n4nd0 n4nd0 n4nd0 --- Log opened Wed Apr 25 00:00:19 2012 -!- Marty28 [~marty@158.181.78.199] has quit [Quit: Colloquy for iPad - http://colloquy.mobi] 00:05 -!- n4nd0 [5aac282c@gateway/web/freenode/ip.90.172.40.44] has joined #shogun 00:12 good evening guys 00:12 from Spain 00:12 -!- harshit_ [~harshit@182.68.43.52] has joined #shogun 00:16 I will see you tom around here, good night! 00:23 -!- n4nd0 [5aac282c@gateway/web/freenode/ip.90.172.40.44] has quit [Quit: Page closed] 00:23 -!- gsomix [~gsomix@188.168.5.178] has quit [Read error: Operation timed out] 00:27 -!- harshit_ [~harshit@182.68.43.52] has quit [Quit: Leaving] 01:01 -!- vikram360 [~vikram360@117.192.161.188] has quit [Ping timeout: 244 seconds] 03:04 -!- pluskid [~pluskid@li381-56.members.linode.com] has joined #shogun 04:42 * pluskid bought more memory for the laptop, need to install a 64bit system now 04:43 -!- pluskid [~pluskid@li381-56.members.linode.com] has quit [Client Quit] 04:47 -!- pluskid [~chatzilla@111.120.57.192] has joined #shogun 05:23 -!- vikram360 [~vikram360@117.192.161.188] has joined #shogun 05:58 -!- pluskid [~chatzilla@111.120.57.192] has quit [Ping timeout: 252 seconds] 06:04 -!- vikram360 [~vikram360@117.192.161.188] has quit [Ping timeout: 260 seconds] 08:01 -!- uricamic [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun 08:49 -!- pluskid [~chatzilla@111.120.57.192] has joined #shogun 08:52 -!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun 08:53 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 09:06 -!- n4nd0 [d480fcac@gateway/web/freenode/ip.212.128.252.172] has joined #shogun 09:08 good morning! 09:08 n4nd0: hey how was your flight? 09:11 blackburn: it went good 09:11 Is it Apr. 25 or Apr. 26 now in your time zone? 09:11 blackburn: first flight ever seen with so many Smith charts :P 09:11 pluskid: 25 09:12 25th 09:12 so the kick off meeting is on 26th, right? 09:12 pluskid: 27 for you probably 09:12 orr 09:12 no 09:12 are you in utc+8? 09:12 yes 09:13 sonney2k: hey! about the kickoff meeting, my exam starts at 13:00 UTC tomorrow, I hope there is no trouble with that since it starts 11:00 UTC 09:13 sonney2k: do you happen to know how long will it take? 09:13 n4nd0: I don't think it would take more than half of hour 09:13 n4nd0: good luck with your exam 09:13 -!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun 09:13 pluskid: wait but is it 26th in your timezone? 09:13 pluskid: thank you! 09:14 blackburn: great 09:14 I don't think so :) 09:14 it should be 15 now? 09:14 25th, now, in my time zone 09:14 I will ask sonne|work also just in case 09:14 afternoon of 25th 09:14 pluskid: looks like you didn't read the subject of the email 09:14 pluskid: it said 26th 09:14 sonne|work: no he is confused with timezones 09:14 so we check watches :D 09:15 sonne|work: oops, you are right 09:15 -!- vikram360 [~vikram360@117.192.161.188] has joined #shogun 09:15 thanks! 09:15 we are all on the same day (25th now) 09:15 sonne|work: is it a problem if my exam tomorrow starts two hours after the kickoff is scheduled? I mean in terms of how long will it be 09:15 not two hours definitely 09:16 it is clear from the beginning that not everyone can attend / some will have to leave early 09:19 but I have to leave after 1 hr myself so... 09:19 no worries then :) 09:20 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 09:27 -!- pluskid [~chatzilla@111.120.57.192] has quit [Remote host closed the connection] 09:30 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun 09:47 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] 09:47 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 09:47 -!- n4nd0 [d480fcac@gateway/web/freenode/ip.212.128.252.172] has quit [Ping timeout: 245 seconds] 09:59 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 09:59 sonne|work: just a note - we need to discuss cases when one is stucked and unhappy 10:08 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 10:08 during kickoff I think 10:08 -!- n4nd0 [~n4nd0@wifi-cct-cisco-r252-172.cct.urjc.es] has joined #shogun 10:08 blackburn: make a list and remind me if I forget sth :) 10:10 sonne|work: shogun meeting @ berlin, middle of august? 10:13 +1000 to that suggestion :) 10:13 heh I am afraid not this year 10:14 why? 10:14 http://nuit-blanche.blogspot.com/2012/04/compressive-sensing-this-week.html 10:23 sonne|work: n4nd0: ^ 10:23 it is pretty unusual to read about myself hehe 10:23 cool 10:24 blackburn: hehe :) 10:24 who is the blogger? 10:24 no idea 10:25 haha you're famous 10:26 -!- karlnapf [~heiko@host86-180-44-86.range86-180.btcentralplus.com] has joined #shogun 10:42 -!- emre-away is now known as emrecelikten 10:46 -!- karlnapf [~heiko@host86-180-44-86.range86-180.btcentralplus.com] has left #shogun [] 10:50 sonne|work: don't you think tutorial should be a submodule? 10:52 like data 10:52 blackburn: of course but you did fail putting your tutorial there :D 10:54 sonne|work: tutorial of non-existent toolkit? should take more time to finish it 10:54 yes sure - we have to start somehow right? 10:56 sonne|work: what about cherry-picking some stuff from your thesis? 