--- Log opened Mon Jun 11 00:00:41 2012 | ||
n4nd0 | yay!! multiclass example with SO is working fine :D | 01:37 |
---|---|---|
n4nd0 | someone out there to share my happiness? | 01:38 |
-!- n4nd0_ [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 01:42 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 245 seconds] | 01:42 | |
-!- n4nd0_ is now known as n4nd0 | 01:44 | |
emrecelikten | n4nd0: Great | 02:18 |
emrecelikten | :P | 02:18 |
n4nd0 | :) | 02:23 |
n4nd0 | good night! | 02:30 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 02:30 | |
-!- ckwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has quit [Remote host closed the connection] | 03:26 | |
-!- emrecelikten [~Anubis@176.40.238.218] has quit [Read error: Connection reset by peer] | 05:47 | |
-!- puffin444 [472e31fb@gateway/web/freenode/ip.71.46.49.251] has quit [Quit: Page closed] | 07:13 | |
-!- CIA-9 [~CIA@cia.atheme.org] has quit [] | 07:50 | |
-!- CIA-18 [~CIA@cia.atheme.org] has joined #shogun | 07:54 | |
-!- CIA-18 [~CIA@cia.atheme.org] has quit [Excess Flood] | 07:59 | |
-!- CIA-18 [~CIA@cia.atheme.org] has joined #shogun | 08:01 | |
-!- CIA-18 [~CIA@cia.atheme.org] has quit [Excess Flood] | 08:02 | |
-!- CIA-18 [~CIA@cia.atheme.org] has joined #shogun | 08:07 | |
-!- gsomix [~gsomix@80.234.31.247] has joined #shogun | 08:12 | |
-!- uricamic [~uricamic@2001:718:2:1634:7816:2ae9:adc4:a6da] has joined #shogun | 08:56 | |
-!- gsomix_ [~gsomix@95.67.167.251] has joined #shogun | 08:57 | |
-!- gsomix [~gsomix@80.234.31.247] has quit [Ping timeout: 256 seconds] | 09:01 | |
-!- heiko [~heiko@host86-177-178-116.range86-177.btcentralplus.com] has joined #shogun | 10:11 | |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 10:35 | |
eric_ | hi all | 10:35 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 10:38 | |
n4nd0 | sonney2k: around? | 10:40 |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds] | 11:28 | |
-!- heiko [~heiko@host86-177-178-116.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 11:41 | |
-!- heiko [~heiko@host86-176-2-162.range86-176.btcentralplus.com] has joined #shogun | 11:42 | |
-!- jps [5536538f@gateway/web/freenode/ip.85.54.83.143] has joined #shogun | 11:47 | |
-!- romi_ [~mizobe@187.57.4.69] has joined #shogun | 12:01 | |
wiking | ah report day! | 12:15 |
n4nd0 | yeah \o/ | 12:17 |
wiking | btw: currently shogun will not compile at all w/o lapack support | 12:43 |
wiking | i've just tried it on a simple machine to clone my repo and test some stuff there and i've realized this.. | 12:43 |
-!- jps [5536538f@gateway/web/freenode/ip.85.54.83.143] has quit [Quit: Page closed] | 12:47 | |
-!- gsomix_ [~gsomix@95.67.167.251] has quit [Ping timeout: 265 seconds] | 13:07 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 13:12 | |
-!- heiko [~heiko@host86-176-2-162.range86-176.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 13:43 | |
-!- gsomix [~gsomix@178.45.79.164] has joined #shogun | 13:49 | |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 14:13 | |
eric____ | hi again | 14:14 |
eric____ | I would like to test smthg: compute the spectrum kernel from strings (i have CStrginFeatures<uint16_t>) with CommWordString kernel | 14:23 |
eric____ | whats the function to use (preproc?) which compute the k-mer frequ feature vector ? | 14:24 |
eric____ | following the doc, CommWordString use SortWordString preproc | 14:25 |
eric____ | and it says : "For this to work the strings have to be mapped into * a binary higher order representation first (cf. obtain_from_*() functions in * CStringFeatures)" | 14:25 |
eric____ | what exaclty does obtain_from_features ? It send me an error about num_symbols .. | 14:26 |
-!- heiko [~heiko@host86-177-4-176.range86-177.btcentralplus.com] has joined #shogun | 14:35 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 248 seconds] | 14:41 | |
-!- wiking [~wiking@208.76.55.196] has joined #shogun | 14:42 | |
-!- wiking [~wiking@208.76.55.196] has quit [Changing host] | 14:42 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 14:42 | |
eric____ | someone here ? | 14:50 |
heiko | me :) | 14:50 |
-!- emrecelikten [~Anubis@176.40.238.218] has joined #shogun | 14:55 | |
eric____ | heiko: :) | 14:57 |
eric____ | heiko: could you help me to run CommWordString kernel ? | 14:58 |
heiko | eric, maybe whats the problem with it? | 14:58 |
eric____ | I would like to know how to compute the ?-mer spectrum kernel fro example | 14:58 |
eric____ | 2-mer | 14:59 |
