--- Log opened Wed May 30 00:00:41 2012 | ||
n4nd0 | good night people | 00:43 |
---|---|---|
blackburn | nite | 00:43 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 00:45 | |
-!- av3ngr [av3ngr@nat/redhat/x-xnbshjkyvrpdknhi] has joined #shogun | 02:11 | |
-!- ckwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has quit [Remote host closed the connection] | 03:22 | |
-!- blackburn [~blackburn@31.28.59.65] has quit [Quit: Leaving.] | 03:28 | |
-!- uricamic [~uricamic@2001:718:2:1634:a0dd:d302:38df:5709] has joined #shogun | 08:50 | |
-!- heiko [~heiko@host86-176-2-66.range86-176.btcentralplus.com] has joined #shogun | 09:51 | |
-!- abn_ [av3ngr@nat/redhat/x-tijaowgwtfpejqol] has joined #shogun | 09:51 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 09:52 | |
-!- av3ngr [av3ngr@nat/redhat/x-xnbshjkyvrpdknhi] has quit [Ping timeout: 245 seconds] | 09:55 | |
sonne|work | heiko: whats up? | 10:00 |
heiko | sonne|work, hi | 10:00 |
sonne|work | hey ... | 10:00 |
heiko | currently: NIPS, exam on friday, (so far: exam yesterday) | 10:01 |
sonne|work | crazy | 10:01 |
sonne|work | hope you survive this | 10:01 |
heiko | but also reading stuff to implement eigenspectrum based threshold estimate for quadratic MMD | 10:01 |
heiko | yeah me too :) | 10:01 |
sonne|work | you did ping me yesterday? | 10:01 |
heiko | yes, I was just asking whether shogun had eigenvalue methods, but then I found them myself | 10:01 |
heiko | I added this little convienience methods using SGVectors though | 10:02 |
heiko | whats up in Berlin? | 10:02 |
sonne|work | heiko: yeah I wanted to move more functions from CMath -> SGVector etc | 10:02 |
sonne|work | so you patch is very welcome | 10:02 |
heiko | so all of them should be converted? | 10:02 |
sonne|work | heiko: in two days I will be on paternal leave | 10:03 |
heiko | I would love to see the display methods | 10:03 |
sonne|work | so just finishing up stuff | 10:03 |
sonne|work | heiko: yes exactly | 10:03 |
heiko | paternal leave? | 10:03 |
sonne|work | you know play time with kids ;-) | 10:03 |
heiko | sonne|work, ok then Ill convert all methods I touch | 10:03 |
heiko | ah ok | 10:03 |
heiko | nice | 10:03 |
heiko | dont forget to do that over all the coding :D | 10:03 |
sonne|work | heiko: no worries will be on holidays too so no time for coding next week | 10:04 |
heiko | sonne|work, nice! | 10:04 |
heiko | sonne|work, I really like these director classes, havent tried them yet though, but its awesome | 10:05 |
sonne|work | yeah I think so too | 10:05 |
sonne|work | only problem is speed penalty | 10:05 |
sonne|work | so factor 10 slower... | 10:05 |
sonne|work | but hey could be worse | 10:05 |
heiko | yes, but I mean | 10:06 |
heiko | sometimes people that dont want to get involved into internals just want a new kernel | 10:07 |
heiko | or a new other thing | 10:07 |
heiko | and now thats possible | 10:07 |
blackburn | heiko: do you need full spectrum? | 10:08 |
sonne|work | yeah. but we have to enable that one by one | 10:09 |
heiko | blackburn, no, eigenvalues are fine so far, still experimenting a bit | 10:09 |
sonne|work | it will only be possible case by case | 10:09 |
blackburn | heiko: all eigenvalues? | 10:09 |
heiko | sonne|work, but thats fine | 10:09 |
sonne|work | we need to figure out for which cases we need this stuff... | 10:09 |
heiko | blackburn, yes, so far, although I will sort them and only use most numerically stable ones | 10:10 |
blackburn | for max/min you can use arpack I mean | 10:10 |
heiko | ah ok | 10:10 |
heiko | currently using the lapack based compute_eigenvalues thingi | 10:10 |
blackburn | a lot of options | 10:10 |
blackburn | heiko: dsyev? | 10:11 |
heiko | yes that one | 10:11 |
heiko | isnt arpack more for sparse matrices? | 10:11 |
heiko | will be back in 5 mins ... | 10:12 |
blackburn | heiko: no - reverse interface - it is like COFFIN for eigenstuff | 10:12 |
blackburn | it depends on structure of matrix | 10:12 |
