney2k, ok, thanks. I'll try it tomorrow. | 21:30 |
gsomix | sonney2k, so what about swig2.0.7? segfault looks strange | 21:31 |
blackburn | we should help swig and report about segfault I think :) | 21:32 |
blackburn | however that is not swig | 21:32 |
blackburn | it is g++ breaks, right? | 21:32 |
gsomix | yep | 21:32 |
blackburn | huh | 21:32 |
blackburn | g++ dev team would be happy to try to recognize the error in 300K LoC file | 21:33 |
blackburn | :D | 21:33 |
blackburn | however I know one way I think | 21:33 |
blackburn | one need to compare modshogun.cxx produced by swig 2.0.4 | 21:34 |
blackburn | and by swig 2.0.7 | 21:34 |
gsomix | aha | 21:35 |
gsomix | wow, thunderstorm | 21:41 |
gsomix | it's good that I'm in the big city now :) | 21:41 |
shogun-buildbot | build #9 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/9 | 21:46 |
shogun-buildbot | build #5 of nightly_all started, including [] | 21:46 |
blackburn | shogun-buildbot: work | 21:48 |
CIA-18 | shogun: Sergey Lisitsyn master * rf1a01b7 / (3 files in 3 dirs): Made refcounting in SGReferencedData thread-safe - http://git.io/SieMkA | 21:48 |
-!- hoijui [~hoijui@dslb-088-074-106-085.pools.arcor-ip.net] has joined #shogun | 22:07 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 22:10 | |
n4nd0 | gsomix: what kind of news we will have to today? :) | 22:13 |
gsomix | n4nd0, do you mean that I should to write weekly report? okay. :] | 22:15 |
gsomix | all news in log | 22:15 |
n4nd0 | haha no no | 22:15 |
n4nd0 | I don't have kick-ass mode | 22:15 |
n4nd0 | :P | 22:15 |
gsomix | but I should :) | 22:16 |
n4nd0 | I am a fan of your comments type "good news everyone" | 22:16 |
n4nd0 | for real, I think they are original :) | 22:16 |
@sonney2k | blackburn, dammed looks like my backup machines power supply died | 22:17 |
@sonney2k | likely the thunderstorm caused this | 22:17 |
blackburn | what is wrong with it? | 22:17 |
@sonney2k | blackburn, turn compu on 'tsss' noise -> off | 22:17 |
shogun-buildbot | build #5 of nightly_all is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/5 | 22:18 |
shogun-buildbot | build #26 of deb1 - libshogun started, including [f1a01b70ec1519a0d586f5588d55b9927df84fc7] | 22:18 |
blackburn | sonney2k: what?! | 22:18 |
@sonney2k | blackburn, ? | 22:18 |
blackburn | sonney2k: so it produces no voltage or what? | 22:18 |
@sonney2k | well for <1 second it does | 22:19 |
@sonney2k | could be that the PSU is OK and the board or whatever is RIP | 22:19 |
@sonney2k | but how can I find out :/ | 22:19 |
@sonney2k | I have to buy a new one to find the cause | 22:19 |
@sonney2k | and most likely is the psu | 22:20 |
shogun-buildbot | build #26 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/26 | 22:20 |
shogun-buildbot | build #34 of deb2 - static_interfaces started, including [f1a01b70ec1519a0d586f5588d55b9927df84fc7] | 22:20 |
blackburn | sonney2k: measure voltage of PSU runned separately | 22:20 |
gsomix | n4nd0, I'll try to `repass' an exam tomorrow. This will increase my uni stipend for 10$ per month. :D | 22:21 |
n4nd0 | gsomix: cool, I wish you good luck with it. What is the subject? | 22:21 |
gsomix | optics, optics, again optics | 22:21 |
gsomix | tnx :) | 22:22 |
n4nd0 | you must be an expert on the topic | 22:22 |
blackburn | sonney2k: just shortcircuit black and green or so (shoold be googleable) :) | 22:22 |
gsomix | n4nd0, this is my future work, hehe | 22:22 |
@sonney2k | blackburn, :P | 22:23 |
blackburn | sonney2k: seriously - I did that twice or so | 22:23 |
@sonney2k | ???? | 22:24 |
blackburn | just run your PSU w/o anything and you will get whether it works | 22:24 |
@sonney2k | blackburn, how so? | 22:24 |
@sonney2k | it dies under load | 22:24 |
