--- Log opened Wed Jan 30 00:00:38 2013 | ||
-!- cronor [~cronor@92.229.56.230] has quit [Quit: cronor] | 00:59 | |
-!- FSCV [~FSCV@187.210.54.165] has quit [Quit: Leaving] | 01:31 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] | 01:55 | |
shogun-buildbot | build #266 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/266 | 03:35 |
---|---|---|
-!- ptizoom [~christian@79-71-90-105.dynamic.dsl.as9105.com] has joined #shogun | 07:44 | |
-!- cronor [~cronor@fb.ml.tu-berlin.de] has joined #shogun | 09:22 | |
-!- sumit_ [ca4eaca2@gateway/web/freenode/ip.202.78.172.162] has joined #shogun | 10:30 | |
sonne|work | ptizoom ! | 10:38 |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun | 10:41 | |
sonne|work | n4nd0: I realize that an extra page won't work - we would somehow need another nesting level | 10:56 |
n4nd0 | sonne|work: mmm not sure why it won't work | 10:56 |
n4nd0 | tell me | 10:56 |
sonne|work | n4nd0: It would be good to have one page with announcements/news about the workshop, one with the program and one for registration and later one for user contributed slides/code or so | 10:57 |
n4nd0 | what is your idea? how do you want it to look like? | 10:57 |
sonne|work | but this will only work for 2013 | 10:57 |
sonne|work | if we have another WS in 2014 what then? | 10:57 |
n4nd0 | ok I see your point now | 10:57 |
n4nd0 | the idea would be to go to events | 10:58 |
n4nd0 | later to workshop 2013 | 10:58 |
n4nd0 | and then find those pages | 10:58 |
n4nd0 | right now workshop 2013 and workshop program are at the same level, which it is not that ^ idea | 10:58 |
sonne|work | yeah... | 11:00 |
-!- sumit_ [ca4eaca2@gateway/web/freenode/ip.202.78.172.162] has quit [Ping timeout: 245 seconds] | 11:24 | |
n4nd0 | sonne|work: I am talking to Bernard about it | 11:46 |
n4nd0 | we will find a solution | 11:46 |
sonne|work | cool thx | 11:47 |
n4nd0 | sonne|work: btw I still have my SO-SVM question :) | 11:48 |
n4nd0 | maybe you have a minute for it? | 11:49 |
sonne|work | which is? | 11:49 |
n4nd0 | does it handle predictions over continuous structured data? | 11:49 |
n4nd0 | for what I have read so far, the framework presented in Tsochantaridis which is what I know for SO-SVM doesn't do it | 11:50 |
n4nd0 | some papers about kernel density estimation, and joint kernel estimation handle that; but those are other things | 11:50 |
sonne|work | you mean e.g. a task like learn where roads end, e.g. split a road after x meters | 11:51 |
sonne|work | so non-discrete segmentation | 11:51 |
n4nd0 | yes | 11:52 |
n4nd0 | think of the HM-SVM too for example, the SO-SVM predicts sequence of states there | 11:53 |
n4nd0 | and each state is identified with an integer | 11:53 |
n4nd0 | what about sequences of states where each state is a real number? | 11:53 |
n4nd0 | it sounds a bit like regression I think, I have seen something called structured output regression too | 11:54 |
sonne|work | n4nd0: I think something like this is around for at least hidden markov models | 11:54 |
sonne|work | something with continuous state space hmms | 11:55 |
n4nd0 | I have been looking around quite a lot but didn't find anything promising to tell the truth | 11:55 |
sonne|work | I just had a look | 11:56 |
sonne|work | wikipedia names it also like this | 11:56 |
sonne|work | no idea about concrete papers though | 11:57 |
sonne|work | didn't need it yet | 11:57 |
n4nd0 | sonne|work: I looke up in wikipedia, they called it linear dynamical system | 11:58 |
blackburn | I am still convinced linear dynamical systems are related to math :D | 11:59 |
sonne|work | sounds like differential equation | 11:59 |
n4nd0 | blackburn: isn't machine learning related to math? :) | 11:59 |
blackburn | I consider machine learning as a tool to *estimate* dependencies not known before | 12:00 |
n4nd0 | state spaces are represented via differential equations | 12:00 |
n4nd0 | blackburn: don't you find that correspondence in this problem? | 12:00 |
n4nd0 | estimate the dependencies between observations and states that have produced those observations | 12:01 |