blackburn | basically it is pretty related to krylov subspace method | 10:13 |
blackburn | which require span of [ Ax A^2 x .... ] | 10:13 |
blackburn | so for example like in power method for largest eigenvalue it would require only few matrix vector products | 10:14 |
sonne|work | blackburn: btw I was thinking about getting DirectorDOTFeatures to work | 10:14 |
blackburn | what do you have in mind | 10:14 |
blackburn | ? | 10:14 |
sonne|work | and I think one way to do it is to define all these add/dense_dot etc operations again but based on SGVector | 10:14 |
sonne|work | so then the low-level functions would call the sgvector ones | 10:15 |
sonne|work | and these are the ones that via typemaps etc are nicely exposed to python and other langs | 10:16 |
blackburn | yes obvious | 10:16 |
blackburn | hmm while you are available | 10:17 |
blackburn | what do you think about openmping it gradually? | 10:17 |
blackburn | I attempted to openmp everything but it needs to be rebased now totally | 10:17 |
heiko | re | 10:18 |
sonne|work | blackburn: regarding openmp - maybe you can openmp the most difficult part, e.g. dotfeatures first. if this works then I think it will work for all stuff | 10:19 |
sonne|work | (works == same speed as w/ pthreads) | 10:20 |
blackburn | yes but shall I commit it then? | 10:21 |
blackburn | really hard to track one more branch | 10:22 |
heiko | blackburn, I will just have a look how it works and compare against my matlab implementation and then get back to you | 10:23 |
blackburn | ok | 10:24 |
heiko | have to study kernel covariance operator now :) | 10:25 |
-!- abn_ [av3ngr@nat/redhat/x-tijaowgwtfpejqol] has quit [Ping timeout: 248 seconds] | 10:40 | |
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 244 seconds] | 11:38 | |
-!- heiko1 [~heiko@host86-183-72-132.range86-183.btcentralplus.com] has joined #shogun | 13:06 | |
-!- heiko [~heiko@host86-176-2-66.range86-176.btcentralplus.com] has quit [Ping timeout: 260 seconds] | 13:07 | |
-!- gsomix [~gsomix@37.61.181.114] has joined #shogun | 14:18 | |
gsomix | hello | 14:19 |
-!- alex_thedatawhis [82955843@gateway/web/freenode/ip.130.149.88.67] has joined #shogun | 14:52 | |
-!- heiko1 [~heiko@host86-183-72-132.range86-183.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 14:58 | |
-!- alex_thedatawhis [82955843@gateway/web/freenode/ip.130.149.88.67] has quit [Quit: Page closed] | 15:18 | |
gsomix | sonne|work, hey. I'm sorry, I will get back to work at evening. Exam... tomorrow. Preparing is needed. | 15:19 |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has joined #shogun | 15:46 | |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 16:16 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 16:38 | |
n4nd0 | summer, yaay!! | 16:39 |
-!- alexdatawhispere [82955843@gateway/web/freenode/ip.130.149.88.67] has joined #shogun | 17:04 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 17:28 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 17:29 | |
-!- uricamic [~uricamic@2001:718:2:1634:a0dd:d302:38df:5709] has quit [Quit: Leaving.] | 17:29 | |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has joined #shogun | 17:36 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 17:52 | |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 18:21 | |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has joined #shogun | 18:36 | |
-!- heiko [~heiko@host86-177-112-17.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 18:40 | |
blackburn | n4nd0: did you finish your exams? | 18:55 |
n4nd0 | blackburn: yeah! | 19:11 |
blackburn | I just came with cool game - everyone say some math word and I have to put some sentence with it to my thesis :D | 19:39 |
gsomix | blackburn, topology | 19:39 |
blackburn | easy | 19:40 |
blackburn | n4nd0: your turn | 19:41 |
n4nd0 | blackburn: let me think... | 19:41 |
gsomix | blackburn, ok. then I think I need to consult with Verbitsky, hehe | 19:42 |
