--- Log opened Sun Aug 14 00:00:28 2011 | ||
shogun-buildbot | build #13 of octave_modular is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org:5347/builders/octave_modular/builds/13 blamelist: shashwat@Xan-Kriegor.(none) | 00:13 |
---|---|---|
f-x | sonney2k: can you identify the error in the log? i'm unable to find anything | 00:19 |
f-x | shogun-buildbot: status | 00:21 |
shogun-buildbot | cmdline_static: idle, last build 26m37s ago: build successful | 00:21 |
shogun-buildbot | csharp_modular: idle, last build 12m02s ago: build successful | 00:21 |
shogun-buildbot | java_modular: idle, last build 16m37s ago: build successful | 00:21 |
shogun-buildbot | libshogun: idle, last build 32m22s ago: build successful | 00:21 |
shogun-buildbot | lua_modular: idle, last build 11m59s ago: build successful | 00:21 |
shogun-buildbot | nightly_all: idle, last build 3h21m39s ago: failed configure | 00:21 |
shogun-buildbot | nightly_none: idle, last build 3h14m35s ago: build successful | 00:21 |
shogun-buildbot | octave_modular: idle, last build 7m56s ago: failed compile | 00:21 |
shogun-buildbot | octave_static: idle, last build 25m17s ago: build successful | 00:21 |
shogun-buildbot | python_modular: idle, last build 3m44s ago: build successful | 00:21 |
shogun-buildbot | python_static: idle, last build 25m33s ago: build successful | 00:21 |
shogun-buildbot | r_modular: idle, last build 3m48s ago: build successful | 00:21 |
f-x | shogun-buildbot: commands | 00:22 |
shogun-buildbot | buildbot commands: commands, dance, destroy, excited, force, hello, help, last, list, mute, notify, source, status, stop, unmute, version, watch | 00:22 |
f-x | shogun-buildbot: last | 00:22 |
shogun-buildbot | last build [cmdline_static]: 27m12s ago: build successful | 00:22 |
shogun-buildbot | last build [csharp_modular]: 12m37s ago: build successful | 00:22 |
shogun-buildbot | last build [java_modular]: 17m11s ago: build successful | 00:22 |
shogun-buildbot | last build [libshogun]: 32m56s ago: build successful | 00:22 |
shogun-buildbot | last build [lua_modular]: 12m34s ago: build successful | 00:22 |
shogun-buildbot | last build [nightly_all]: 3h22m14s ago: failed configure | 00:22 |
shogun-buildbot | last build [nightly_none]: 3h15m10s ago: build successful | 00:22 |
shogun-buildbot | last build [octave_modular]: 8m30s ago: failed compile | 00:22 |
shogun-buildbot | last build [octave_static]: 25m52s ago: build successful | 00:22 |
-!- alesis-novik [~alesis@cpat001.wlan.net.ed.ac.uk] has quit [Read error: Operation timed out] | 00:22 | |
shogun-buildbot | last build [python_modular]: 4m19s ago: build successful | 00:22 |
shogun-buildbot | last build [python_static]: 26m08s ago: build successful | 00:22 |
shogun-buildbot | last build [r_modular]: 4m23s ago: build successful | 00:22 |
shogun-buildbot | last build [r_static]: 26m14s ago: build successful | 00:22 |
shogun-buildbot | last build [ruby_modular]: 12m27s ago: build successful | 00:22 |
-!- alesis-novik [~alesis@cpat001.wlan.net.ed.ac.uk] has joined #shogun | 00:25 | |
-!- f-x [~user@117.192.212.140] has quit [Remote host closed the connection] | 01:22 | |
-!- alesis-novik [~alesis@cpat001.wlan.net.ed.ac.uk] has quit [Ping timeout: 258 seconds] | 08:58 | |
-!- alesis-novik [~alesis@cnat038.wlan.net.ed.ac.uk] has joined #shogun | 08:59 | |
-!- in3xes [~in3xes@27.107.159.204] has joined #shogun | 10:58 | |
-!- blackburn [~blackburn@188.122.238.50] has joined #shogun | 11:02 | |
-!- in3xes [~in3xes@27.107.159.204] has quit [Ping timeout: 258 seconds] | 11:34 | |
CIA-87 | shogun: Shashwat Lal Das master * rc9a0d16 / (5 files in 2 dirs): | 12:20 |
CIA-87 | shogun: Incorporated dump regressor in VowpalWabbit. | 12:20 |
CIA-87 | shogun: Added vw_math. - http://bit.ly/pnIBwP | 12:20 |