eric____ | (i have CStrginFeatures<uint16_t> | 14:59 |
eric____ | docs advise to run preprocess SortWordString which just sort the string | 14:59 |
eric____ | I am wondering where is the function which compute the 2-mer frequency feature vector ?? | 15:00 |
heiko | I am not really into that stuff, however, lets see | 15:01 |
heiko | so you just want to use this common word kernel? | 15:01 |
heiko | have you tried the example for it? | 15:01 |
heiko | why do you want to compute a feature vector? | 15:02 |
eric____ | lets explain what I have understand | 15:03 |
eric____ | for now my samples are stringfeatures: string list of uint16_t | 15:03 |
eric____ | I would like to compute a spectrum kernel of 2-mer | 15:04 |
heiko | ok | 15:04 |
eric____ | and as the docs says : in this feature vector each entry denotes how often the k-mer appears in that string | 15:05 |
eric____ | so there is some preproc to do I think | 15:05 |
heiko | where is that written in the doc? | 15:05 |
heiko | could you send a link so that I can see the context? | 15:05 |
eric____ | I have seen obtain_from_features method in CStringFeatures but I dont know how to use it . | 15:05 |
eric____ | http://www.shogun-toolbox.org/doc/en/current/classshogun_1_1CWeightedCommWordStringKernel.html | 15:06 |
eric____ | or http://www.shogun-toolbox.org/doc/en/current/classshogun_1_1CCommWordStringKernel.html for simple k fixed version | 15:06 |
heiko | ok | 15:07 |
heiko | first: you usually dont compute fetaure vectors directly, this is just implicitly. If you *would* compute feature vectors, they had this described structure | 15:09 |
heiko | imagine you use 10-mers over an alphabet with 10 letters | 15:09 |
heiko | thats 10^10 dimensions | 15:09 |
heiko | this sorting just means, that in order to use these weights, you would have to sort the data that you attach to the kernel | 15:10 |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds] | 15:10 | |
-!- K0stIa [~kostia@2001:718:2:1634:76f0:6dff:fe92:4013] has joined #shogun | 15:11 | |
heiko | the obtain_from_* methods are just to convert string features/char features to numerical representations | 15:13 |
heiko | I would suggest to make the example run and then experiment a bit with it | 15:13 |
K0stIa | hi all, I have problem with installing shogun from github code. I'm doing ./configure and then make. And my make fails, it says | 15:14 |
K0stIa | statistics/QuadraticTimeMMD.cpp:164:34: error: 'compute_eigenvectors' is not a member of 'shogun::SGMatrix<double>' | 15:14 |
K0stIa | can anyone say smth about it ? | 15:14 |
heiko | Hi K0stIa | 15:14 |
heiko | yes, I will have a look, just a second | 15:14 |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun | 15:15 | |
heiko | K0stIa, are you using the current git ? | 15:15 |
heiko | when did you clone/update last? | 15:16 |
heiko | ah I think I know whats wrong ... | 15:16 |
heiko | you can fix that by installing lapack libraries but I will also add a fix | 15:17 |
CIA-18 | shogun: Heiko Strathmann master * r27f5fd1 / (2 files): added lapack check for eigenspectrum based test to solve compile error - http://git.io/C5xsmQ | 15:24 |
CIA-18 | shogun: Heiko Strathmann master * r1532e15 / (2 files): Merge pull request #579 from karlnapf/master - http://git.io/dli_Ng | 15:24 |
CIA-18 | shogun: Heiko Strathmann master * r0e9be8c / src/shogun/statistics/QuadraticTimeMMD.cpp : another check for LAPACK - http://git.io/ziupmw | 15:24 |
heiko | K0stIa, try with current git code, I added a fix if LAPACK is not installed | 15:24 |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 15:59 | |
eric____ | heiko: re | 15:59 |
-!- alexlovesdata [c25faea9@gateway/web/freenode/ip.194.95.174.169] has joined #shogun | 16:00 | |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds] | 16:03 | |
-!- K0stIa [~kostia@2001:718:2:1634:76f0:6dff:fe92:4013] has left #shogun [] | 16:32 | |
-!- K0stIa [~kostia@2001:718:2:1634:76f0:6dff:fe92:4013] has joined #shogun | 16:53 | |
K0stIa | @heiko, thnx, now it works :) | 16:53 |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Quit: Page closed] | 17:01 | |
heiko | K0stIa, nice :) | 17:18 |
heiko | eric, your connection is even worse than mine :) | 17:18 |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has joined #shogun | 17:22 | |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] | 17:22 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 17:22 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 256 seconds] | 17:25 | |