11:00 makes sense but still lots and lots of work... 11:01 sonne|work: can I obtain source? 11:02 sure I can give it to you 11:02 -!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Connection reset by peer] 11:02 sonne|work: you thesis is well-written so it makes sense to cherry-pick some parts like string kernels 11:03 -!- wiking [~wiking@we02c096.ugent.be] has joined #shogun 11:03 -!- wiking [~wiking@we02c096.ugent.be] has quit [Changing host] 11:03 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 11:03 I should obtain chris' thesis heh 11:04 damn my machine crashed... :( 11:07 wiking: do you think dot product of homogay kernel map can be optimized somehow? 11:08 sh I can't stop call it homogay it is not funny at all :D 11:08 blackburn: well homo gay kernel map should be first extended to the new stuff ;) 11:08 blackburn: remember the 2 papers i've sent 11:08 wiking: I mean they use sin and cos right? 11:09 yes 11:09 so in dot product we could reduce it probably? 11:09 ? 11:09 sin(x)*sin(y) = 2 sin(x+y) or so? 11:09 1/2 11:09 heh 11:09 no idea if it worths trying 11:10 mmmm 11:11 now i'm starting to see what you mean ;) 11:11 could be interesting 11:11 wiking: actually it seems we will use some identical multipliers 11:12 so may be we can just precompute squares? 11:12 eh? 11:14 ok now i'm lost blackburn 11:14 so expand your story :) 11:14 each feature is being multiplied 11:14 with sin and cos and 1? 11:14 not exactly this straightforward 11:15 ah I checked code and it is not as simple as I thought hehe 11:16 wiking: okay I am lost :D 11:17 but the idea stays - is dot product easier to compute than full mapping? 11:17 it needs to be checked 11:18 wiking: have you time for that? 11:18 yeah no worries 11:19 i can hope on that train ;P 11:19 I'm going to start with slep porting tomorrow 11:19 yeey 11:20 shogun: Heiko Strathmann master * rb47d55b / src/shogun/features/SimpleFeatures.cpp : fixed chache bug reported by - http://git.io/su1RUQ 11:27 shogun: Heiko Strathmann master * r0afeee4 / src/shogun/features/SimpleFeatures.cpp : 11:27 shogun: Merge pull request #485 from karlnapf/master 11:27 shogun: bug fix reported by Koen van de Sande - http://git.io/nY3K5A 11:27 Hey guys, what's the primary difference between a Markov chain and a Markov random field? One is a probabilistic FSM while the other is like a undirected Bayesian network? 11:27 emrecelikten, are you just wondering or do you have to choose between both of them for something in particular? 11:31 n4nd0: Wondering. Could MRFs be applied to speech/music domain like HMMs for example 11:32 I think so 11:32 emrecelikten, I am not sure, I have never really worked with MRFs far beyond definition and toy examples 11:33 emrecelikten: markov chain is a chain right? :) 11:33 while random field is a graph structure 11:33 emrecelikten, I don't really see direct application of MRF to it but it might be 11:33 blackburn, isn't it a chain a particular case of directed weighted graph? 11:34 n4nd0: yes but it is a chain in markov chains - not a graph 11:34 Hmm thanks 11:35 blackburn, I partially agree, I think that a graph is something more general than a chain 11:36 yes exactly what I mean 11:36 but I'd say that a chain is a graph 11:36 it is 11:36 acyclic with nodes with only two or one output edge ? 11:37 something like that I guess 11:38 a Markov chain can be cyclic I think 11:39 right sorry 11:39 no problem :) 11:39 but no multiple outputs 11:39 yeah, they can have multiple 11:39 if today is sunny 11:40 wait but when I am fucked up - what is the difference then? 11:40 tomorrow might be rainy, sunny, cloudy or pouring cats and dogs for example 11:40 there you have more than two output edges 11:40 the difference between which ones? 11:40 MC and MRF 11:41 I think MRF are something similar to Bayesian nets 11:41 so for example the graph that represent them is not weighted 11:41 there must be more differences 11:42 at least I have found them in different realms 11:42 damn I should get back to markov stuff and study it :D 11:42 emrecelikten: probably I am wrong 11:42 I think MRF is more advanced stuff 11:44 one type of probabilistic graphical model 11:44 blackburn: It's fine :) 11:44 I have to learn more of them for the project since I'll have to implement conditional random fields 11:44 blackburn: Longtimenostudy effect :P :D 11:45 emrecelikten: oh it's been a long without some effects and paradoxes ;) 11:45 MRF supports higher dimensions I think 11:45 blackburn: :P 11:45 -!- vikram360 [~vikram360@117.192.161.188] has quit [Read error: Connection reset by peer] 11:46 -!- vikram360 [~vikram360@117.192.161.188] has joined #shogun 11:46 * blackburn is reading bishop book 11:48 n4nd0: Are you taking PGM class from Coursera? 11:49 emrecelikten, I didn't get time to do it now but I would like to follow it during summer 11:49 emrecelikten, are you? 11:49 n4nd0: Yes, but its difficult 11:50 emrecelikten, do you recommend it? 11:50 -!