CIA-87 | shogun: Shashwat Lal Das master * r14946c5 / (3 files in 2 dirs): Added VowpalWabbit class. - http://bit.ly/nltDft | 12:20 |
CIA-87 | shogun: Shashwat Lal Das master * r1fe7294 / (4 files in 3 dirs): | 12:20 |
CIA-87 | shogun: Corrected and documented StreamingSparseFeatures. | 12:20 |
CIA-87 | shogun: Added an example for OnlineSVMSGD. - http://bit.ly/nWyT7u | 12:20 |
CIA-87 | shogun: Soeren Sonnenburg master * rcb0a6ca / (10 files in 4 dirs): | 12:20 |
CIA-87 | shogun: Merge pull request #294 from frx/streaming_vw | 12:20 |
CIA-87 | shogun: VW Class - http://bit.ly/pQ4clp | 12:20 |
CIA-87 | shogun: Sergey Lisitsyn master * r5f776e8 / src/shogun/preprocessor/PCA.cpp : Got PCA to work in the right way again - http://bit.ly/o0mBVw | 12:55 |
-!- in3xes [~in3xes@180.149.49.230] has joined #shogun | 13:16 | |
-!- in3xes [~in3xes@180.149.49.230] has quit [Remote host closed the connection] | 13:16 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 13:17 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 14:20 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 258 seconds] | 14:24 | |
-!- in3xes__ [~in3xes@180.149.49.227] has joined #shogun | 14:47 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 276 seconds] | 14:51 | |
-!- in3xes__ is now known as in3xes | 16:22 | |
-!- blackburn [~blackburn@188.122.238.50] has quit [Quit: Leaving.] | 17:45 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 18:15 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 276 seconds] | 18:19 | |
-!- in3xes_ is now known as in3xes | 18:47 | |
-!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has quit [Read error: Connection reset by peer] | 19:44 | |
-!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has joined #shogun | 19:53 | |
-!- blackburn [~blackburn@188.122.238.50] has joined #shogun | 20:13 | |
@sonney2k | blackburn, so what was wrong in PCA? | 20:15 |
blackburn | sonney2k: order of eigenvalues | 20:15 |
blackburn | better say side of eigenvalues | 20:16 |
@sonney2k | but these come from lapack? | 20:16 |
blackburn | sonney2k: CMath::compute_eigenvectors is used there | 20:17 |
@sonney2k | ok but that uses lapack too | 20:18 |
@sonney2k | so what was wrong? smallest insteaf of biggest ev's taken? | 20:19 |
blackburn | yes | 20:19 |
@sonney2k | ok but than this was only used for whitening | 20:19 |
@sonney2k | or? | 20:19 |
blackburn | nono | 20:19 |
blackburn | recall that eigenvectors are used too | 20:20 |
@sonney2k | wait so also wrong eigenvectors where used? | 20:20 |
blackburn | cov matrix is near singular so it's smallest eigenvector is 1 | 20:20 |
blackburn | well it is the same | 20:20 |
CIA-87 | shogun: Soeren Sonnenburg master * rea26eaa / src/shogun/mathematics/Math.h : use CALLOC instead of malloc and fill_vector - http://bit.ly/nh4GBu | 20:23 |
@sonney2k | blackburn, dammit I have stability issues - either because the buildbot was to greedy on my machine taking it down because it did compile 6 interfaces simultaneously or ?!?! | 20:34 |
@sonney2k | serialhex, around? | 20:34 |
serialhex | yes i am | 20:35 |
serialhex | whats up sonney2k?? | 20:36 |
@sonney2k | serialhex, how far are you with setting up the machine? | 20:36 |
@sonney2k | I could give you some periodic build thing if you are still interested... | 20:37 |
@sonney2k | if not - not problem - you could aswell do it the easy way | 20:37 |
@sonney2k | and just git clone shogun / configure / make all the time :) | 20:37 |
serialhex | it's burning in today, compiling gnome (or halted for some retardo error) but i'll be there tomorrow setting things up & will be able todo whatever | 20:37 |
@sonney2k | make -j 3 or so even :) | 20:37 |
serialhex | lol :P | 20:37 |
serialhex | dude, it's a 4 core xeon proc... 8 threads... j=3??? more like j=10!!!! | 20:38 |
serialhex | though being able to autobuild id probably better | 20:39 |