n4nd0 | blackburn: curl | 19:43 |
blackburn | n4nd0: hah ok! | 19:44 |
blackburn | ok currently diffeomorphism, curl, Darboux sum, topology (gsomix want to change?) | 19:48 |
blackburn | puffin444: surprise me | 19:48 |
blackburn | sonney2k: and you :) | 19:48 |
puffin444 | Noetherian Rings | 19:48 |
gsomix | blackburn, nope. | 19:49 |
blackburn | oh sh | 19:49 |
puffin444 | Did I break your thesis? | 19:49 |
blackburn | yes probably it is not about svms anymore | 19:49 |
blackburn | :D | 19:49 |
blackburn | however one guy said Klein group already | 19:51 |
blackburn | uh I asked him to change :D | 19:51 |
blackburn | affinity now | 19:51 |
blackburn | puffin444: could you please change? :D | 19:51 |
blackburn | sounds like a bit of hardcore to put rings to svms :D | 19:52 |
puffin444 | Uhh... Dense Set | 19:52 |
blackburn | hah ok! | 19:52 |
blackburn | thanks :) | 19:53 |
blackburn | in summary, I need to add affinity, curl, dense set, topology, diffeomorphism, darboux sum | 19:55 |
blackburn | okk sounds crazy :D | 19:57 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 20:02 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 20:13 | |
puffin444 | Anybody here an expert in blas? | 21:53 |
blackburn | well not expert but have some experience | 21:54 |
blackburn | what is the issue? | 21:54 |
puffin444 | So I am calling cblas_dposv, Solves Ax=B where A is a positive symmetric matrix | 21:54 |
blackburn | right | 21:55 |
-!- cwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has joined #shogun | 21:55 | |
puffin444 | It is somehow possible that the contents of A get modified during the function? | 21:55 |
blackburn | yes it modifies A | 21:55 |
blackburn | stores cholesky upper triangular L | 21:55 |
puffin444 | Should I be calling clapack_dposv or dposv_? | 21:57 |
blackburn | clapack_dposv | 21:59 |
blackburn | puffin444: in mathematics/lapack.h we provide impl of clapack_dposv that calls usually fortran extern dposv_ | 22:14 |
blackburn | if atlas is available no need to define clapack_dposv so it will be used from atlas with the same interface | 22:14 |
puffin444 | I think I have been just using the wrong function all along | 22:15 |
blackburn | where? | 22:16 |
puffin444 | There got it now. I should have been using clapack_dpotrs as I already had the factorization | 22:18 |
blackburn | ah yes | 22:18 |
blackburn | dpotrf and dpotrs | 22:18 |
blackburn | dposv is dpotrf and dpotrs basically | 22:18 |
blackburn | however multiple calls of dposv make no sense | 22:19 |
blackburn | it would factor factors | 22:19 |
puffin444 | Exactly | 22:19 |
puffin444 | Yay! I can calculate the gradients now! | 22:27 |
blackburn | heh cool | 22:28 |
puffin444 | Ok. I think its time to take a break. See you later. | 22:30 |
blackburn | cu | 22:35 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed] | 22:37 | |
-!- alexdatawhispere [82955843@gateway/web/freenode/ip.130.149.88.67] has quit [Quit: Page closed] | 22:38 | |
n4nd0 | blackburn: hey, I got one question | 22:44 |
n4nd0 | blackburn: why is it that we cannot have modular interface for matlab? | 22:44 |
blackburn | easy question - swig does not support it :D | 22:44 |
n4nd0 | aham I see | 22:44 |
n4nd0 | my friend is using shogun for his thesis | 22:45 |
n4nd0 | he started with octave and it was all right | 22:45 |
n4nd0 | but he needed some OOP and it looks like octave is somewhat limited there | 22:45 |
blackburn | python is the key to everything | 22:45 |
n4nd0 | so he's now in matlab but found that there's no modular interface there | 22:45 |
n4nd0 | I think he will manage with static in any case | 22:46 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 22:59 | |
-!- heiko [~heiko@host86-169-107-96.range86-169.btcentralplus.com] has joined #shogun | 23:05 | |
gsomix | good night guys | 23:25 |
gsomix | exam is tomorrow... networks | 23:26 |
blackburn | good luck | 23:27 |
