--- Log opened Wed Jul 27 00:00:31 2011 | ||
blackburn | sonney2k: could you please paste similar one for delete? | 00:02 |
---|---|---|
blackburn | ah nevermind | 00:02 |
blackburn | will do it manually | 00:02 |
@sonney2k | blackburn, what did the hooo* stand for? | 00:02 |
blackburn | LTSA is ready | 00:03 |
blackburn | and working | 00:03 |
@sonney2k | you did git clean -dfx w/o adding? | 00:03 |
@sonney2k | SCNR ;-) | 00:03 |
blackburn | this time I will add | 00:03 |
blackburn | sonney2k: could you please translate to german? :D | 00:07 |
blackburn | "Hi, Chris | 00:07 |
blackburn | Today I've finished implementing LTSA and finally learned german. Laplacian eigenmaps is on the way to be implemented." | 00:07 |
@sonney2k | Hi Chris, | 00:08 |
blackburn | he could be a little amazed with speed I do algos and languages haha | 00:08 |
@bettyboo | smart ;D | 00:08 |
@sonney2k | Heute habe ich endlich LTSA fertig implementiert und nun endlich auch deutsch gelernt. | 00:08 |
@sonney2k | Laplacian Eigenmaps ist nun das nächste auf meiner Lister. | 00:09 |
@sonney2k | Liste | 00:09 |
@sonney2k | blackburn, not exactly your words but mostly | 00:09 |
blackburn | :D | 00:09 |
blackburn | thanks | 00:09 |
blackburn | our discussion become boring, I need to joke a little :D | 00:10 |
blackburn | with chris I mean | 00:10 |
@sonney2k | argh | 00:12 |
@sonney2k | it is past midnight | 00:12 |
@sonney2k | I am getting hungry | 00:12 |
blackburn | why? | 00:13 |
@sonney2k | for some reason the power plug doesn't provide me with enough energy | 00:14 |
@sonney2k | and I tried... | 00:14 |
* sonney2k is on the way to the fridge | 00:15 | |
blackburn | hahaa google has indexed secret docs from gov.ru | 00:20 |
blackburn | oh in one doc it is said that 29609000 roubles was stealen with 30219000 available | 00:26 |
blackburn | more than 90% stealen, awesome :D | 00:26 |
@sonney2k | what a sign of competence ... | 00:28 |
blackburn | in fact most of russian people wants to leave russia heheh | 00:29 |
blackburn | and it makes no sense to any other country, I guess | 00:30 |
CIA-87 | shogun: Sergey Lisitsyn master * r5918132 / (5 files in 2 dirs): Introduced Local Tangent Space Alignment preprocessor - https://github.com/shogun-toolbox/shogun/commit/5918132d0383e371969afc9afb7593891bb08ba0 | 00:34 |
CIA-87 | shogun: Sergey Lisitsyn master * r16acbc2 / (2 files): SG_CALLOC usage for LLE and HLLE - https://github.com/shogun-toolbox/shogun/commit/16acbc2a648c7c522badba20e76849c0107e7ea0 | 00:34 |
blackburn | sonney2k: one algo to go ;) | 00:36 |
blackburn | http://dl.dropbox.com/u/10139213/shogun/ltsa-k12-2000.png | 00:36 |
@sonney2k | blackburn, your best figure so far... | 00:40 |
blackburn | no, HLLE is ~ | 00:40 |
blackburn | sonney2k: I guess it is the best a cat can get | 00:41 |
blackburn | :D | 00:41 |
blackburn | the only thing - normalization | 00:41 |
CIA-87 | shogun: Sergey Lisitsyn master * rb61710c / (2 files): Fixed HLLE python example and added LTSA one - https://github.com/shogun-toolbox/shogun/commit/b61710c72ab5550c7a6b859a1f93e85ff65fc3b4 | 01:03 |
blackburn | sonney2k: http://dl.dropbox.com/u/10139213/shogun/ltsa-k12-5000.png took 79s to compute | 01:15 |
@sonney2k | nice carpet | 01:15 |
blackburn | when people asks me what I do in GSoC I tell them I'm unrolling swiss roll | 01:16 |
blackburn | hey what's wrong with the weather? water in volga is +27C!! | 01:20 |
@sonney2k | ?! | 01:20 |
@sonney2k | here it is cold | 01:20 |
blackburn | today was +36 | 01:20 |
@sonney2k | serialhex, awake? | 01:21 |
blackburn | tomorrow will be +38 | 01:21 |
@sonney2k | but I guess at work... | 01:21 |
serialhex | yes, whats up? | 01:21 |
@sonney2k | serialhex, I am ruby wise nonfunctional | 01:21 |
@sonney2k | I've loaded modshogun | 01:21 |
serialhex | and it dosnt work? | 01:21 |
@sonney2k | but now how do I print any constant that is in this module? | 01:21 |
serialhex | it should be: | 01:22 |
serialhex | puts modshogun::Klass::c | 01:22 |
@sonney2k | for example I see | 01:22 |