serialhex | cause it's going to be running til wed or so... so even at night it can do it | 20:39 |
@sonney2k | serialhex, yeah but compiling shogun can easily take 3 GB of memory now | 20:41 |
@sonney2k | so unless you have >30G better don't do it :) | 20:42 |
serialhex | ...and it has 8gigs atm... | 20:42 |
serialhex | of ram... and a pair of 500tb hdd's | 20:42 |
serialhex | and like i said, i want to burn this sucker in... why not stress it? | 20:42 |
serialhex | the thing is getting wiped and windows installed on it in a few days anyway | 20:42 |
blackburn | ehh will 10 threads take 10x memory? | 20:43 |
serialhex | idk... but that is what swap is for :P | 20:44 |
blackburn | serialhex: hmm could I send you some python script later to run on it? | 20:46 |
serialhex | yeah, np | 20:46 |
serialhex | i won't be able to get to it til tomorrow... but yeah that's cool | 20:46 |
blackburn | serialhex: but I probably need to have arpack installed there | 20:47 |
serialhex | ok... hopefully that's comething i can just emerge onto the system... | 20:47 |
blackburn | in debian/ubuntu it is pretty simple to install | 20:47 |
blackburn | I think the same for gentoo | 20:47 |
serialhex | yeah i installed gentoo just to be more masochistic to the machine :P | 20:48 |
serialhex | ...or sadistic? yes, sadistic | 20:48 |
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Quit: buildmaster reconfigured: bot disconnecting] | 20:50 | |
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun | 20:50 | |
@sonney2k | blackburn, yeah, 10x gcc will be 30G | 20:51 |
@sonney2k | (when all compile octave_modular) | 20:52 |
-!- f-x [~user@117.192.192.146] has joined #shogun | 20:53 | |
serialhex | ok, maybe not 10... maybe just 8... (though 10 would be fun!) | 20:53 |
@sonney2k | shogun-buildbot, status all | 20:55 |
shogun-buildbot | cmdline_static: idle, last build 3m22s ago: build successful | 20:55 |
shogun-buildbot | csharp_modular: building(no current step), (no current step) | 20:55 |
shogun-buildbot | java_modular: building(no current step), (no current step) | 20:55 |
shogun-buildbot | libshogun: building(no current step) | 20:55 |
shogun-buildbot | lua_modular: building(no current step), (no current step) | 20:55 |
shogun-buildbot | nightly_all: idle, last build 23h55m38s ago: failed configure | 20:55 |
shogun-buildbot | nightly_none: idle, last build 23h48m34s ago: build successful | 20:55 |
shogun-buildbot | octave_modular: building(compiling), (compiling) | 20:55 |
shogun-buildbot | octave_static: idle, last build 1m59s ago: build successful | 20:55 |
shogun-buildbot | python_modular: building(no current step), (no current step) | 20:55 |
shogun-buildbot | python_static: idle, last build 8h18m33s ago: build successful | 20:55 |
shogun-buildbot | r_modular: building(no current step), (no current step) | 20:55 |
shogun-buildbot | r_static: idle, last build 8h18m38s ago: build successful | 20:55 |
shogun-buildbot | ruby_modular: building(no current step), (no current step) | 20:55 |
@sonney2k | all weird | 20:56 |
@sonney2k | I hope I managed to implement interlocks to have just 1 gcc running per builder | 20:56 |
* blackburn thinks of non-scalability of shogun | 21:00 | |
serialhex | hey, do we have anything that can run on GPU's or are we all CPU??? | 21:01 |
blackburn | no :( | 21:01 |
blackburn | no GPU, no distributed memory | 21:01 |
serialhex | hmm.......... | 21:03 |
blackburn | sonney2k: hey we are gonna miss the train with no scalability.. | 21:04 |
serialhex | so blackburn & sonney2k, if you can e-mail me a list of what i may need to do to get these projects running onmonday it'd help me get them up and running quickly | 21:04 |
@sonney2k | blackburn, then go stealing :) | 21:04 |
blackburn | stealing what? | 21:04 |
@sonney2k | code | 21:04 |
serialhex | scalability stuffs!!! | 21:04 |