-!- wiking_ is now known as wiking | 17:25 | |
-!- emrecelikten [~Anubis@176.40.238.218] has quit [Quit: Leaving.] | 17:27 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 17:30 | |
-!- blackburn [d55781be@gateway/web/freenode/ip.213.87.129.190] has joined #shogun | 17:30 | |
-!- K0stIa [~kostia@2001:718:2:1634:76f0:6dff:fe92:4013] has left #shogun [] | 17:40 | |
-!- blackburn [d55781be@gateway/web/freenode/ip.213.87.129.190] has quit [Quit: Page closed] | 17:40 | |
-!- eric [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 17:47 | |
-!- eric is now known as Guest57868 | 17:47 | |
-!- blackburn [~blackburn@188.168.3.9] has joined #shogun | 17:50 | |
-!- uricamic [~uricamic@2001:718:2:1634:7816:2ae9:adc4:a6da] has quit [Quit: Leaving.] | 17:52 | |
-!- romi_ [~mizobe@187.57.4.69] has quit [Ping timeout: 260 seconds] | 18:03 | |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 18:08 | |
-!- Guest57868 [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds] | 18:09 | |
-!- emrecelikten [~emrecelik@176.40.238.218] has joined #shogun | 18:09 | |
-!- eric____ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds] | 18:15 | |
blackburn | heiko: one trick - you may pull from fork w/o pull request | 18:22 |
heiko | blackburn, ok, how? :) | 18:23 |
blackburn | ah you have no clone of shogun | 18:23 |
blackburn | then no way :) | 18:24 |
heiko | no :) | 18:25 |
heiko | If it annoys you I can do that | 18:25 |
blackburn | noo | 18:25 |
blackburn | I just mean would be faster probably | 18:25 |
blackburn | for you | 18:25 |
blackburn | 3 more reports to go | 18:26 |
heiko | reports? | 18:27 |
heiko | the way I am doing it is also fast :) I always get confused if there is too much stuff | 18:28 |
heiko | ah weekly | 18:29 |
blackburn | yeah | 18:31 |
n4nd0 | blackburn: I am doing my soon, I promise :) | 18:33 |
n4nd0 | blackburn: I am leaving my repo clean of new code first | 18:33 |
blackburn | heh | 18:34 |
blackburn | I have seen you get multiclass example to work alreadY? | 18:34 |
n4nd0 | blackburn: yeah! | 18:36 |
n4nd0 | blackburn: I have to use bigger datasets though | 18:38 |
n4nd0 | Nico told me that the evaluation license of MOSEK has an expiration date | 18:38 |
n4nd0 | so I could just solve very small problems | 18:38 |
blackburn | baad | 18:39 |
n4nd0 | but if I get a new one I will be able to solve big ones | 18:39 |
blackburn | is there any chance to use that bmrm stuff? | 18:39 |
blackburn | I am not into SO actually you know :) | 18:39 |
n4nd0 | blackburn: to use it for what do you mean? | 18:40 |
blackburn | instead of mosek | 18:40 |
n4nd0 | blackburn: ah ok | 18:41 |
n4nd0 | they are different things | 18:41 |
n4nd0 | so we use mosek for the optimization algorithm | 18:41 |
n4nd0 | bmrm is the training algorithm | 18:41 |
n4nd0 | it relies in a optimizer as well | 18:41 |
n4nd0 | but it is a different one and for bmrm libqp is enough | 18:42 |
n4nd0 | at least I think it is like this, I'm not 100% sure | 18:42 |
blackburn | aham | 18:42 |
-!- emrecelikten [~emrecelik@176.40.238.218] has left #shogun [] | 18:58 | |
@sonney2k | n4nd0, so what was the issue? what did you have to change to get it to work? | 19:08 |
@sonney2k | heiko, thanks for the fix | 19:08 |
@sonney2k | heiko, I guess you have seen that I reworked CStatistics/CMath/SGVector/SGMatrix... | 19:08 |
@sonney2k | I think it is much better now | 19:09 |
heiko | yes, my examples broke :) | 19:09 |
heiko | I agree | 19:09 |
n4nd0 | sonney2k: the issue I wanted to tell you about is related to modular interface, I cannot make work my SO there | 19:09 |
heiko | much nicer | 19:09 |
n4nd0 | sonney2k: but I have not fixed it yet | 19:09 |
@sonney2k | heiko, ahh and please keep the scheme w/ doing pull requests and then merges | 19:09 |
@sonney2k | I think I should do the same | 19:09 |
n4nd0 | sonney2k: I did the example in libshogun interface | 19:09 |
heiko | sonney2k, well, ok :) it feels more comfortable for me anyway | 19:09 |
heiko | and this way I dont break the main repo | 19:09 |
heiko | (which I sometimes manage to do with my fork :) | 19:10 |
@sonney2k | n4nd0, so waht is the issue w/ modulear here? | 19:10 |
blackburn | is BFGS any cool for svms? | 19:10 |
blackburn | any evidences? | 19:10 |
@sonney2k | the buildbot is back to live | 19:10 |
@sonney2k | shogun-buildbot, you rock! | 19:10 |
shogun-buildbot | What you say! | 19:10 |
@sonney2k | n4nd0, I got positive feedback wrt mosek btw | 19:11 |