blackburn | sonney2k: are you sure it dies under load? | 22:24 |
@sonney2k | blackburn, yes | 22:24 |
blackburn | hmm | 22:25 |
@sonney2k | at least the mainboard led is on / network led too afterwards | 22:25 |
@sonney2k | so the board seems to be OK | 22:25 |
@sonney2k | but who knows... | 22:25 |
@sonney2k | blackburn, I was just looking at your patch | 22:25 |
blackburn | and? | 22:25 |
@sonney2k | I am a bit concerned that init_data() is called in a pthread lock thing | 22:26 |
blackburn | why? | 22:26 |
@sonney2k | (which is a busy loop) | 22:26 |
@sonney2k | better store the refcount before | 22:27 |
blackburn | init_data is basically two 'mov' in all our structures | 22:27 |
@sonney2k | then free etc | 22:27 |
blackburn | free should be inside LOCK | 22:27 |
@sonney2k | no | 22:27 |
blackburn | whoops | 22:28 |
blackburn | wrong here for sure | 22:28 |
@sonney2k | I mean if it is 0 | 22:28 |
@sonney2k | then there really is no one else fiddling with the object | 22:28 |
@sonney2k | not now and cannot be in future | 22:28 |
blackburn | sonney2k: bug here anyway | 22:28 |
blackburn | return before unlock | 22:28 |
@sonney2k | yes | 22:29 |
@sonney2k | better do int c = m_refcount->rc; directly after the decrement | 22:29 |
@sonney2k | and then unlock | 22:29 |
@sonney2k | and do whatever necessary | 22:29 |
blackburn | I feel too lazy to think about it | 22:29 |
blackburn | will just check whether it works | 22:29 |
@sonney2k | blackburn, err what is line 107ff doing? | 22:30 |
blackburn | sonney2k: so you mean in thready case rc will never be zero? | 22:30 |
@sonney2k | shouldn't it be ifndef ? | 22:30 |
blackburn | oops | 22:30 |
@sonney2k | blackburn, no why? | 22:31 |
blackburn | sonney2k: I am concerned with free out of pthread lock | 22:31 |
@sonney2k | blackburn, if the refcount is 0 | 22:32 |
@sonney2k | then we free right? | 22:32 |
blackburn | yes | 22:32 |
@sonney2k | so it means that this is the only object still available | 22:33 |
blackburn | sonney2k: but how do we guarantee we free only once? | 22:33 |
@sonney2k | no one else can have one | 22:33 |
blackburn | looks like | 22:33 |
blackburn | sonney2k: what if we have two copies | 22:35 |
blackburn | ref = 2 | 22:35 |
blackburn | they both decrement ref | 22:35 |
blackburn | to zero | 22:35 |
@sonney2k | that cannot happen | 22:35 |
@sonney2k | one will decrement first | 22:35 |
@sonney2k | so only one of them can call free | 22:35 |
blackburn | sonney2k: should be better now | 22:40 |
CIA-18 | shogun: Sergey Lisitsyn master * r8bc0fdf / src/shogun/lib/SGReferencedData.h : Updated references thread-safe handling - http://git.io/22qlmw | 22:40 |
shogun-buildbot | build #34 of deb2 - static_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/34 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 22:42 |
shogun-buildbot | build #31 of deb3 - modular_interfaces started, including [f1a01b70ec1519a0d586f5588d55b9927df84fc7] | 22:42 |
blackburn | are you kidding??! | 22:44 |
@sonney2k | ? | 22:46 |
blackburn | sonney2k: it hangs on classifier_libsvm | 22:46 |
-!- hoijui [~hoijui@dslb-088-074-106-085.pools.arcor-ip.net] has quit [Quit: Leaving] | 22:54 | |
@sonney2k | gsomix, when is your exam tomorrow? | 23:09 |
-!- naywhaya1e is now known as naywhayare | 23:09 | |
gsomix | sonney2k, 10 UTC. | 23:10 |
@sonney2k | gsomix, ok then good luck for that :) | 23:10 |
@sonney2k | gsomix, do you have time somewhen afterwards to chat? | 23:11 |
gsomix | of course. e.g. tomorrow | 23:11 |
@sonney2k | when? | 23:12 |
gsomix | after 15UTC, is this convenient for you? | 23:13 |
@sonney2k | I only have time before that or later in the evening | 23:14 |
gsomix | sonney2k, ok, than later in the evening. I'll be available. | 23:15 |