blackburn | not the code problem I'd say | 21:05 |
@sonney2k | blackburn, problem is that most algos in shogun do not easily benefit form GPU's etc | 21:05 |
@sonney2k | in fact GPUs are totally useless for the kind of string processing we do | 21:05 |
blackburn | but distributed memory systems.. | 21:06 |
@sonney2k | blackburn, commodity hardware will be multi-core | 21:06 |
@sonney2k | (or is already) | 21:06 |
@sonney2k | do you have a system like this at home? | 21:06 |
@sonney2k | or access to one? | 21:07 |
@sonney2k | I don't | 21:07 |
@sonney2k | well you can rent amazon ec2 / azure ... | 21:07 |
@sonney2k | but what for... | 21:07 |
serialhex | hopefully i'll have one soon... (maybe not an AWESOMEZOMGUBERFASTAHHHHHHH!!!! machine, but something pretty spiffy anyway) | 21:07 |
@sonney2k | blackburn, at max planck they have sth. like 128GB machines with 64 cores or so... | 21:08 |
blackburn | you are right but I'm worried of it anyway | 21:08 |
@sonney2k | and this will soon be in everyones notebook | 21:08 |
@sonney2k | so I would rather concentrate on what 'normal' people have | 21:08 |
@sonney2k | all the others (==yahoo, google, ...) can do tweaked stuff for their massive amounts of data | 21:09 |
@sonney2k | I mean they wouldn't even compute anything much more complex then mean / cov | 21:09 |
@sonney2k | because it is too expensive even on 10k nodes | 21:10 |
@sonney2k | for their petabytes of date | 21:10 |
@sonney2k | data | 21:10 |
blackburn | sure | 21:10 |
blackburn | hooray scikits learn have kernel pca | 21:11 |
blackburn | sometimes I think we do the same as scikitlearn guys do and sometimes it is worse | 21:13 |
@sonney2k | blackburn, probably not as good as voijtechs but hey :) | 21:13 |
@sonney2k | blackburn, in which respect? | 21:13 |
blackburn | at least structure | 21:13 |
blackburn | K = self.centerer.transform(self._get_kernel(X, self.X_fit_)) | 21:14 |
blackburn | return np.dot(K, self.alphas_ / np.sqrt(self.lambdas_)) | 21:14 |
blackburn | that's all for kPCA transform! | 21:14 |
blackburn | some perfectionism is going to kill me | 21:15 |
@sonney2k | c++ libraries are always more involved... | 21:16 |
@sonney2k | blackburn, I agree structure wise we can always learn from other ML toolboxes | 21:16 |
blackburn | something is wrong with me :D | 21:16 |
@sonney2k | blackburn, look we are below 1000 warnings now | 21:18 |
@sonney2k | so everything is working out just fine :D | 21:18 |
blackburn | sonney2k: yeah I killed ~200 yesterday | 21:18 |
blackburn | not fine to me, our doc is a crap and code too sometimes :( | 21:18 |
@sonney2k | great job :) | 21:19 |
@sonney2k | blackburn, however we have examples for everything | 21:20 |
@sonney2k | blackburn, and this is what people usually look at | 21:20 |
@sonney2k | true it is tough to combine things without proper documentation | 21:21 |
blackburn | I'd say they are not cover even 30% | 21:21 |
@sonney2k | but documenting everything is sooooo much work | 21:21 |
@sonney2k | examples? | 21:21 |
blackburn | yes | 21:21 |
@sonney2k | I think they cover everything | 21:21 |
blackburn | I mean as tests | 21:21 |
@sonney2k | not all options of everything of course | 21:21 |
@sonney2k | blackburn, I hope the weather improves in Russia soon - you seem to be in that hmmhh mood | 21:22 |
blackburn | the weather is good :( | 21:23 |
@sonney2k | blackburn, Re tests - impossible to cover everything | 21:23 |
@sonney2k | raining here | 21:23 |
blackburn | my brother said it is raining in st.petersburg three days long | 21:23 |
blackburn | but here is sunny | 21:24 |
blackburn | it is* | 21:24 |
@sonney2k | blackburn, anyway to improve all the issues you raise (documenttaion/ tests / design / speed(distributed algs/memory/...)) we simply need more contributors | 21:25 |