- uricamic [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Ping timeout: 245 seconds] 11:50 n4nd0: Just failed my first programming assignment last night, very time consuming (and slightly annoying)... But its a good class 11:51 oh 11:51 n4nd0: I have seen a "proof by example" though, so I should warn you in advance if you like rigor :D 11:54 -!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has joined #shogun 12:00 -!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has left #shogun [] 12:00 -!- gsomix [~gsomix@85.26.164.208] has joined #shogun 12:24 sonney2k, around? 12:29 sonne|work, 12:29 -!- vikram360 [~vikram360@117.192.161.188] has quit [Ping timeout: 260 seconds] 12:30 gtg guys, see you later! 12:31 n4nd0: see you 12:31 -!- n4nd0 [~n4nd0@wifi-cct-cisco-r252-172.cct.urjc.es] has quit [Quit: Ex-Chat] 12:33 -!- vikram360 [~vikram360@117.192.163.57] has joined #shogun 12:47 -!- gsomix [~gsomix@85.26.164.208] has quit [Ping timeout: 272 seconds] 12:51 -!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] 13:08 -!- gsomix [~gsomix@188.168.5.195] has joined #shogun 13:14 gsomix: yes? 13:23 sonne|work, I'm so sorry, but I not able go to kickoff meeting, because I have classes at this time. What to do? I can prepare text for tomorrow, if it's needed. 13:24 sonne|work, moin 13:24 gsomix: not a problem - you can read the chat logs later 13:24 gsomix: it was clear from the beginning that not everyone can attend 13:24 sonne|work, thanks a lot. :) 13:24 it is very hard to schedule a meeting for 16 participants w/o planning years into the future :) 13:25 sonne|work, huh, I see. 13:26 -!- harshit_ [~harshit@182.68.43.52] has joined #shogun 13:39 * gsomix http://www.planetaryresources.com/mission/ 13:48 -!- vikram360 [~vikram360@117.192.163.57] has quit [Ping timeout: 240 seconds] 13:49 -!- vikram360 [~vikram360@117.192.189.186] has joined #shogun 13:49 gsomix: So it was truly asteroid mining then 13:51 Why don't we start constructing a Dyson sphere instead :P 13:52 -!- harshit_ [~harshit@182.68.43.52] has quit [Remote host closed the connection] 14:05 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 14:05 -!- pluskid_ [~chatzilla@173.254.214.60] has joined #shogun 14:16 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun 14:17 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] 14:17 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 14:17 -!- pluskid_ is now known as pluskid 14:19 mmm 14:24 pluskid: here? 14:24 wiking: hi 14:25 oh great hi! 14:26 so i was just wondering about your refactoring of multi class 14:26 as i'm now implementing the latent svm 14:27 so what's your question? 14:27 it's a derivative of LinearMachine 14:27 and as a first step it's binary 14:27 hmm 14:28 but if i want to extend it to multi class machine, do we have line LinearMulticlass? 14:28 does your latent svm support multiclass *naturally* ? 14:28 I mean 14:29 for example, traditional svm only support binary case 14:29 but we can use one-vs-rest or one-vs-one to make it multiclass 14:29 -!- blackburn [~qdrgsm@188.168.2.179] has joined #shogun 14:29 pluskid: no it does not support multi class naturally 14:29 -!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has joined #shogun 14:29 if your latent svm also rely one one-vs-res or one-vs-one for multiclass 14:30 but i'd like it to be able to use for multi class case as well 14:30 then I guess you don't have to do anything 14:30 ah ok 14:30 because MulticlassLinearMachine can accept a "base linear machine" 14:30 yeah as a first step i guess it should be OvR OvO 14:30 and turn that base linear machine into multiclass 14:30 ok 14:30 that's great 14:31 there's an example code I guess 14:31 and in case i can extend it to a true multi class machine? 14:31 i.e. like a crammer-singer 14:31 if it becomes a true multiclass machine, you will derive a subclass from MulticlassLinearMachine 14:31 it cannot be subclass of both linearmachine and multiclasslinearmachine 14:32 that's our problem 14:32 also happens in MulticlassSVM 14:32 aha 14:32 cannot be subclass of SVM and MultickassKernelmachine simultaneously 14:32 but why not? :) 14:32 we do not like multiple inheritance here 14:32 i mean i guess the current design does not allow it 14:32 why? :) 14:32 -!- n4nd0 [~n4nd0@wifi-cct-cisco-r252-4.cct.urjc.es] has joined #shogun 14:33 it's not that of a 'big magic' :) 14:33 because sonney2k hate multiple inheritance :p 14:33 joking haha 14:33 I guess if we really want multiple inheritance, we will need "virtual inheritance" 14:34 :>>> 14:34 not necesarrily 14:34 but of course it depends a lot on the base classes 14:34 because both super class has a common super class (CSGObject) 14:34 ah yeah 14:34 heheh in that case yes 14:35 the inheritance graph becoms complicated 14:35 yeah a diamond inharitance 14:35 yes, like std::iostream ? :p 14:36 anyhow that's another story for later 14:36 first i want to see how it performs with a OvR setup 14:36 and as you said that should be fairly easy to do 14:37 wiking: check this example 14:37 https://github.com/shogun-toolbox/shogun/blob/master/examples/undocumented/libshogun/classifier_multiclasslinearmachine.cpp 14:37 it use a LibLinear as a base classifier 14:37 and use OneVsOne to turn it into multiclass 14:37 what you need is to new a CMulticlassOneVsOneStrategy() haha 14:37 there will be ECOC support soon :) 14:41 great! 