blackburn | well differential equation is known dependency with unknown parameters | 12:01 |
n4nd0 | in addition to dependencies between states in a chain themselves | 12:01 |
blackburn | okay, I just can't imagine so-svm find something for that :) | 12:02 |
n4nd0 | blackburn: why not? | 12:02 |
n4nd0 | blackburn: does it seem natural for you if the states are discrete? i.e. in a more common HMM | 12:03 |
blackburn | I don't know - a little more | 12:04 |
sonne|work | blackburn: well my example above a task like learn where roads end, e.g. split a road after x meters would be continuous state space | 12:05 |
blackburn | sonne|work: and estimate that from say video? | 12:06 |
sonne|work | yes | 12:06 |
blackburn | that sounds pretty complex to handle with anything linear | 12:07 |
sonne|work | though video is descrete :) | 12:07 |
blackburn | well I have seen SO used for stereo disparity estimation | 12:07 |
blackburn | n4nd0: is that the case? | 12:07 |
n4nd0 | blackburn: yes, it is an example | 12:08 |
blackburn | then I am more convinced it is possible | 12:08 |
n4nd0 | people use SO for human pose estimation all the time | 12:08 |
n4nd0 | there are a bunch of papers about it | 12:08 |
n4nd0 | that is continuous state space | 12:08 |
blackburn | the main problem is features construction I guess | 12:08 |
n4nd0 | I would guess for SIFT and SURF type of features | 12:09 |
blackburn | with keypoints? | 12:10 |
blackburn | or dense? | 12:10 |
n4nd0 | idk | 12:11 |
ptizoom | hi... I am lost.... | 13:05 |
n4nd0 | hi ptizoom | 13:09 |
n4nd0 | what is it? | 13:09 |
ptizoom | I spend a lot of time trying to link different idea on auto-learning instead of trying to finishing off this swig-perl with sparse matrix... | 13:09 |
n4nd0 | ok | 13:10 |
n4nd0 | you probably learnt a lot on auto-learning :) | 13:10 |
ptizoom | also still looking for a smart way to translate 'python' tests into 'perl' tests! | 13:11 |
n4nd0 | I see | 13:11 |
ptizoom | please direct me into a nice group who does this kind of stuff "learning and matching parsing relations"... | 13:13 |
n4nd0 | I have never used that | 13:14 |
ptizoom | yep, parsers are nice candidates to be auto-learned! I thought using hidden markov approx (like every one), but really how to picking up local features an work relevant relations in an orthogonal (kernelized) fashion... | 13:21 |
ptizoom | eg. does someone know how to create a stemmer ? I believe HMM algo is not enough. | 13:23 |
ptizoom | even more it seems shogun can only use a 2-gram HMM..... | 13:25 |
ptizoom | ho well... may be see you in Berlin! | 13:28 |
n4nd0 | :) | 13:47 |
sonne|work | ptizoom: yeah come to berlin :) | 13:52 |
n4nd0 | ptizoom: regarding the stemmer | 13:55 |
n4nd0 | I suggest you to take a look to http://nlp.stanford.edu/IR-book/ | 13:56 |
n4nd0 | they have very good material in there, slides, book, etc | 13:56 |
-!- FSCV [~FSCV@187.210.54.165] has joined #shogun | 14:16 | |
ptizoom | thanks n4nd0,(I like this name), I have been rumbling my soul at nlp.standford.edu indeed. I even got up to taste the madness of Wolfram...and his amazing cellular automatons... that shall be the way to program (he said), I ended up in looking into the F4 algo to reduce critical pairs...yes, one to select features needs to make some choices.... | 15:38 |
n4nd0 | it sounds like you had lot of fun :) | 15:40 |
ptizoom | yes, but I still do not get it. | 15:40 |
ptizoom | it is something beautiful , about ways of rewriting equations, involving the golden ratio....! | 15:45 |
-!- sumit [ca4eaca2@gateway/web/freenode/ip.202.78.172.162] has joined #shogun | 15:46 | |
ptizoom | anyway, yes http://nlp.stanford.edu/IR-book/ will do! Thanks for the tip. | 15:50 |
-!- sumit [ca4eaca2@gateway/web/freenode/ip.202.78.172.162] has quit [Ping timeout: 245 seconds] | 16:09 | |
n4nd0 | you are welcome | 16:10 |
-!- kakashi_ [~kakashi_@nltk/kakashi] has joined #shogun | 16:14 | |
kakashi_ | Hi, does shogun have an SVD implementation? If so, is there a performance analysis of the same? | 16:14 |
sonne|work | blackburn: ^ | 16:17 |
kakashi_ | well, I was just browsing through the code and foundout a rough jacobi implementation? | 16:18 |