@sonney2k | totally impossible with just the few of us | 21:25 |
blackburn | bad bad bad | 21:25 |
blackburn | I feel depressed about that crap | 21:26 |
@sonney2k | blackburn, on the bright side there isn't any toolbox supporting that many $lang / algorithms | 21:26 |
@sonney2k | blackburn, which things do depress you most? | 21:26 |
blackburn | doc/code | 21:26 |
blackburn | structure is pretty ok | 21:27 |
@sonney2k | which code? | 21:27 |
@sonney2k | there is a lot of code you know... | 21:27 |
blackburn | there are a lot of crappy code | 21:27 |
blackburn | btw my code is a crap too | 21:27 |
blackburn | :D | 21:27 |
@sonney2k | blackburn, please be more specific | 21:27 |
@sonney2k | ok your code is crap and which algos are too? | 21:27 |
blackburn | the whole preprocessor thing seems to be a crap for me :D | 21:28 |
@sonney2k | blackburn, *lol* | 21:28 |
@sonney2k | documentation is something no one wants to do | 21:28 |
blackburn | what to say about classifier/svm/* | 21:28 |
@sonney2k | all the documentation there is (with formulas etc) was written be me (and I hated doing it) | 21:29 |
@sonney2k | I know some really bad code in shogun - the Trie.h and in the Weighted* kernels | 21:29 |
@sonney2k | blackburn, rest assured that I have read a lot of code from other toolboxes - I've seen much worse code | 21:31 |
blackburn | okay.. | 21:32 |
@sonney2k | blackburn, anyway it is no use to be overly critic - if you want to improve things - great - if not live goes on. As long as the algorithms work and are correct and fast - no one cares | 21:32 |
@sonney2k | the interface itself is nice IMHO | 21:32 |
blackburn | why the fuck I did that LandmarkIsomap and ClassicIsomap | 21:32 |
@sonney2k | and documentation will always be a problem | 21:32 |
@sonney2k | I mean you can write a book of >1000 pages for shogun | 21:33 |
@sonney2k | introducing all the algorithms (math behind it), implementation, examples ... | 21:33 |
@sonney2k | extending / languages ... | 21:33 |
blackburn | I would work on doc more | 21:33 |
blackburn | but my english is too russian :D | 21:34 |
@sonney2k | and mine too german :) | 21:34 |
blackburn | it is pretty difficult to describe these things in russian though | 21:34 |
@sonney2k | blackburn, why not draft sth up and ask serialhex to quickly proofread it - I mean he is a native and very interested in learning all this? | 21:34 |
blackburn | don't know if serialhex agree :) | 21:35 |
serialhex | what who where why?? | 21:35 |
* serialhex reads preceeding convo... | 21:35 | |
@sonney2k | serialhex, reading documentation that blackburn writes | 21:35 |
@sonney2k | serialhex, blackburn is afraid of his russian english not being parseable | 21:36 |
serialhex | ahh, yeah i just read | 21:36 |
serialhex | i wouldn't mind that | 21:36 |
serialhex | his english has been getting much better from when i met him at the beginning of summer, and i want to learn all that fun stuff anyway, so it's a win-win!! :D | 21:37 |
@sonney2k | serialhex, yeah I think so too. I can even understand blackburn from time to time | 21:37 |
@sonney2k | (though not the wining today) | 21:37 |
serialhex | lol :P | 21:37 |
blackburn | sonney2k: ну ничего себе комплимент :D | 21:37 |
serialhex | first rule of documentation club: no russian! :P | 21:38 |
@sonney2k | hehe | 21:42 |
@sonney2k | hab ich mir fast gedacht :) | 21:43 |
blackburn | I'll change MDS/Isomap thing now | 21:44 |
@sonney2k | blackburn, btw as you know I am one of the action editors of jmlr reviewing machine learning oss | 21:46 |
@sonney2k | blackburn, 99% of them don't even have tests when they are submitted | 21:46 |
blackburn | aha | 21:46 |
@sonney2k | nor developer documentation | 21:47 |
blackburn | hehe | 21:47 |
@sonney2k | blackburn, certainly no buildbot | 21:47 |
blackburn | sonney2k: I'm thinking about removing HLLE.. | 21:47 |