14:41 thanks a lot i'll do the testing with this 14:41 i mean based on this 14:41 :) 14:41 yeeey this should be fun to see how it performs :))) 14:41 I didn't tested the performance of the shogun machines really 14:41 be back in a while 14:41 wiking: hey what is your issue? 14:44 blackburn: well i just want to have multi class support for latent svm ;) 14:45 wiking: hmm can you describe in a few words how does it extend? 14:46 well currently as you can see only with OvR 14:46 but i must check whether i can use crammer-singer 14:47 but as for now i don't see a problem with it 14:47 wiking: crammer-singer is done in both ocas and liblinear multiclass svms 14:47 yeah i know 14:47 and since i currently use ocas 14:47 it should be fine 14:48 wiking: I already asked vojtech about some improvement of ocas btw 14:48 but then i guess i have to the the 'reimplementation' of latent svm 14:48 I need to solve 0.5||w||^2 + + C \sum_i \xi_i problems 14:48 as current implementation is derived from linearmachine 14:49 (domain adaptation) 14:49 wiking: yes sad truth 14:49 ah w_0 is the input learned w? 14:49 linear machine is pretty 'far away' from linearmulticlassmachine 14:49 wiking: yes kind of pre-svm 14:49 source domain svm 14:49 yep 14:49 (but in multiclass case) 14:49 so \sum_m w_m everywhere 14:50 btw: i see that we have SVMOcas for binary case 14:52 but there's no such for multi class case 14:52 ah no 14:53 i'm wrong 14:53 we do 14:53 wiking: multiclass/MulticlassOCAS 14:53 yes yes found it sorry 14:53 wiking: I have pretty recently implemented it actually 14:53 that should be interesting to test 14:53 i'll have to change the api of my implementation as far as i can see 14:54 since maybe the best thing is to somehow supply the svm you want to use 14:54 for latent svm 14:54 but i'm wondering how will an s-svm look like 14:54 n4nd0 will tell me more about that soon ;) 14:54 yeah once he is finished with some univ stuff :) 14:55 yeah :) 14:55 :>>> 14:55 wiking, in any case, yesterday I devoted some time to read some of the Struct SVM's doc 14:55 -!- pluskid_ [~chatzilla@111.120.57.192] has joined #shogun 14:55 wiking, I like some of the diagrams they have there and I think that it is exactly what we want to do in shogun 14:56 :)))) 14:56 I am waiting for an answer from Nico to know that though 14:56 wiking, have you taken a look to Struct SVM? 14:56 yep 14:56 cool 14:56 -!- pluskid [~chatzilla@173.254.214.60] has quit [Read error: Connection reset by peer] 14:56 currently i'm doing runs 14:56 n4nd0: which diagrams? 14:57 my implementation vs the other latent implementation 14:57 -!- pluskid_ is now known as pluskid 14:57 blackburn, a moment 14:57 that is based on structsvm 14:57 but i just have -1,1 labeling so i don't use SO 14:57 blackburn, just scroll down a bit http://svmlight.joachims.org/svm_struct.html 14:57 wiking, I have to think about how to fit into shogun the labels hierarchy we need for SO 14:58 hmm where? 14:58 n4nd0: i'm changing as well labels... 14:59 since i have latent output 14:59 so it might be good to talk about it when u have some time 14:59 n4nd0: are you sure it is the link? 14:59 blackburn, maybe diagrams is a bad word to refer to them 14:59 I see only text :) 15:00 blackburn, the "coloured boxes" on the left with arrows in between 15:00 a wait 15:00 http://www.cs.cornell.edu/~tomf/svmpython2/ 15:00 ah 15:01 the link remains the same in the browser's textbox for the address :D 15:01 yes any kind of diagrams help 15:01 I think that SVM struct is a good reference for what we want to do 15:02 n4nd0: yes yes 15:02 but I wonder if I got something wrong since it was not referenced in the project ideas 15:02 pretty general 15:02 that's why I asked Nico about it, just to be sure 15:02 hmm I don't think thorsten has fucked up 15:03 :D 15:03 :D 15:03 n4nd0: wiking: it should be of interest right? http://svmlight.joachims.org/svm_struct.html 15:05 blackburn: well yeah that's a typical structural svm 15:05 I am very excited about having all that jazz in shogun 15:06 hehehe 15:06 it's going to be very cool 15:07 what will you use as labels in SO learning? 15:10 I mean the current CLabels wouldn't be enough to describe structures 15:10 pluskid: yeah it needs to be extended 15:10 btw when are we planning to do a release? 15:10 i mean shogun release 15:10 any concrete idea? I'm thinking of extending CLabels for multiclass, too. though it can support multiclass in the current status 15:11 pluskid: yes would be nice to have multiclass lables 15:11 probabilities is the thing we can store there 15:11 wiking: that would be hard to release before end of gsoc currently 15:12 heheh ok 15:12 afaik the last release was 8 months ago 15:12 however this can be done 15:12 and there's a lot of improvement/fixes since then 15:12 yes but we are in unstable state 15:12 true 15:12 all is broken 15:12 :D 15:12 api is changing a lot :)) 15:12 have you considered some benchmark test for our machines in shogun? 