kakashi_ | or it seems to use eigen++, hmmm | 16:19 |
n4nd0 | kakashi_: we mainly use eigen or arpack for linear algebra | 16:22 |
kakashi_ | n4nd0: aha, can you give me any links? also, can I use sparse matrices? | 16:26 |
n4nd0 | kakashi_: where do you want to use the sparse matrices? | 16:26 |
n4nd0 | we have some nice sparse features :) | 16:26 |
n4nd0 | links to where? examples of where svd is used in shogun? | 16:27 |
kakashi_ | yes, well, liknks to some examples | 16:27 |
kakashi_ | at present I have a sparse matrix in scipy.sparse or pysparse format (and ofcourse their combinations including a (row, column, data) as a csv is also doable) | 16:29 |
kakashi_ | dimensions are approx (10000x200000) | 16:29 |
n4nd0 | I will be back in a few minutes | 16:30 |
kakashi_ | sure | 16:31 |
kakashi_ | take your time | 16:31 |
n4nd0 | back | 16:38 |
n4nd0 | kakashi_: in multiclass/QDA.cpp SVD is used IIRC | 16:38 |
n4nd0 | yeah exactly | 16:38 |
n4nd0 | wrap_svd | 16:38 |
n4nd0 | wrap_dgesvd sorry | 16:39 |
kakashi_ | I see, is there any python wrapper, by anychance? | 16:41 |
sonne|work | nope | 16:42 |
kakashi_ | hmm, btw, this is SVD of a dense matrix, right? | 16:44 |
n4nd0 | kakashi_: I think that example probably is | 16:46 |
n4nd0 | what do you want to do? svd for a sparse matrix from python? | 16:46 |
n4nd0 | shouldn't there be something in scipy for that? | 16:47 |
kakashi_ | python would be great (numpy/scipy/sparsesvd are pretty slow) | 16:47 |
kakashi_ | scipy.sparse uses arapack implementation, a single svd of the 10000x200000 takes approx 2-3 days | 16:48 |
-!- heiko [~heiko@nat-162-125.internal.eduroam.ucl.ac.uk] has joined #shogun | 16:51 | |
n4nd0 | hehe that's lot of time | 16:51 |
n4nd0 | I am not aware if there are faster implementations | 16:51 |
n4nd0 | maybe there are some approximate methods out there | 16:51 |
kakashi_ | hmm, thanks anyways! | 16:54 |
sonne|work | kakashi_: might be that blackburn knows though... | 16:56 |
kakashi_ | aha! sure, I will keep an eye on it ... | 16:57 |
kakashi_ | I have not seen anything that is as good as mahout's implementation, but the pity is all my code is in java and JPype may obfuscate the code :( | 16:58 |
kakashi_ | s/java/python | 16:58 |
-!- cavalier- [~cavalier-@70.118.110.73] has quit [Ping timeout: 260 seconds] | 17:25 | |
-!- cavalier- [~cavalier-@70.118.110.73] has joined #shogun | 17:26 | |
n4nd0 | kakashi_: do you happen to know what kind of implementation do the use in mahout? | 17:35 |
-!- cronor [~cronor@fb.ml.tu-berlin.de] has quit [Quit: cronor] | 18:00 | |
-!- cavalier- [~cavalier-@70.118.110.73] has quit [Ping timeout: 276 seconds] | 18:09 | |
-!- cavalier- [~cavalier-@70.118.110.73] has joined #shogun | 18:09 | |
kakashi_ | I think they use jacobi implementation | 18:24 |
kakashi_ | n4nd0: i will be back in a while | 18:27 |
blackburn | kakashi_: we have some code ported from redsvd | 18:39 |
blackburn | may be it is of interest for you | 18:40 |
-!- cronor [~cronor@g226044049.adsl.alicedsl.de] has joined #shogun | 18:43 | |
blackburn | kakashi_: in my experience it doesn't work that well for sparse matrices due to randomization but I think that's quite ok still | 18:44 |
blackburn | kakashi_: http://code.google.com/p/redsvd/ | 18:46 |
-!- heiko [~heiko@nat-162-125.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 18:49 | |
-!- heiko [~heiko@nat-162-125.internal.eduroam.ucl.ac.uk] has joined #shogun | 18:50 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Ping timeout: 245 seconds] | 19:26 | |
-!- heiko [~heiko@nat-162-125.internal.eduroam.ucl.ac.uk] has left #shogun [] | 20:26 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun | 20:37 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] | 21:31 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun | 21:49 | |
@sonney2k | n4nd0, what did bernard say? | 23:34 |
n4nd0 | sonney2k: that we have to change some css and python code | 23:36 |
n4nd0 | he said he is a bit busy now | 23:36 |
n4nd0 | but I will try to do it with his mentorship :D | 23:36 |
blackburn | sonney2k: so does that mean I should answer you? | 23:41 |