blackburn | it is equal to LTSA | 21:47 |
@sonney2k | why? | 21:47 |
blackburn | but less stable | 21:47 |
blackburn | :D | 21:48 |
@sonney2k | then just document it - I wouldn't remove it | 21:48 |
blackburn | okay | 21:48 |
@sonney2k | blackburn, every algo has its strengths | 21:48 |
@sonney2k | we have *many* svm solvers in there | 21:48 |
@sonney2k | and there is no general winner | 21:48 |
@sonney2k | some are very stable | 21:48 |
blackburn | no, results are the same! | 21:49 |
@sonney2k | some are very fast in some situation but not stable in others etc | 21:49 |
@sonney2k | yeah results *should* be the same w/ svms too but aren't | 21:49 |
blackburn | but HLLE can fuck up where LTSA will do ok | 21:49 |
blackburn | rm shogun/preprocessor/ClassicMDS.h shogun/preprocessor/ClassicMDS.cpp | 21:50 |
blackburn | what a great command i did | 21:50 |
blackburn | :D | 21:50 |
@sonney2k | still - it has its value | 21:51 |
@sonney2k | e.g. when learning about such algos | 21:52 |
@sonney2k | it is good to learn that one is not so stable | 21:52 |
@sonney2k | blackburn, did you read hhg2tg? You really sound like Marvin today. | 21:52 |
blackburn | nope, haven't readed it | 21:53 |
@sonney2k | blackburn, there is this highly skilled robot named Marvin - that becomes depressed because everything around him is so stupid... | 21:54 |
blackburn | ЖВ | 21:54 |
blackburn | :D | 21:54 |
@sonney2k | blackburn, there is a chapter with Marvin and a battle tank | 21:55 |
@sonney2k | that you should read | 21:55 |
blackburn | okay :) | 21:55 |
@sonney2k | but yes Marvin managed a spaceship to commit suicide because he made it to become depressed... | 21:56 |
@sonney2k | dammed nightly_all still fails | 21:57 |
@sonney2k | http://www.shogun-toolbox.org:5347/builders/nightly_all/builds/2/steps/compile/logs/stdio | 21:57 |
@sonney2k | everyday a new fix | 21:58 |
CIA-87 | shogun: Soeren Sonnenburg master * rfe8cd5c / src/shogun/mathematics/Math.h : call CMath::is_finite() instead of CMath::finite() - http://bit.ly/nVzFr0 | 21:58 |
blackburn | I don't like CMath | 21:59 |
@sonney2k | blackburn, so what do you propose? | 22:00 |
blackburn | nothing :( | 22:00 |
@sonney2k | blackburn, where would you put functions like finite() ? | 22:01 |
@sonney2k | or max() ? | 22:01 |
blackburn | to CMath | 22:01 |
@sonney2k | certainly things like display_matrix / etc will go | 22:01 |
@sonney2k | (to SGMatrix etrc) | 22:01 |
blackburn | I don't like anything | 22:02 |
@sonney2k | blackburn, it is about time that you /nick marvin | 22:03 |
blackburn | it is used already :( | 22:04 |
@sonney2k | blackburn, http://www.deeplearning.net/software/theano/ ? | 22:04 |
blackburn | I would say it is call if it wasn't "Theano is a Python library" | 22:05 |
blackburn | cool | 22:05 |
blackburn | call? why I wrote | 22:05 |
@sonney2k | blackburn, but it is just a matrix library... | 22:10 |
blackburn | sonney2k: some warnings here are for things under DOXYGEN_SHOULD_SKIP_THIS #ifs..? | 22:12 |
@sonney2k | blackburn, ohh then we just have to fix the Doxyfile / Makefile to really ignore them | 22:13 |
blackburn | sonney2k: shuffle should go to statistics? | 22:13 |
@sonney2k | blackburn, could be an extra random module with all the rand / randn etc stuff in there | 22:14 |
@sonney2k | in mathematics/Random.h | 22:14 |
blackburn | okay so i'd put shuffle there when it is created | 22:15 |
@sonney2k | blackburn, for now we should really focus on getting things stable and documented | 22:16 |
blackburn | yeah | 22:16 |
blackburn | but it is used in landmark mds | 22:16 |
blackburn | that's why I asked | 22:16 |
@sonney2k | then once 1.0 is release we will create some 1.0 branch | 22:16 |
@sonney2k | and probably bug fix releases will be 1.0.1 etc | 22:17 |