15:13 I'd call it shogun 2.0 15:13 :D 15:13 pluskid: what to benchmark? 15:13 for examples, all the examples only run classification etc. 15:13 pluskid: i do with my own... 15:13 but do not evaluate 15:13 performance evaluation? 15:13 yes 15:13 that would also be a good reference when we are doing refactoring of the code 15:14 if the performance are different before and after, there must be something wrong 15:14 ah you mean to compare things using there benchmarks? 15:14 yes 15:14 I don't think it would work very well 15:15 and this is also useful for the users to choose machines to use 15:15 blackburn: why? 15:15 pluskid: what would you measure for classifiers? 15:15 for example, accuracy on a standard dataset 15:15 but why not to compare learned parameters? 15:16 for the purpose of refactoring, your way is better. But performance is more intuitive to see, and has other uses 15:17 I mean, for example, I'm a user 15:18 of shogun 15:18 pluskid: where would you like to add it? 15:18 I downloaded the shogun, compile it and run the example code 15:18 when I see it produces performance as indicated in the reference, I would be sure that everything is fine, and I can start my work 15:18 blackburn: no idea yet, just thinking about this :p 15:19 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 15:23 something like this: http://yuml.me/f14284c4.jpg 15:29 pluskid: hehehe 15:29 pluskid: binary=real 15:30 ah 15:30 regression 15:30 hmm 15:30 sorry no 15:30 makes sense 15:30 pluskid: btw multiple output regression is something real too 15:30 so RealVectorLabels 15:30 -!- wiking [~wiking@we02c096.ugent.be] has joined #shogun 15:30 -!- wiking [~wiking@we02c096.ugent.be] has quit [Changing host] 15:30 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 15:30 hmm, make sense 15:31 the problem is 15:31 how to implement it 15:31 we have get_labels() 15:31 what should it return? 15:31 I see templates as a choice actually 15:31 get_labels() is a function of what object? 15:31 of labels 15:31 or get_label(i) 15:31 sonney2k might hate template? :p 15:31 no, I don't think 15:32 I can't see any other way 15:32 pluskid: how did you draw it? 15:32 blackburn: http://yuml.me/edit/f14284c4 15:33 all the uml stuff is so damn heavy 15:33 that's fscking awesome 15:33 haha 15:33 this is not *drawing* , though :p 15:34 yeah 15:34 hehehe 15:35 you can include LatentLabels as well ;P 15:36 and it's not drawing for me 15:36 wiking: what would a latent label looks like? 15:36 pluskid: well it can be basically anything 15:36 it's user defined 15:36 hmm 15:37 since the latent variable should be defined by the user 15:37 we are going to have great hierarchy there 15:37 I am afraid we need shogun/labels 15:37 yes, I feel very strange the first time when I found CLabels in features folder :D 15:37 sonney2k: it is going to be tough to support back compatibility for this release 15:37 everything will be broken 15:38 pluskid: did the yuml.me code generate you the diagram? 15:38 maybe everything is already broken ? :p 15:38 is already and will be more 15:38 wiking: yes, scroll down the page you will see links to the pictures 15:39 of various type, png, pdf, jpg, svg, etc 15:39 lool 15:39 yeah that works 15:39 but not within the page 15:39 pluskid: would you mind describing multiclass part @ wikipedia? 15:46 wikipedia of shogun? 15:46 yes 15:46 http://en.wikipedia.org/wiki/Shogun_(toolbox) 15:46 http://en.wikipedia.org/wiki/Shogun_(toolbox) 15:47 mmm 15:47 funny 15:47 it missed for me the ) 15:47 hmm, OK 15:47 I'll update it after I add each new multiclass support 15:48 algorithm 15:48 -!- vikram360 [~vikram360@117.192.189.186] has quit [Ping timeout: 260 seconds] 15:49 I'm going to add some dim reduction part there once 15:49 simply lets make it best article of wikipedia 15:49 haha 15:49 ambitious goal 15:49 no - we know this field 15:50 :) 15:50 n4nd0: wiking: join me in this effort ;) 15:52 hehehehe 15:52 here's some pseudo code: https://gist.github.com/2489859 15:52 of new CLabels 15:52 we need promotion - it worths doing 15:52 but the problem is that 15:52 in the end your names will be well known :) 15:53 if we use template here, they no longer share a *single* super class 15:53 pluskid: I would skip binary labels 15:53 it is inconvient at some point 15:54 i.e. you train svm using binary 15:54 but it will always return real 15:54 hmm 15:54 that's why we have both LabelStoreType and LabelType template variable 15:55 ah 15:55 brb 15:55 -!- harshit_ [~harshit@182.68.43.52] has joined #shogun 16:15 wiking: last release was on 1, December 2011 btw 16:37 :> 16:39 -!- emrecelikten [~emre@176.40.232.233] has quit [Quit: Leaving.] 16:47 -!- harshit_ [~harshit@182.68.43.52] has quit [Quit: Leaving] 17:07 blackburn, wiking: Nico has approved the diagrams I linked you guys, svm struct may be an inspiration for our SO base 17:14 hehehe 17:14 n4nd0: heh 17:14 nice 17:14 can u do a sketch of that 17:14 how that actually would look like in case of shogun 17:14 * wiking is lazy ;) 17:14 wiking, yes, we have discussed and would like to prepare some UML diagrams and stuff 17:15 i mean in my case 17:15 i'm using psi function as well 17:15 but it's 17:15 PSI(x,y,h) 17:15 what is your h? 17:15 so "maybe" it's a good idea to sync 17:16 h = latent variable 17:16 yes, it's a good idea 17:16 I don't know much about latent variables but 17:17 n4nd0: psi is defined by the user right? 