@sonney2k | rb_define_const(mModshogun, "STRING_LEN", SWIG_From_int(static_cast< int >(256))) | 01:23 |
@sonney2k | in the modshogun_wrap.cxx | 01:23 |
@sonney2k | so I should be able to do puts modshogun::STRING_LEN ? | 01:23 |
serialhex | erm... i think so | 01:23 |
serialhex | i'm not too familiar with the C-side of ruby... hd on | 01:24 |
@sonney2k | yes that works Modshogun::MSG_DEBUG | 01:24 |
serialhex | ok cool | 01:25 |
@sonney2k | hurray | 01:25 |
@sonney2k | irb(main):015:0> gk=Modshogun::GaussianKernel.new() | 01:25 |
@sonney2k | => #<Modshogun::GaussianKernel:0x7f93977f67f0> | 01:25 |
@sonney2k | irb(main):016:0> puts gk.get_width() | 01:25 |
@sonney2k | 1.0 | 01:25 |
serialhex | for some reason i'm unable to load the module, when i try it in REPL session i get this error: https://gist.github.com/1105955 | 01:25 |
serialhex | SWEET!! | 01:25 |
@sonney2k | serialhex, I think you didn't do LD_LIBARY_PATH=pathtolibshogun.so | 01:26 |
serialhex | quite possibly, i'll try that later | 01:26 |
@sonney2k | serialhex, I can tell it works here now | 01:26 |
@sonney2k | serialhex, sploving wrote some typemaps already so vectors / string lists could potentially work | 01:27 |
serialhex | ok cool... so when did Modshogun pop up?? that namespace wasn't there before | 01:27 |
@sonney2k | he also did a bit of narray | 01:27 |
serialhex | yes, i've been talking to him a bit | 01:27 |
@sonney2k | serialhex, we merged all of shogun into one module -> modshogun | 01:27 |
serialhex | nice... good idea! | 01:28 |
@sonney2k | now we don't have naming conflicts | 01:28 |
@sonney2k | and it works under octave_modular too | 01:28 |
serialhex | and sploving was having problems with finding narray.h right? | 01:28 |
serialhex | i read that this am i think... it's been hardcoded IIRC | 01:29 |
serialhex | i'll figure it out & post it. NArray is a gem and thus shouldn't be too hard to parse out where it's hiding it's goods | 01:30 |
@sonney2k | serialhex, is there any way to figure out which methods are defined for an object? | 01:34 |
@sonney2k | for example I do: | 01:34 |
@sonney2k | require 'modshogun' | 01:34 |
@sonney2k | x = Modshogun::Labels.new | 01:35 |
@sonney2k | => #<Modshogun::Labels:0x7ff9481d1ca0> | 01:35 |
@sonney2k | y = [1, 3, 5, 7] | 01:35 |
@sonney2k | x.set_labels(y) | 01:35 |
@sonney2k | can I now figure out what methods x has? | 01:35 |
CIA-87 | shogun: Soeren Sonnenburg master * r91c9490 / (4 files): get the ruby examples to work again - https://github.com/shogun-toolbox/shogun/commit/91c9490d766564d8a82947f5b4cbc362319315b9 | 01:41 |
CIA-87 | shogun: Soeren Sonnenburg master * r8f5a9eb / (7 files in 3 dirs): get TRACE_MALLOCS to compile again (but still crashes for some reason) - https://github.com/shogun-toolbox/shogun/commit/8f5a9ebd403f0cedd97c21f95f385769ec532a32 | 01:41 |
CIA-87 | shogun: Soeren Sonnenburg master * r9bb26bc / (9 files in 3 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - https://github.com/shogun-toolbox/shogun/commit/9bb26bc315ab8d8645391df3ab87c2a3538985d2 | 01:41 |
@sonney2k | serialhex, yay vector and matrix work again | 01:41 |
@sonney2k | even narray vectors | 01:42 |
@sonney2k | serialhex, I didn't manage to get your example to run 100% though - I don't understand the syntax | 01:42 |
@sonney2k | it complains that sign is not defined on an Narray | 01:42 |
@sonney2k | blackburn, good n8 | 01:42 |
@sonney2k | I am tired... | 01:42 |
blackburn | see you | 01:43 |
blackburn | me too | 01:43 |
-!- blackburn [~blackburn@109.226.76.87] has quit [Quit: Leaving.] | 01:46 | |
serialhex | sonney2k: to inspect / play around with ruby i use a tool called pry - https://github.com/pry/pry you can ls & cd around a program like a shell & do a bunch of other things. you can also do things like x.methods (or x.methods(false) for singleton methods) and x.class | 02:41 |
serialhex | and i'm working on my example... pulling out the parts that aren't nessisary & such | 02:42 |
-!- f-x [~user@117.192.194.0] has joined #shogun | 04:25 | |
-!- gsomix [~gsomix@109.169.154.188] has joined #shogun | 05:44 | |