blackburn | :D | 23:41 |
@sonney2k | blackburn, ? | 23:41 |
blackburn | you submitted the form | 23:41 |
@sonney2k | n4nd0, ok | 23:41 |
@sonney2k | blackburn, yes I am trying to automate the stuff | 23:42 |
@sonney2k | like that the mailinglist gets the form | 23:42 |
blackburn | sonney2k okay - am I subscribed to that list btw? | 23:43 |
@sonney2k | A user made changes from 1/31/13 2:40 AM to 2:40 AM (Azerbaijan Standard Time) ? | 23:43 |
@sonney2k | what? | 23:43 |
@sonney2k | blackburn, IDK | 23:43 |
blackburn | sonney2k: yeah so you live in azerbaijan | 23:43 |
blackburn | it is not the best notification we could see | 23:44 |
@sonney2k | yeah | 23:44 |
@sonney2k | not very helpful | 23:44 |
blackburn | yeah we can view the lnie but still | 23:44 |
@sonney2k | blackburn, would even say that this is useless | 23:44 |
@sonney2k | so I think it would be better you reply everyone submitting manually | 23:45 |
blackburn | sonney2k: yeah I see no other way so far | 23:46 |
@sonney2k | it is no use to create any kind of automagic notification | 23:46 |
blackburn | sonney2k: some js could help | 23:46 |
@sonney2k | so about the mailinglist | 23:47 |
blackburn | but nobody wants to do that | 23:47 |
@sonney2k | how should we do it? | 23:47 |
@sonney2k | subscribe people to it? | 23:47 |
blackburn | hmm can you let me send mails from that list? | 23:47 |
@sonney2k | I originally wanted to use it for notifications | 23:47 |
@sonney2k | such that heiko/you/me etc get an email | 23:48 |
@sonney2k | but now idk | 23:48 |
@sonney2k | blackburn, you can use whatever address you want in from: | 23:48 |
blackburn | well probably the best way could be - I subscribe for these notices | 23:49 |
blackburn | once somebody submits the form I write him a mail from @shogun-toolbox.org | 23:49 |
blackburn | and CC it to workshop@shogun-toolbox.org | 23:49 |
blackburn | which has you, me, .... in subscribers | 23:50 |
blackburn | gut? | 23:50 |
blackburn | sleep time | 23:52 |
@sonney2k | blackburn, yup! | 23:53 |
@sonney2k | blackburn, or alternatively you subscribe everyone registering to that mailinglist and we can use it for updates later | 23:53 |
blackburn | sonney2k: we need something private then | 23:53 |
@sonney2k | n4nd0, hmmhh I tried to embed the form | 23:53 |
blackburn | to BCC | 23:53 |
@sonney2k | blackburn, ? | 23:54 |
@sonney2k | blackburn, mailinglist is not giving out email addresses of others... | 23:54 |
blackburn | I mean if we subscribe they will receive answers purposed for other people | 23:54 |
@sonney2k | n4nd0, but it does not like me | 23:54 |
@sonney2k | blackburn, ahh I see yes sure | 23:55 |
blackburn | lets just use it for our purposes | 23:55 |
@sonney2k | blackburn, might be enough to just send around these emails among us few (all in CC) | 23:55 |
n4nd0 | mmm | 23:55 |
blackburn | hmm yes | 23:55 |
blackburn | then mailing list is not needed for all | 23:55 |
n4nd0 | sonney2k: what can it be? | 23:55 |
@sonney2k | n4nd0, it converts the iframe to some text | 23:55 |
n4nd0 | funny | 23:55 |
@sonney2k | that is the code we need to insert to have it directly on the page <iframe src="https://docs.google.com/spreadsheet/embeddedform?formkey=dEFoS0Q2U19qYkdjT2VReFNvNDBoclE6MQ" width="760" height="1629" frameborder="0" marginheight="0" marginwidth="0">Loading...</iframe> | 23:56 |
@sonney2k | n4nd0, but it removes iframes | 23:56 |
blackburn | isn't iframe something from 1992? | 23:56 |
@sonney2k | blackburn, only for people that attend | 23:56 |
blackburn | hmm I am probably wrong | 23:57 |
@sonney2k | blackburn, click 'embed on website' in the spreadsheet | 23:57 |
@sonney2k | blackburn, I guess you meant frames | 23:57 |
blackburn | sonney2k: yeah I mix it up with something else | 23:57 |
blackburn | yeah | 23:57 |
blackburn | frames are left behind before your first kiss | 23:57 |
blackburn | :D | 23:57 |
blackburn | sonney2k: do you mean integrating it to the page? | 23:59 |
blackburn | isn't better to just put a url? | 23:59 |
--- Log closed Thu Jan 31 00:00:37 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!