17:17 n4nd0: i mean in case of SO 17:17 wiking, yes 17:17 so you'll have somewhere in the class of features or something 17:17 that is user defined 17:17 wiking, just a kind of interface hast to be followed 17:17 currently my CLatentFeatures is something like that 17:17 you add x, y and h 17:17 and you have to define PSI 17:17 so maybe that could be used 17:18 just then we need to rename it 17:18 so that it works for SO 17:18 PSI(x,y) 17:18 I think that for SO h should be within x 17:19 I mean, the general algorithm shouldn't really case whether some features there are latent or not right? 17:19 really care*, sorry 17:19 well 17:20 you can take it like that as well 17:20 for me it looks more unified in that way, what do you think? 17:20 the idea i had is just that we make a new feature or whatever for SO 17:20 -!- vikram360 [~vikram360@117.192.189.186] has joined #shogun 17:20 and from that i'll inherit latent... to be able to introduce the latent variable 17:20 because not CLatentFeatures : public CDotFeatures 17:21 *now 17:21 s/not/now/g ;) 17:21 :) 17:21 so i guess you'd do the same 17:21 but only have x and y 17:21 where y can be more than a simple -1,1 17:21 why to inherit from CDotFeatures? 17:22 and you'll need to be able to define the delta(y,y') as well somehow 17:22 w*PSI(x,y) 17:22 or in other way 17:23 so that's a dot product 17:23 my plan is to have a general class for the labels where that class has a pure virtual method for the cost function (this delta(y, y')) 17:23 ah yeah 17:23 that's cool 17:24 what I have not clear yet is where this class should fit 17:24 so in case of any application of SO 17:24 if inheriting from sth else in Shogun or not 17:24 you'd need to derive from that class and define that pure virtual func 17:24 exactly 17:24 well 17:24 that good but then again 17:25 you can start the refactoring on the level of CLabels 17:25 so that CLabels has a delta as well or? :) 17:25 yes, I wanted to check thoroughly the hierarchy there 17:25 i mean it's straightforward what is the delta 17:25 wiking, but it is application dependent again 17:26 wiking, or at least dependent on the structure used for the labels 17:27 that's why we need SequenceLabels, TreeLabels, etc. etc. :p 17:27 pluskid, those should inherit from this general class labels 17:28 yes, did you see the figure and pseudo code? 17:28 oh, I think I missed that 17:28 yuml.me/f14284c4.jpg 17:29 and https://gist.github.com/2489859 17:29 very preliminary, though 17:29 and there's a problem of using template here: the superclass CLabels is no longer *one* class 17:29 but many many classes depending on the template parameter 17:30 so maybe we will not use this type of hierarchy, not sure currently 17:30 pluskid, mmm I don't understand why do you want to have the base class CLabels templated 17:33 n4nd0: otherwise, we do not know what to return in the function get_label(int32_t) 17:33 it might be int, float, or some tree or sequence 17:34 will be complex object in SO I guess 17:34 pluskid, well but one should know in the moment of using that function 17:34 should know it* 17:34 n4nd0: but how do you define the prototype of that function? 17:34 in a common superclass CLabels 17:35 pluskid, doing something similar to what we have in CFeatures 17:36 n4nd0: what's that? 17:36 you mean we do not define this function in the base class 17:36 but we dynamic_cast to subclasses when calling some specific functions? 17:37 pluskid, wait a second I am looking for an example within the code so you can see what I mean 17:37 ok 17:37 yes my idea would be to dynamic_cast to subclasses 17:42 we don't really need base CLabels 17:42 multiclass work with multiclass labels (only) 17:42 OK, I understand 17:42 binary with binary 17:42 however this can be a problem again 17:42 cause we have generic machine 17:43 blackburn, but there might some common functionality to all of them interesting to have in a base class 17:43 what is generic machine? 17:43 CMachine 17:43 it has labels.. 17:43 blackburn, and I think it offers more flexibility to have a base class 17:43 we can have a pure virtual CLabel super class 17:44 with almost nothing in it 17:44 yes 17:44 yes but we have to be careful 17:44 I imagine it like that 17:44 blackburn, in what sense? 17:44 with future casts 17:44 can you put an example where a problem may arise? 