n4nd0 | sonney2k: what feedback? | 19:12 |
n4nd0 | sonney2k: about modular | 19:14 |
n4nd0 | sonney2k: TypeError: in method 'StructuredOutputMachine_set_model', argument 2 of type 'CStructuredModel *' | 19:14 |
@sonney2k | n4nd0, that we will get licenses | 19:14 |
n4nd0 | sonney2k: I get errors like this | 19:14 |
blackburn | n4nd0: didn't you fix that? | 19:14 |
n4nd0 | sonney2k: it seems that it doesn't recognize the inheritance properly | 19:14 |
n4nd0 | blackburn: no :O | 19:14 |
blackburn | okay lets try to get rid of it | 19:15 |
@sonney2k | n4nd0, that is easily fixed - just put a class CStructuredModel; in the header | 19:15 |
n4nd0 | blackburn: the last we said about it was ... we'll ask sonney2k | 19:15 |
blackburn | yes I recall | 19:15 |
blackburn | iirc it ^ didn't helped? | 19:15 |
n4nd0 | sonney2k: I tried with that, blackburn told me about it | 19:15 |
@sonney2k | n4nd0, that will help | 19:15 |
n4nd0 | but it didn't make the trick | 19:15 |
n4nd0 | I'll try again then | 19:15 |
@sonney2k | it will | 19:15 |
@sonney2k | if not show me the changed header | 19:16 |
n4nd0 | sonney2k: ok, give me some minutes, I have just rebased and I have to adapt the changes done in CMath | 19:16 |
@sonney2k | for some reason swig forgets that we are in shogun:: namespace if we don't add that everywhere :( | 19:16 |
blackburn | inheritance problems emerge essentially when you have wrong order of includes | 19:16 |
@sonney2k | n4nd0, k | 19:16 |
@sonney2k | blackburn, it is not an inheritance problem here | 19:17 |
blackburn | some kind of | 19:17 |
blackburn | however yes rather polymorphism :D | 19:17 |
@sonney2k | blackburn, today I got the galaxy s3 | 19:23 |
@sonney2k | only to realize that it needs a micro sim card | 19:24 |
blackburn | cool | 19:24 |
@sonney2k | so I had to cut my sim card with a scissors | 19:24 |
blackburn | however I still prefer my robust phone | 19:25 |
n4nd0 | seriously? | 19:25 |
n4nd0 | the scissors thing | 19:25 |
@sonney2k | yes | 19:25 |
@sonney2k | it works | 19:25 |
n4nd0 | hahaha | 19:25 |
blackburn | why not | 19:25 |
n4nd0 | c'mon, it sounds funny | 19:26 |
@sonney2k | n4nd0, easier than transitioning CLabels :D | 19:26 |
n4nd0 | sonney2k: :D so the CStructuredModel I should include it in the child class right? | 19:26 |
@sonney2k | ? | 19:27 |
@sonney2k | n4nd0, in the header of StructuredOutputMachine | 19:27 |
n4nd0 | sonney2k: ah ok | 19:27 |
@sonney2k | you put class CStructuredModel; directly after the namespace shogun { | 19:27 |
n4nd0 | sonney2k: yeah, I just thought that I had to do that in the classes that extend CStructuredModel | 19:28 |
@sonney2k | n4nd0, in every class that needs a CStructuredModel as input | 19:28 |
n4nd0 | sonney2k: ok got it, thanks! I am going to try it | 19:28 |
n4nd0 | sonney2k: about the mosek licenses | 19:29 |
n4nd0 | sonney2k: nico told me that the evaluation ones expire after three months | 19:29 |
n4nd0 | sonney2k: but that I should be able to get a new evaluation one | 19:30 |
n4nd0 | sonney2k: that has all the capabilities for yet three more months | 19:30 |
wiking | n4nd0 yes it does | 19:42 |
wiking | n4nd0: if u want i can send u a new license | 19:42 |
wiking | i needed a new one as well | 19:42 |
n4nd0 | sonney2k: all right, you fixed my problem with modular interface :) | 19:42 |
CIA-18 | shogun: Sergey Lisitsyn master * rf5163e8 / (7 files in 2 dirs): Merge branch 'multiclass' of git://github.com/pluskid/shogun (+5 more commits...) - http://git.io/hn3mxQ | 19:45 |
@sonney2k | n4nd0, dont' ask how long it took me to find the workaround/solution | 19:46 |
n4nd0 | :O | 19:47 |
CIA-18 | shogun: Sergey Lisitsyn master * r90641c6 / src/shogun/lib/slep/SpInvCoVa/invCov.cpp : Made sparse inverse covariance not that verbose - http://git.io/5RFMOw | 19:48 |
CIA-18 | shogun: Sergey Lisitsyn master * r7cd8964 / examples/undocumented/python_modular/mathematics_sparseinversecovariance_modular.py : Added Sparse Inverse Covariance example - http://git.io/amEXRg | 20:06 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 248 seconds] | 20:08 | |
CIA-18 | shogun: Sergey Lisitsyn master * rdf58c63 / examples/undocumented/python_modular/mkl_linadd_regression.py : Dropped linadd mkl test example - http://git.io/6NxhqQ | 20:09 |
-!- vikram360 [~vikram360@117.192.187.238] has joined #shogun | 20:55 | |
-!- vikram360 [~vikram360@117.192.187.238] has quit [Ping timeout: 245 seconds] | 21:00 | |