-!- f-x [~user@117.192.194.0] has quit [Ping timeout: 260 seconds] | 06:29 | |
@sonney2k | f-x_, if you want you can overhaul your new[] / resize memory stuff now | 09:31 |
@sonney2k | f-x_, I transitioned everything but I guess you can now use SG_REALLOC | 09:31 |
-!- sploving1 [~sploving@124.16.139.134] has joined #shogun | 09:57 | |
sploving1 | sonney2k, any thing? | 09:57 |
@sonney2k | sploving1, good morning :D | 09:57 |
@sonney2k | could you now run the ruby examples? | 09:58 |
sploving1 | yeap. except mini_svm | 09:58 |
sploving1 | others can run | 09:58 |
sploving1 | sonney2k, current upstream can work well? | 09:59 |
sploving1 | if that, I will pull it | 09:59 |
@sonney2k | yeah good - I fixed them yesterday - so it makes sense to now just port a few examples like you did for lua and do some testing | 09:59 |
@sonney2k | sploving1, yes | 09:59 |
@sonney2k | if your examples run then you have latest upstream I think | 09:59 |
sploving1 | I changed it by myself | 09:59 |
sploving1 | okay. I will pull it | 10:00 |
@sonney2k | I see - too bad then because I did the same thing too yesterday | 10:01 |
sploving1 | tonight we will have a meeting? 1:00? | 10:08 |
sploving1 | okay. | 10:08 |
-!- sploving1 [~sploving@124.16.139.134] has left #shogun [] | 10:08 | |
gsomix | Yeah. There are lists of applicants to my university. I'll see who will succeed me. :) | 10:09 |
@sonney2k | gsomix, ? | 10:13 |
gsomix | sonney2k, this is offtop. | 10:16 |
@sonney2k | gsomix, I couldn't parse it... | 10:17 |
gsomix | Hm. Lists wishing to study at university. Samara State Aerospace University. Their show today. | 10:19 |
@sonney2k | gsomix, but you are still there right? | 10:20 |
gsomix | I'll be a long time. | 10:26 |
@sonney2k | gsomix, which is good :) university time is one of the best in life :) | 10:26 |
gsomix | I mean they will alternate me on "the first year". | 10:26 |
@sonney2k | I don't understand - alternate you? | 10:27 |
gsomix | They will replace. Will be taught by people who taught me the first year. | 10:31 |
gsomix | Never mind. I continue to speak bad. | 10:31 |
gsomix | :) | 10:31 |
@sonney2k | gsomix, :) | 10:33 |
@sonney2k | gsomix, btw are you currently doing any shogun work / or want to do any? | 10:34 |
gsomix | Cover tree. Lisitsyn needs it. | 10:35 |
@sonney2k | gsomix, ok great :) | 10:35 |
-!- f-x [~user@117.192.218.222] has joined #shogun | 11:03 | |
@sonney2k | f-x, did I talk to your twin f-x_ ? | 11:21 |
@sonney2k | f-x, so did you see the lines above? | 11:21 |
@sonney2k | <sonney2k> f-x_, if you want you can overhaul your new[] / resize memory stuff now | 11:21 |
@sonney2k | <sonney2k> f-x_, I transitioned everything but I guess you can now use SG_REALLOC | 11:21 |
f-x | sonney2k: possibly - he doesn't talk much | 11:21 |
@sonney2k | ? | 11:21 |
@sonney2k | f-x, is that you manage to work 24h/ day ? | 11:21 |
f-x | hadn't seen those - i didn't check f-x_'s session | 11:21 |
f-x | i wish! | 11:21 |
@sonney2k | ok, so feel free to do those | 11:22 |
f-x | sonney2k: sure.. it's a better system now | 11:22 |
f-x | and btw | 11:22 |
f-x | vw is working now | 11:22 |
f-x | (but mostly using the default options) | 11:22 |
f-x | sonney2k: i will test it some more and remove some unneeded functions | 11:23 |
-!- blackburn [~blackburn@109.226.76.87] has joined #shogun | 11:52 | |
gsomix | <offtop> I looked over the lists. Entrants have a great scores for the exams. There will be a strong study group. On my major I mean. </offtop> | 11:53 |
-!- heiko [~heiko@134.91.54.253] has joined #shogun | 11:55 | |
-!- heiko [~heiko@134.91.54.253] has quit [Client Quit] | 11:55 | |
-!- heiko [~heiko@134.91.54.253] has joined #shogun | 11:55 | |
@sonney2k | f-x, very nice | 11:56 |
@sonney2k | f-x, I think this is the most intrusive change now still on the TODO | 11:57 |
@sonney2k | f-x, then it is probably only clean / speed ups | 11:57 |
@sonney2k | heiko, hi | 11:57 |
heiko | sonney2k, hi | 11:57 |
heiko | sorry that I wasnt there yesterday | 11:57 |