17:45 n4nd0: it would be nice to have set_labels(MulticlassLabels) for multiclass machines 17:45 but how to do that? 17:45 no way probably so it should be checked probably 17:46 yes dynamic_cast everywhere 17:47 pluskid: n4nd0: wiking: could you please elaborate tihs problem at the mailing list? 17:48 blackburn, sure, but I have to think more of it first 17:48 yes no hurry 17:48 hehehe ok 17:49 brb 17:49 sure 17:49 but we all can forget something so would be nice to have it on the mailing list 17:49 thanls 17:49 ks* 17:49 pluskid, blackburn, also, as far as I have understand from SO, SO generalizes multiclass 17:49 doesn't shogun has an issue list? 17:49 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 17:49 n4nd0: yes 17:49 pluskid: github issues work well too 17:49 feel free to use it 17:49 may be even better 17:49 yes 17:49 blackburn: no body looks at it I guess? :p 17:49 I think that should be reflected in the diagram somehow 17:49 pluskid: will if it is used 17:50 n4nd0: in my view, SO is multiclass, with structure in the class space 17:50 n4nd0: pluskid: please create new issue - we can even refer code modifications to it 17:50 multiclass is SO I'd say 17:50 ehmm 17:50 subset of SO 17:50 I agree with blackburn 17:50 simple finite 17:50 yes, SO without structure haha 17:51 no there are some structure 17:51 just like structure of N set 17:51 there is an order and distance 17:51 no, simple discrete set 17:51 I don't think there is order in generic multiclass 17:51 however d(y,y') = [y=y'] is strange distance 17:51 class 1 is not before class 2 17:51 heh right 17:51 no reason to keep it in mind 17:52 there is no really structure in the sense that is important for SO where there are important dependencies between the parts that form a label 17:52 I just thought of implementation 17:52 ok 17:52 actually it can be considered as consistent way 17:52 to define distance between numbers 17:52 like in multiclass 17:52 regression can also be thought to be multiclass: with infinite number of classes, and very nice structure (orders, distance, innerproduct, etc.) in the real line :p 17:53 brb 17:53 midnight here 17:54 good night all 17:54 wtf 17:54 pluskid, good night 17:54 you all are going berserk here? 17:54 :DD 17:54 I can hardly read the whole log 17:54 pluskid, indeed no to multiple inheritance 17:55 I am just preparing by databases exam for tomorrow :P 17:55 it will complicate things beyond repair 17:55 and I cannot even think of how one could possibly use swig with that 17:55 sonney2k: OK, I'll tell wiking that if he wants to use multiple inheritance, sonney2k will kill him :D 17:55 (think java, python that all have no MI) 17:55 I will kill everyone who wants MI 17:56 :D 17:56 good night 17:56 -!- pluskid [~chatzilla@111.120.57.192] has quit [Quit: ChatZilla 0.9.88.2 [Firefox 12.0/20120420145725]] 17:56 pluskid, n8 17:56 blackburn, about the label redesign - I am very much in favor of it - but I have no idea how we even can have an interface there 17:57 I mean basically multiclass, binary and real-valued labels share all functionality 17:57 only that they accept different values 17:57 sonney2k, I know that I am on risk of getting killed but ... I think python supports MI 17:57 like +1/-1 17:57 n4nd0, no 17:57 ok 17:58 http://docs.python.org/tutorial/classes.html 17:59 search for multiple inheritance 17:59 limited form of it they say 17:59 n4nd0, yeah but look at it - you will more likely shoot yourself in the foot then benefiting from this 18:00 but it exists :) 18:00 this is awesome 18:01 "all cases of multiple inheritance exhibit one or more diamond relationships" 18:01 :O 18:01 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun 18:05 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] 18:05 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 18:05 -!- n4nd0 [~n4nd0@wifi-cct-cisco-r252-4.cct.urjc.es] has quit [Read error: Connection reset by peer] 18:12 sonney2k: I told you it will be crazy 18:22 -!- vikram360 [~vikram360@117.192.189.186] has quit [Ping timeout: 240 seconds] 18:49 -!- blackburn [~qdrgsm@188.168.2.179] has quit [Quit: Leaving.] 18:53 -!- harshit_ [~harshit@182.68.43.52] has joined #shogun 19:13 -!- harshit_ [~harshit@182.68.43.52] has quit [Client Quit] 19:14 -!- vikram360 [~vikram360@117.192.173.180] has joined #shogun 19:19 -!- vikram360 [~vikram360@117.192.173.180] has quit [Ping timeout: 260 seconds] 19:29 -!- vikram360 [~vikram360@117.192.172.250] has joined #shogun 19:30 -!- Marty28 [~marty@cable-158-181-78-199.cust.telecolumbus.net] has joined #shogun 20:00 -!- Marty28 [~marty@cable-158-181-78-199.cust.telecolumbus.net] has quit [Quit: Colloquy for iPad - http://colloquy.mobi] 20:11 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun 20:47 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] 20:56 blackburn - all your fault ;-) 21:31 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun 21:45 sonney2k: in which means? 21:46 you seem to want to do 1000 things at the same time ;) 21:58 -!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has quit [Remote host closed the connection] 21:59 blackburn, btw I've seen you label class hierarchy 22:00 which functions should go into the new CLabels? 22:00 sonney2k: mine? it was pluskid's I think 22:05 sonney2k: this should be discussed actually 22:06 about 1000 things - yes but we have enough horsepower^W human resources 22:07 so it is a question of management :) 22:09 sonney2k: one guy from my school is this year's gsoc student - so a chance I'll get some opencv expert in touch 22:11 blackburn, because I think what we currently have in CLabels would work for binary/multiclass/reallabels 22:14 only that we should have some check_consistency / validity of the labels function 22:15 sonney2k: I agree but what about structured labels.. 