-!- alexlovesdata [c25faea9@gateway/web/freenode/ip.194.95.174.169] has quit [Quit: Page closed] | 21:02 | |
-!- vikram360 [~vikram360@117.192.170.22] has joined #shogun | 21:04 | |
-!- vikram360 [~vikram360@117.192.170.22] has quit [Ping timeout: 245 seconds] | 21:11 | |
shogun-buildbot | build #605 of csharp_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/605 | 21:15 |
-!- gsomix [~gsomix@178.45.79.164] has quit [Quit: Ex-Chat] | 21:52 | |
@sonney2k | ok sth killed the buildbot again hmmhh | 22:01 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 22:02 | |
n4nd0 | wiking: hey! do you have a moment now? | 22:03 |
wiking | yeah sure | 22:03 |
wiking | shoot | 22:03 |
wiking | u need a new license? | 22:03 |
n4nd0 | wiking: it's about SO framework | 22:03 |
wiking | ok | 22:03 |
n4nd0 | wiking: I think I can manage to download a license | 22:03 |
wiking | well i can just fwd u again as last time | 22:04 |
wiking | so that you don't have to ask for a new one | 22:04 |
n4nd0 | wiking: ok then | 22:04 |
n4nd0 | wiking: did you get a new one? | 22:04 |
wiking | yeps | 22:04 |
wiking | what was the email again | 22:04 |
wiking | ? | 22:04 |
wiking | done | 22:05 |
wiking | should be good for 2 more months | 22:05 |
n4nd0 | wiking: so Nico found some flaws about having the StructuredModel separated into different classes | 22:06 |
wiking | aaah | 22:06 |
wiking | hit me | 22:06 |
n4nd0 | one for the argmax, another for the psi, etc | 22:06 |
wiking | yep the one we've talked about last week | 22:06 |
n4nd0 | the problem is that they normally rely on each other and it's difficult to make them fully separated | 22:06 |
blackburn | oh 8/8, thanks guys! | 22:07 |
wiking | hehehe blackburn :) | 22:07 |
wiking | n4nd0: ok i get it that they are relying on each other | 22:07 |
wiking | but still why would it be difficult to separate them? | 22:07 |
n4nd0 | wiking: I am not sure exactly about the reasons | 22:08 |
n4nd0 | wiking: but what I found is that for example | 22:08 |
n4nd0 | if the argmax needs the loss and joint feature function | 22:08 |
n4nd0 | then you have with the argmax a class that contains the same logic as the whole model | 22:08 |
n4nd0 | and it loses a bit of sense | 22:09 |
wiking | oh | 22:09 |
wiking | i see what you mean | 22:09 |
n4nd0 | :) | 22:09 |
wiking | that we might end up | 22:09 |
wiking | having a class that contains a class that contains the the parent class | 22:09 |
wiking | right? | 22:09 |
wiking | A has B has A | 22:09 |
wiking | or? | 22:10 |
n4nd0 | mmm I don't think that problem really arises | 22:10 |
n4nd0 | but it would be more like | 22:10 |
n4nd0 | you have the Model that contains the argmax, the loss and psi | 22:11 |
n4nd0 | but argmax needs loss and psi too | 22:11 |
wiking | blackburn: btw: russia will win this eurocup | 22:11 |
wiking | n4nd0: yep that's more or less what i'm saying | 22:11 |
n4nd0 | wiking: ok then | 22:11 |
wiking | that you have class A that has B, C,D | 22:11 |
wiking | and B will have C, D as well | 22:11 |
n4nd0 | exactly | 22:11 |
wiking | c = loss, d = psi | 22:11 |
n4nd0 | that was the problem I found, but I think that Nico had other reasons too | 22:12 |
wiking | well just pass the reference to the parent? | 22:12 |
wiking | n4nd0: was it discussed here? | 22:12 |
wiking | i.e. can i find it in irc logs? | 22:12 |
n4nd0 | wiking: yes, I'll send you the link now | 22:12 |
wiking | ok cool | 22:13 |
n4nd0 | wiking: but even if you pass the reference to the parent | 22:13 |
n4nd0 | wiking: the thing would be that the model or the argmax loses its meaning | 22:13 |
wiking | well not necessarily | 22:13 |
wiking | i mean if u need psi and loss for argmax | 22:13 |
wiking | then u need it | 22:14 |
wiking | thus it's still argmax | 22:14 |
wiking | the only question is how do u have the reference on these | 22:14 |
n4nd0 | wiking: http://www.shogun-toolbox.org/irclogs/%23shogun.2012-06-07.log.html the converation about it starts around 15:45 | 22:15 |
wiking | ok | 22:15 |
wiking | reading/parsing | 22:15 |
wiking | "latent guys" | 22:17 |
wiking | :D | 22:17 |
n4nd0 | :) | 22:17 |
n4nd0 | people cannot see you | 22:18 |
wiking | "nicococoexample: for latent svms i think we also need a delta(y_i,y_bar,h_latent)" | 22:18 |
wiking | that's not true | 22:18 |
n4nd0 | it's just the joint feature right? | 22:18 |
wiking | no i mean | 22:18 |
wiking | i don't need that delta | 22:18 |
wiking | only delta(y, y_bar) | 22:18 |
n4nd0 | yeah ok, I meant like it's the joint feature the only one that accounts for h | 22:19 |