heiko | good luck | 23:27 |
gsomix | tnx | 23:28 |
blackburn | heiko: are you busy right now? | 23:29 |
heiko | no, resting :) | 23:30 |
heiko | blackburn, but in generell yes | 23:30 |
heiko | NIPS till friday | 23:30 |
heiko | and I got an exam on friday also | 23:30 |
blackburn | I have one ML question | 23:30 |
blackburn | heh | 23:30 |
blackburn | what is it is about? | 23:30 |
heiko | exam or nips? | 23:30 |
heiko | I mean, I if I can help you, I will enjoy doing it :) whats the question? | 23:31 |
blackburn | heiko: do you know what is L1/Lq norm? | 23:32 |
blackburn | both exam and nips | 23:32 |
heiko | blackburn, exam is about kernel theory: RKHS theory, some kernel algorithms, distributions in feature space, MMD, a bit of convex optimisation | 23:32 |
blackburn | heh cool | 23:32 |
heiko | nips working title is kernel selection for large scale two sample tests | 23:32 |
heiko | yes the course was pretty cool | 23:32 |
heiko | I know Lq norm | 23:33 |
blackburn | I have troubles deriving dual of some special task | 23:33 |
heiko | using Lagrangian? | 23:33 |
blackburn | L1/Lq is basically if you divide your vector to some cells and compute Lq of it | 23:33 |
blackburn | summarizing it with L1 | 23:33 |
blackburn | L1/L2 of [1 2 3 4] = (1+4) + (9+16) | 23:34 |
blackburn | if groups are first twos and second | 23:34 |
heiko | oh that is used in the group lasso right? | 23:34 |
blackburn | heh exactly | 23:35 |
blackburn | the problem is I do not understand how ||w||^2 would looks like in terms of alphas :) | 23:35 |
blackburn | would look* | 23:35 |
heiko | do you have like a formal description of the problem? | 23:36 |
blackburn | no but let me try to latex it a little | 23:36 |
heiko | you can use gmailtex | 23:36 |
heiko | (thats faster /easier than whole latex procedure) | 23:37 |
blackburn | http://latex.codecogs.com/gif.latex?||w||_{1/2}%20=%20\sum_{g%20\in%20G}%20||w_g||_2 | 23:37 |
blackburn | heiko: is that idea clear for you? | 23:37 |
blackburn | let G some indices space :) | 23:38 |
heiko | mmh that graphic isnt properly displayed here | 23:38 |
blackburn | why?? | 23:38 |
blackburn | wait I start to think my question is stupid | 23:39 |
heiko | I cannot read it :( | 23:39 |
blackburn | argh | 23:40 |
heiko | background is black instead of white | 23:40 |
blackburn | :D | 23:40 |
heiko | and font is also black | 23:40 |
blackburn | heiko: ok nevermind I have to think about it a little more | 23:40 |
heiko | I am only seeing the antialiased pixels | 23:40 |
blackburn | the thing I was confused is how to compute derivate of dual lagrangian by w | 23:41 |
blackburn | if norm is complicated | 23:41 |
heiko | ah ok | 23:41 |
heiko | derivatives of norms | 23:41 |
heiko | I always do it component wise if this happens | 23:42 |
heiko | and the re-assemble the result later once I understand whats going on | 23:42 |
blackburn | in case of 2 norm it is pretty basic | 23:44 |
blackburn | just like scalars | 23:44 |
heiko | are you talking about the SVM dual? | 23:45 |
blackburn | yes | 23:46 |
blackburn | just want to get liblinear easy patched | 23:47 |
blackburn | it is based on gradient so I think if I had it I would patch it | 23:47 |
heiko | mmh, I dont really get it, but nevermind, have you solved it though? | 23:48 |
blackburn | solved what? | 23:48 |
blackburn | I think I have to formulate questions more precisely :D | 23:48 |
blackburn | just confused you | 23:48 |
heiko | yes :) | 23:49 |
heiko | well, I gotta go to bed anyway | 23:49 |
heiko | We can discuss tomorrow if you want | 23:49 |
blackburn | ok :) | 23:50 |
heiko | I will be st | 23:50 |
heiko | I will be online during the day since I have to study :) | 23:50 |
heiko | good night | 23:51 |
-!- heiko [~heiko@host86-169-107-96.range86-169.btcentralplus.com] has left #shogun [] | 23:51 | |
blackburn | good night | 23:51 |
--- Log closed Thu May 31 00:00:41 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!