@sonney2k | while we continue to work on shogun 1.1 with even better structure :D | 22:17 |
@sonney2k | and your new shiny tutorial / documentation :D | 22:17 |
@sonney2k | blackburn, it looks like the buildbot now really only uses 2 cores here... now I hope that it runs longer... | 22:19 |
@sonney2k | I guess I should also ask gunnar if they have resources for more build slaves | 22:19 |
blackburn | is he really interested in shogun? | 22:20 |
blackburn | can't see his activity here and I don't like it too | 22:23 |
@sonney2k | blackburn, he is using it and sponsored quite some development... | 22:24 |
@sonney2k | blackburn, he is a big shot now so he will never touch the code again himself I guess | 22:24 |
blackburn | can't see any his student | 22:25 |
blackburn | of | 22:25 |
@sonney2k | cwidmer... | 22:25 |
blackburn | he is not active now too | 22:25 |
@sonney2k | fabio de bona before | 22:25 |
@sonney2k | and jonas behr | 22:25 |
blackburn | yeah but it is past | 22:26 |
@sonney2k | cwidmer will continue | 22:26 |
blackburn | I hope | 22:27 |
alesis-novik | sonney2k, I've been out of it for a few days, what's the current plan for Shogun (I remember bug-hunting and such being mentioned) | 22:28 |
@sonney2k | alesis-novik, please write documentation and fix undocumented things - list is here http://shogun-toolbox.org:5347/builders/python_modular/builds/14/steps/compile/logs/warnings%20(987)/text | 22:28 |
alesis-novik | sonney2k, thanks | 22:29 |
@sonney2k | blackburn, yeah but look f-x will extend his online learning framework, heiko his modelselection stuff, you and me will do random things | 22:29 |
@sonney2k | blackburn, so it will continue to improve over time | 22:30 |
blackburn | I hope so too | 22:30 |
@sonney2k | blackburn, if we manage to get into next years gsoc we can have new cool 'big' projects. | 22:31 |
blackburn | yeah I would like to participate :) | 22:31 |
alesis-novik | That would be interesting | 22:31 |
@sonney2k | but even without - it is pretty neat already now | 22:31 |
blackburn | I don't see any reasons why we can't get into gsoc | 22:32 |
@sonney2k | blackburn, I think this release has the biggest advancements we ever had between releases so... | 22:32 |
blackburn | it made huge impact on shogun | 22:32 |
@sonney2k | I agree | 22:33 |
blackburn | so it is exactly what is GSoC was intended | 22:33 |
@sonney2k | for example next year one could try to get many boosting algorithms in there | 22:33 |
alesis-novik | sonney2k, about the list of undocumented things, should I just infer what the documentation should be for stuff there? | 22:33 |
@sonney2k | blackburn, gsoc is in particular to get new long term contributers | 22:33 |
@sonney2k | alesis-novik, well document only what you think you understand :) | 22:34 |
@sonney2k | blackburn, or structured output learnign for real | 22:35 |
@sonney2k | or gaussian processes | 22:35 |
alesis-novik | gaussian processes would be interesting | 22:36 |
@sonney2k | alesis-novik, you don't have to wait for next GSoC to contribute - right? | 22:36 |
@sonney2k | alesis-novik, just do it like blackburn does if you are interested. your contributions are warmly welcome. | 22:37 |
alesis-novik | sonney2k, yeah, I'd be interested in continuing contributing | 22:37 |
@sonney2k | alesis-novik, any plans what you intend to contribute in the future? | 22:38 |
alesis-novik | sonney2k, I guess it partially depends on what is needed. There is some more stuff I could potentially implement to do with EM. | 22:38 |
alesis-novik | sonney2k, I was also wondering if time-series would possibly be in the scope of shogun? | 22:39 |
@sonney2k | alesis-novik, I wouldn't mind them in shogun - it is a new area not yet in there though | 22:40 |
alesis-novik | sonney2k, yeah, I think if it were to be added, it would be best to keep it outside of stable release until it would have some substantial features | 22:41 |