-!- heiko [~heiko@host86-177-176-121.range86-177.btcentralplus.com] has joined #shogun | 23:31 | |
CIA-18 | shogun: Soeren Sonnenburg master * r053bdde / src/shogun/lib/SGReferencedData.h : destroy lock on free and fix race condition - http://git.io/ZDWpZQ | 23:32 |
blackburn | ah destroy.. | 23:33 |
blackburn | yeah makes sense | 23:33 |
@sonney2k | gsomix, please still send the weekly report... | 23:34 |
blackburn | sonney2k: btw I would like to get rid of spinlocks | 23:34 |
@sonney2k | blackburn, no! | 23:34 |
@sonney2k | why that? | 23:34 |
gsomix | sonney2k, aha | 23:34 |
blackburn | they are unsafe - 100% cpu load for ever in case of glitch | 23:34 |
@sonney2k | why unsafe? | 23:34 |
@sonney2k | pthread mutexes have huge overhead | 23:35 |
@sonney2k | in particular for these *fast* refcount inc/decrefs | 23:35 |
@sonney2k | spinlocks are much much better suited | 23:35 |
shogun-buildbot | build #31 of deb3 - modular_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/31 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:36 |
shogun-buildbot | build #27 of deb1 - libshogun started, including [8bc0fdfbf8a21ffd83a772043a3bd187f31223ca, 053bdde484b5f277de9acc7482931ee2fecb25e6] | 23:36 |
blackburn | sonney2k: we don't need fast in sgobject in particular | 23:37 |
blackburn | I'll become mad soon - when will it work finally? | 23:37 |
@sonney2k | well we use it to compute combined kernels etc | 23:37 |
blackburn | sonney2k: any idea why it hangs? | 23:38 |
@sonney2k | blackburn, relax and do sth else maybe - doc stuff? | 23:38 |
shogun-buildbot | build #27 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/27 | 23:38 |
shogun-buildbot | build #35 of deb2 - static_interfaces started, including [f1a01b70ec1519a0d586f5588d55b9927df84fc7] | 23:38 |
blackburn | sonney2k: I have gsoc project here you know :D | 23:38 |
@sonney2k | blackburn, ahum that is why you do SGSparse :D | 23:39 |
blackburn | sonney2k: I do not really remember why did I do that | 23:39 |
blackburn | ahhh chris pushed me in these swarms full of monsters and dragons | 23:39 |
@sonney2k | I guess because you promised at some point and then because you merged the broken commit from chris? | 23:39 |
blackburn | oh you are so positive :D | 23:41 |
@sonney2k | blackburn, I finally know why R fails | 23:42 |
blackburn | why? | 23:42 |
@sonney2k | it requires R >= 2.13 | 23:42 |
@sonney2k | and on bb is 2.11 | 23:42 |
@sonney2k | it would have worked with 2.11 if we had the old .r-install.sh script though | 23:42 |
blackburn | no idea about it | 23:43 |
@sonney2k | well R 2.11 is from 2010 | 23:43 |
@sonney2k | and they do such nice things like renaming .saveRDS to saveRDS | 23:43 |
blackburn | sonney2k: I have seen R code a few times | 23:45 |
blackburn | and I don't want to anymore | 23:46 |
blackburn | :D | 23:46 |
@sonney2k | I finally understand why it worked all good here | 23:46 |
@sonney2k | R 2.15 ... | 23:46 |
@sonney2k | I think we need to support the old version again :/ | 23:49 |
gsomix | good night guys | 23:53 |
n4nd0 | good night gsomix | 23:56 |
-!- heiko [~heiko@host86-177-176-121.range86-177.btcentralplus.com] has left #shogun [] | 23:57 | |
@sonney2k | nite | 23:58 |
shogun-buildbot | build #35 of deb2 - static_interfaces is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/35 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:59 |
shogun-buildbot | build #36 of deb2 - static_interfaces started, including [8bc0fdfbf8a21ffd83a772043a3bd187f31223ca, 053bdde484b5f277de9acc7482931ee2fecb25e6] | 23:59 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed] | 23:59 | |
--- Log closed Tue Jul 03 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!