wiking | mmm | 22:21 |
wiking | i've read it | 22:21 |
wiking | tilld 16:57 | 22:21 |
wiking | *till | 22:21 |
n4nd0 | so? would you still like more to have the division into classes? | 22:21 |
n4nd0 | CArgMax, etc | 22:21 |
wiking | well | 22:21 |
wiking | for me that makes my life easier | 22:21 |
wiking | but then again | 22:21 |
wiking | if u are strongly against it | 22:22 |
n4nd0 | I think that to have just a CStructuredModel and extend that one is good | 22:22 |
wiking | then i guess we'll just have to live with that | 22:22 |
wiking | and then yeah | 22:22 |
wiking | just derive my own stuff from CStructuredModel | 22:22 |
n4nd0 | why do you think it makes your life easier for you? | 22:22 |
wiking | because i don't have to care about anything else in CStructuredModel | 22:23 |
wiking | just about psi and argmax ;) | 22:23 |
n4nd0 | wiking: https://github.com/iglesias/shogun/blob/so/src/shogun/so/MulticlassModel.h take a look here | 22:23 |
wiking | exactly | 22:23 |
n4nd0 | so the plan would be to add an additional get_joint_feature_vector | 22:23 |
n4nd0 | for latent | 22:23 |
wiking | this is complicated ;) | 22:23 |
n4nd0 | that gets int32_t, StructuredData and another int32_t | 22:23 |
wiking | i mean if u think about that i only want to define a psi and an argmax function | 22:23 |
wiking | this is way too complicated | 22:23 |
n4nd0 | but you will need to do just that | 22:24 |
wiking | and then each time | 22:24 |
wiking | actually if u wanna be fair | 22:24 |
wiking | you'll need to override const char* get_name() const function as well | 22:24 |
wiking | since the given class is a new one... ;) | 22:24 |
n4nd0 | aned get_dim | 22:24 |
wiking | so it's just weird in my head | 22:25 |
wiking | that for a simple 2 things | 22:25 |
wiking | actually 2 functions | 22:25 |
wiking | i have to override a whole class | 22:25 |
wiking | and ok i get why shouldn't we use function pointers... i.e. swig | 22:25 |
n4nd0 | yes, I mean the reasons you are telling me are just saved using function pointers | 22:26 |
n4nd0 | and we cannot do that | 22:26 |
wiking | hehe yeah | 22:26 |
wiking | but we could do still structs or classes ;) | 22:26 |
wiking | but then again | 22:26 |
n4nd0 | so we can either use a class for each thing (for argmax, for delta, for psi) | 22:27 |
n4nd0 | or a class where we can put everything together | 22:27 |
wiking | if u strongly feel about this and as far as i can see nico is quite assertive about this | 22:27 |
wiking | then go with a one class option | 22:27 |
n4nd0 | exactly | 22:27 |
n4nd0 | this is what we are going for | 22:27 |
wiking | but i do not agree ;) | 22:27 |
n4nd0 | haha | 22:27 |
wiking | but it's your code and your stuff | 22:27 |
wiking | so i do not want to interfere too much :) | 22:28 |
n4nd0 | I do not agree there :) | 22:28 |
n4nd0 | I mean, it is my code ok | 22:28 |
wiking | but then again i'm gonna use it | 22:28 |
wiking | and i want to make my life easier :P | 22:28 |
n4nd0 | but it is the "only" option we have | 22:28 |
n4nd0 | because we cannot use function pointers | 22:28 |
wiking | friend classes :) | 22:28 |
n4nd0 | and discarding them, the option I choose is the one class option | 22:28 |
n4nd0 | I suggested that quite at the beginning, but blackburn and sonney2k disagreed about them :S | 22:29 |
wiking | hahahhaha | 22:29 |
wiking | u see | 22:29 |
blackburn | what did I diagree? | 22:29 |
wiking | blackburn: baaastard! | 22:29 |
n4nd0 | using friend classes | 22:29 |
blackburn | :D | 22:29 |
blackburn | hmm sorry I am lazy to read your discussion - is there any problem related with not having friends? | 22:30 |
blackburn | not for real but in your classes I mean | 22:30 |
blackburn | :D | 22:30 |
wiking | blackburn: :> | 22:30 |
wiking | hahahahah https://fbcdn-sphotos-a.akamaihd.net/hphotos-ak-ash3/547749_472264696132937_557005192_n.jpg | 22:30 |
n4nd0 | we don't even get to install the 10% in Sweden ... | 22:31 |
wiking | blackburn: well using friend for classes would make our lives easier :) | 22:31 |
blackburn | haha | 22:31 |
wiking | n4nd0: :> | 22:31 |
blackburn | I think being flexible is cool so if you really think friends would make it better do it | 22:32 |
n4nd0 | wiking: how would you like to use them in any case? | 22:32 |
wiking | n4nd0: well have argmax as a friend class in structuredmodel | 22:34 |
wiking | so anything u have in structuredmodel you'll be able to call from argmax | 22:34 |