@sonney2k | alesis-novik, one feature at a time. | 22:42 |
alesis-novik | I guess it depends on the definition of feature :) Maybe one model at a time? | 22:43 |
@sonney2k | alesis-novik, yeah - e.g. one ICA algorithm at a time :) | 22:44 |
@sonney2k | blackburn, regarding the DOXYGEN_SHOULD_SKIP_THIS | 22:46 |
@sonney2k | blackburn, which file uses this and still generated a warning | 22:46 |
@sonney2k | (I have a potential fix so would like to check) | 22:46 |
blackburn | classifier/svm/ssl.h | 22:46 |
@sonney2k | ok then the fix works | 22:47 |
CIA-87 | shogun: Soeren Sonnenburg master * r4d757d9 / src/interfaces/modular/modshogun.doxy : set DOXYGEN_SHOULD_SKIP this when generating documentation - http://bit.ly/q5xYEV | 22:48 |
@sonney2k | blackburn, ~950 warnings now | 22:50 |
blackburn | uh | 22:51 |
blackburn | refactored ClassicMDS,LandmarkMDS,ClassicIsomap, LandmarkIsomap -> MultidimensionalScaling, Isomap | 22:51 |
@sonney2k | blackburn, it is becoming a blonde Russian beauty :D | 22:53 |
alesis-novik | blackburn, can you tell me what paper you used for LTSA? | 22:57 |
blackburn | yeah | 22:58 |
blackburn | wait a min | 22:58 |
blackburn | Z. Zhang, H. Zha, Principal manifolds and nonlinear dimensionality | 22:59 |
blackburn | * reduction via tangent space alignment, SIAM J. Sci. Comput. 26 (1) | 22:59 |
blackburn | * (2004) 313–338. | 22:59 |
blackburn | alesis-novik: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.4.3693&rep=rep1&type=pdf | 23:00 |
blackburn | it was hard to extract url from google stuff hehe | 23:00 |
alesis-novik | Thanks | 23:00 |
blackburn | np | 23:00 |
blackburn | alesis-novik: btw how is your dimreduction experience? | 23:09 |
alesis-novik | interesting, though nothing to report yet :) | 23:09 |
blackburn | big commit in a min | 23:51 |
@sonney2k | poor buildbot | 23:52 |
@sonney2k | no rest never ever | 23:52 |
@sonney2k | blackburn, when heiko is back on monday - I hope we can resolve the example doesn't run issues in python | 23:53 |
@sonney2k | then we can turn on make check-examples on buildbot | 23:53 |
blackburn | хорошо | 23:53 |
@sonney2k | blackburn, genau :D | 23:53 |
@sonney2k | then prime goal is to get all examples to work in all $LANG | 23:54 |
blackburn | I would be happy if someone took kernelpca.. | 23:54 |
blackburn | for some strange reason I feel depressed like marvin about kpca | 23:55 |
@sonney2k | blackburn, we should do this ASAP - we still have 1 week till the end of gsoc so e.g. alesis-novik and sploving1 f-x could help too | 23:55 |
@sonney2k | blackburn, then ignore it for now | 23:55 |
blackburn | I don't find any implementation I could port easily and have no really comprehension how to do this | 23:56 |
blackburn | didn't find* | 23:56 |
@sonney2k | blackburn, I would just port the apply function from vojtechs toolbox | 23:56 |
@sonney2k | not so difficult but I didn't have time to finish it | 23:56 |
blackburn | some strange algo is used here | 23:56 |
blackburn | or I have miscomprehension about it | 23:57 |
@sonney2k | no it is the original kPCA algorithm. Only apply() is a bit more clever - he did some centering IIRC | 23:57 |
@sonney2k | btw, http://shogun-toolbox.org:5347/waterfall gets wtf?! | 23:58 |
@sonney2k | hard to comprehend now | 23:58 |
@sonney2k | I think this is the only useful one now http://shogun-toolbox.org:5347/console | 23:58 |
alesis-novik | sonney2k, when's the intended release date? | 23:58 |
@sonney2k | alesis-novik, end of august | 23:58 |
@sonney2k | alesis-novik, for that we need the old test suite to pass and all warnings removed, everything documented | 23:59 |
blackburn | 15 aug already? | 23:59 |
blackburn | oh shit | 23:59 |
--- Log closed Mon Aug 15 00:00:08 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!