22:15 they are totally different 22:15 there is nothing except checking 'is_valid()' that they have in common 22:16 I mean even get_label() cannot be shared among label classes 22:16 so it cannot be in the new CLabels() 22:16 so I am actually not so sure if we really need a CLabels base class 22:17 agree 22:17 sonney2k: actually it could be shared using templates 22:18 but some non-trivial thing 22:18 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] 22:33 blackburn - with templates overloading sucks 22:35 -!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] 22:39 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun 22:46 -!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] 22:46 -!- wiking [~wiking@huwico/staff/wiking] has joined #shogun 22:46 dooh 22:51 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun 22:51 https://github.com/shogun-toolbox/shogun/graphs/contributors 22:51 wonder where am i :) 22:51 github aaate meee! 22:53 that sucks 22:55 ;) 22:55 interesting stats 22:55 sonney2k: yeah the punchcard is funny as well :) 22:55 sonney2k: have you seen my mail to slep? do you think I ask for reasonable things? 22:56 sonney2k: around? 23:01 blackburn, yeah but free for academic use means - not open source 23:01 sonney2k: gsomix needs some guidance 23:01 so you need some statement from them that you can integrate it under the terms of GPL 23:02 sonney2k: do you think they would deny it? 23:02 ah 23:02 currently it looks like this 23:02 yes so I asked they 23:02 gsomix, wassup? 23:02 -!- harshit_ [~harshit@182.68.43.52] has joined #shogun 23:02 sonney2k: that would be pretty bad 23:02 sonney2k, what we plan to store in objects at end? references to SGVectors or pointers, or copies? 23:03 -!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun 23:03 hmm 23:03 browser has died 23:03 gsomix, good question 23:06 sonney2k: why would they want to keep code available but not real open source? 23:06 blackburn, why is svmlight/struct not opensource? 23:06 -!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] 23:07 sonney2k: isn't it? 23:07 of course not 23:07 that's crazy 23:07 blackburn_, that is why svmlight is not part of official shogun 23:08 sonney2k: yes I know it has some special status 23:08 can SLEP has similar one then? 23:08 and I was begging TJ for way too many times - he simply doesn't like OSS 23:08 blackburn_, ask them: tell them that shogun is under GPLv3 and that you need a statement saying SLEP is released under a GPL3 compatible license or under GPL3 directly 23:09 sonney2k: can reimplementation of anything has different license? 23:10 gsomix, let me think loudly: If we were using pointers we would keep ptrs everywhere. 23:10 sonney2k: yeah I asked already but not about statement - lets wait for answer first 23:10 blackburn_, re-implementing without using / looking at their code - of course 23:10 sonney2k: that license stuff is pretty vomitting isn't it? 23:11 blackburn_, sure it is - in particular because even OSS licenses might conflict with each other - so we cannot use MPL / CPL licensed code in shogun 23:12 gsomix, so in principle we want exactly that - pass around object ptrs only 23:12 lets hope they will license it using gpl 23:13 I think to allow to be integrated to shogun is the only way to survive :D 23:17 gsomix, now when we have a function SGVector return_new_vector() we have to return some copy 23:17 gsomix, but otherwise we return the reference SGVector& return_vector_from_class() 23:18 -!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] 23:21 sonney2k, aha. 23:24 sonney2k, sonney2k, btw, what about const SGVector& return_vector_from_class() const;'? Should we have two version (one for const objects)? 23:24 sonney2k, ok, I'll think about it yet. I will end review of code for SGVector to refs' tomorrow. 23:37 I'm going to bed now =___= 23:38 nite 23:38 -!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun 23:49 errr connection sucks 23:49 when I see mails like that last at the gsoc students mailing list 23:50 I ask myself how can people be that stupid 23:50 blackburn_, I start to wonder if we shouldn't use SGVector* ... 23:54 I feel uneasy by returning copies of SGVector and not being able to store references... 23:55 sonney2k: I am afraid we can stuck there 23:55 what do you mean? 23:55 sonney2k: I mean we can lose much time doing this (being unsure it is useful) 23:56 I am talking about any transitions 23:56 blackburn_, sure 23:56 that is why I am asking 23:57 blackburn_, btw why do you use the webchat? 23:57 sonney2k: using public wifi - irc is blocked 23:57 sonney2k: we would need to really get sure we need it 23:58 do we have any problem with it so far? 23:58 yes 23:58 which one? 23:58 everything is copied and we have no proper way of cleaning up memory 23:58 --- Log closed Thu Apr 26 00:00:01 2012