n4nd0 | wiking: but still, you have to make a class for argmax | 22:34 |
wiking | and why's that a problem in that case? | 22:34 |
n4nd0 | and you would do this in order to avoid doing a class child of structuredmodel | 22:34 |
wiking | yes because argmax class would just do the calculation of argmax | 22:35 |
wiking | thus if you inherit from it | 22:35 |
wiking | u want to change the argmax calculation | 22:35 |
wiking | and nothing else | 22:35 |
wiking | not like in case of structuredmodel | 22:35 |
wiking | if u inherit from that | 22:35 |
n4nd0 | you can inherit from structuredmodel and just do that too | 22:35 |
wiking | u either change argmax or psi | 22:35 |
wiking | and here or is not exclusive | 22:35 |
wiking | and u might not want to change anything | 22:36 |
wiking | just argmax | 22:36 |
n4nd0 | you can do that as it is right now | 22:36 |
wiking | yes | 22:36 |
n4nd0 | then? | 22:36 |
wiking | by inheriting from structuredmodel | 22:36 |
wiking | right? | 22:36 |
n4nd0 | yes | 22:36 |
blackburn | n4nd0: bahh 2-1 | 22:36 |
n4nd0 | blackburn: :( | 22:36 |
n4nd0 | blackburn: I am suffering it | 22:36 |
wiking | blackburn: sweden loose? | 22:36 |
n4nd0 | blackburn: but to tell the truth ... they really suck | 22:37 |
blackburn | yeah | 22:37 |
wiking | yeah cool | 22:37 |
wiking | finally some fun :) | 22:37 |
n4nd0 | yeah, against Ukraine | 22:37 |
blackburn | I am not watching it just read in twitter :) | 22:37 |
n4nd0 | Ukraine seems a good team so imagine .... | 22:37 |
n4nd0 | :D | 22:37 |
wiking | n4nd0: so the thing is that i'm lazy.. so i don't want to reimplement a lot of things, just because of one function change :P | 22:37 |
n4nd0 | wiking: yeah, I understand that | 22:38 |
wiking | it's going to be russia i have 10 euros on that one :) | 22:38 |
n4nd0 | wiking: but my point is that you don't need to do it with CStructuredModel either | 22:38 |
blackburn | n4nd0: well they have pretty strong teams in national league | 22:38 |
wiking | n4nd0: currently i can only change that way | 22:38 |
blackburn | and well, they have shevchenko | 22:38 |
wiking | or? | 22:38 |
n4nd0 | wiking: yeah | 22:39 |
n4nd0 | wiking: you can extend the model and just implement the argmax if you want | 22:39 |
wiking | extend u mean class Whatever : public CStructuredModel | 22:39 |
wiking | right? | 22:39 |
n4nd0 | yes | 22:39 |
n4nd0 | sorry ... java naming :P | 22:39 |
wiking | ok that is a huge overhead | 22:40 |
wiking | for a simple functional change :P | 22:40 |
wiking | imho | 22:40 |
wiking | :) | 22:40 |
wiking | because this will require me | 22:40 |
wiking | defining ctors and dtors | 22:40 |
wiking | calling the parent ctors | 22:40 |
wiking | but then again | 22:41 |
wiking | yeah it'd be easier in case of using structs | 22:41 |
wiking | ;) | 22:41 |
wiking | so argmax is a struct | 22:41 |
wiking | http://www.swig.org/Doc1.3/SWIG.html#SWIG_nn30 | 22:42 |
wiking | wtf is this? | 22:42 |
wiking | :D | 22:42 |
n4nd0 | haha ok | 22:42 |
n4nd0 | if you want to fight for function pointers then sonney2k is your man not me :) | 22:43 |
wiking | no worries | 22:43 |
wiking | that'll never happen | 22:43 |
@sonney2k | exactly :D | 22:43 |
wiking | ok go with the implementation that u and nico agreed on | 22:43 |
* wiking knows that there's no way to convince sonney2k of anything that he hasn't agreed with from the beginning :P | 22:44 | |
n4nd0 | sonney2k: if it could work with SWIG, is it just a problem with the documentation? | 22:44 |
wiking | n4nd0: "Occasionally, a C library may include functions that expect to receive pointers to functions--possibly to serve as callbacks. SWIG provides full support for function pointers provided that the callback functions are defined in C and not in the target language." | 22:44 |
n4nd0 | wiking: yeah, I read that | 22:44 |
wiking | so one could not define a function for argmax in python | 22:44 |
@sonney2k | wiking, sure I read that doc | 22:45 |
@sonney2k | but have you seen how much effort it is? | 22:45 |
wiking | and pass it | 22:45 |
wiking | sonney2k: it's bs | 22:45 |
@sonney2k | you will need to add some rules for every function ptr | 22:45 |
n4nd0 | wiking: for that we would use director classes | 22:45 |
wiking | sonney2k: so i'm really not pushing this agenda :P | 22:45 |
wiking | i mean it's really a nightmare | 22:45 |
@sonney2k | exactly | 22:46 |
@sonney2k | I have no idea how directors (which are crazy enough already) would work with this stuff | 22:46 |
blackburn | I admit I was pretty near to commit code containing for_each and std::list day before yesterday | 22:47 |
n4nd0 | blackburn: std rebellion!! | 22:47 |
n4nd0 | I vote for it | 22:48 |
wiking | blackburn: :DDD | 22:48 |
blackburn | however later I have changed it to CList :( | 22:49 |
@sonney2k | which is a good idea if you want to have serialization | 22:50 |
@sonney2k | finally apple releases a high-resolution macbook :D | 22:52 |
@sonney2k | 2880x1800 pixel - yay! | 22:52 |
wiking | :D | 22:53 |
@sonney2k | keeping font size whole shogun fits on one screen :D | 22:53 |
wiking | hahah and it can handle more than 8 gigs of ram | 22:53 |
blackburn | http://www.tresnormale.com/components/com_virtuemart/shop_image/product/fb45191b411f14614b63ea9368cac7bf.jpg | 22:53 |
blackburn | sonney2k: 2.3K$, right? | 22:54 |
@sonney2k | I guess I need to go back to academia to get it funded :D | 22:55 |
blackburn | hahha | 22:55 |
blackburn | good reason | 22:55 |
wiking | loooool | 22:55 |
@sonney2k | or I have to blackmail any of you ;-) | 22:55 |
n4nd0 | lol | 22:55 |
wiking | in KL everything was a lot cheaper than here in europe :) | 22:56 |
blackburn | KL? | 22:56 |
wiking | mac stuff at least by 200 euros | 22:56 |
wiking | (malaysia) | 22:56 |
blackburn | ah | 22:56 |
wiking | 1715 euros vs 2300 euros for the 15-inch: 2,3-GHz Retina-display | 22:57 |
wiking | :> | 22:57 |
wiking | that's almost a whooping 600 euros | 22:58 |
blackburn | it will cost 3.5K here I think | 22:58 |
blackburn | w/o ebay :D | 22:58 |
@sonney2k | I guess the display is really pricey... >1k | 22:59 |
wiking | lol wtf | 22:59 |
wiking | why is there 2 gnu in that machine? | 22:59 |
wiking | Intel HD Graphics 4000 and a NVIDIA GeForce GT 650M? | 22:59 |
wiking | *gpu | 23:00 |
blackburn | it is something common now | 23:00 |
wiking | really? cool i guess u can opencl the shit out of those ) | 23:00 |
wiking | :D | 23:00 |
blackburn | shit I have exam day after tomorrow | 23:01 |
blackburn | problem is that it will be about everything :D | 23:05 |
@sonney2k | wiking, yeah even my mac has this | 23:05 |
@sonney2k | and it is 4 years old... | 23:05 |
n4nd0 | do you like mac guys? | 23:06 |
n4nd0 | I mean if you like mac computers ... it was a bit ambiguous :P | 23:06 |
wiking | sonney2k: not mine.... :S | 23:06 |
blackburn | I have good old samsung | 23:07 |
@sonney2k | n4nd0, I do - excellent hardware. only the os is crap :D | 23:08 |
n4nd0 | sonney2k: how do manage with linux on it? | 23:08 |
n4nd0 | lot of problems? | 23:09 |
@sonney2k | n4nd0, I am running linux on macs since 8 years or so | 23:09 |
@sonney2k | for 8 years :D | 23:09 |
@sonney2k | back then (titanium powerpc based mac) | 23:09 |
@sonney2k | there was no other notebook that could handle 5 hours battery time | 23:09 |
@sonney2k | long before intel built reasonable mobile cpus | 23:10 |
-!- vikram360 [~vikram360@117.192.161.30] has joined #shogun | 23:10 | |
@sonney2k | ahh and I could hot plug a battery once the other one got empty | 23:10 |
n4nd0 | the battery in my computer sucks so badly... | 23:10 |
@sonney2k | so I had ~8 hours of battery and that is how shogun became to existence | 23:11 |
n4nd0 | nice | 23:11 |
@sonney2k | looong train rides | 23:11 |
n4nd0 | :) | 23:11 |
n4nd0 | did you code most of it in the train? | 23:11 |
@sonney2k | n4nd0, back then almost everything | 23:12 |
n4nd0 | haha | 23:12 |
blackburn | deutsche railroad should be in (C) | 23:12 |
@sonney2k | sponsord by deutsche bahn delays :D | 23:12 |
blackburn | sonney2k: did you live in poland? why so long train rides? | 23:12 |
n4nd0 | did people look what you were doing or something? | 23:12 |
blackburn | :D | 23:12 |
@sonney2k | others where having windows notebooks | 23:13 |
@sonney2k | I got asked a lot how long these macs can run on battery ... | 23:13 |
blackburn | bumps while riding should determine bad formatting at some places | 23:13 |
n4nd0 | haha | 23:14 |
@sonney2k | blackburn, I can assure you that we don't have jumping ICE trains :D | 23:16 |
blackburn | ah poor you have no bumpy roads :( | 23:17 |
@sonney2k | crazy! | 23:22 |
@sonney2k | on this galaxy s3 the led color is fully programmable | 23:22 |
@sonney2k | each app can have its own notification color... | 23:22 |
-!- heiko [~heiko@host86-177-4-176.range86-177.btcentralplus.com] has quit [Quit: Leaving.] | 23:48 | |
--- Log closed Tue Jun 12 00:00:41 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!