heiko | will work on saturday instead :) | 11:58 |
heiko | friends came visiting me spontaneously | 11:58 |
@sonney2k | heiko, I see - I just wanted to hear your opinion | 11:58 |
@sonney2k | because we did some intrusive change yesterday | 11:59 |
@sonney2k | (you were not there to vote for or against ) | 11:59 |
heiko | what did you do? | 11:59 |
@sonney2k | basically we removed all occurences of new[] / delete[] | 11:59 |
@sonney2k | and replaced it with SG_MALLOC / SG_REALLOC | 11:59 |
@sonney2k | SG_CALLOC | 11:59 |
@sonney2k | SG_FREE | 11:59 |
heiko | didnt you do the opposite recently? | 12:00 |
@sonney2k | heiko, ? no | 12:00 |
heiko | mmh | 12:00 |
heiko | ok and why did you do it? | 12:00 |
heiko | to catch errors and stuff? | 12:00 |
@sonney2k | it has the big advantage that we can trace mallocs later easily | 12:00 |
* heiko checks out current git | 12:00 | |
heiko | yes, this is really good | 12:01 |
@sonney2k | and that we have no interoperability issues with other libraries | 12:01 |
@sonney2k | other C-libs always use malloc/free so we could never pass a ptr of a shogun object | 12:01 |
@sonney2k | object as in double* matrix | 12:02 |
heiko | ok | 12:02 |
heiko | ok then, why should I be against it?= | 12:02 |
@sonney2k | heiko, it is just better when it is a community decision | 12:03 |
@sonney2k | not just me doing random seemign changes | 12:03 |
@sonney2k | seeming | 12:03 |
heiko | ok, well then I think this is good | 12:03 |
heiko | tracing is always usefull | 12:03 |
heiko | I still have the SGObject trace stuff on my TODO list | 12:03 |
heiko | have a list of all allocated SGObjects would be cool | 12:04 |
@sonney2k | heiko, it is there (but not working aka crashing) when you pass to configure --enable-trace-mallocs | 12:04 |
@sonney2k | heiko, the cool thing is that we can even make it like list all SG_MALLOC allocated things + the line nr where these where allocated | 12:05 |
@sonney2k | heiko, for SGObjects I dont' see that it will work | 12:05 |
heiko | yes, very good | 12:05 |
heiko | ok but it would be good to have a list | 12:05 |
heiko | like a DynArray of all allocated SGObjects | 12:06 |
heiko | then you can check if the list is empty if valgrind reports leaks | 12:06 |
@sonney2k | heiko, I think this would only work if we decide to introduce SG_NEW and SG_DELETE macros | 12:06 |
heiko | and you can ask for the names of the objects to get a clue where the leak is | 12:06 |
heiko | for SGObject these macros? | 12:06 |
@sonney2k | with such macro you could even get the file name + line nr where the memory got allocated | 12:07 |
heiko | would be very useful | 12:07 |
@sonney2k | heiko, for any C++ object the new() operator... | 12:07 |
heiko | when I wrote this tree set multiplication stuff, I really wished I had something like this | 12:07 |
@sonney2k | well heiko we could do this big change too | 12:11 |
@sonney2k | I guess we should have a vote then again with blackburn and f-x ... | 12:12 |
heiko | ok, sounds good | 12:13 |
@sonney2k | this would enable us only better debugging - right? | 12:13 |
heiko | yes | 12:14 |
@sonney2k | I mean we could display a table with sgobjects | 12:14 |
@sonney2k | ther ref-counts | 12:14 |
heiko | yes exactly | 12:14 |
@sonney2k | where these where allocated | 12:14 |
heiko | and names | 12:14 |
@sonney2k | and names | 12:14 |
heiko | and it would be great to have access to the adresses of the objects | 12:14 |
@sonney2k | and it would avoid global new operator overloading | 12:14 |
@sonney2k | which is also good | 12:14 |
heiko | true | 12:15 |
@sonney2k | heiko, the only disadvantage is that there might be 1000-2000 occurances of new() | 12:20 |
-!- f-x [~user@117.192.218.222] has quit [Ping timeout: 260 seconds] | 12:20 | |
heiko | yes, mmh, a lot of work | 12:20 |
heiko | perhaps we do this a little bit later to have some more time for the hot stuff this week? | 12:21 |
@sonney2k | heiko, new[] occurred ~1200 times | 12:21 |
heiko | sonney2k, did I mix up my fork again? or is the pull-request possible? | 12:21 |
@sonney2k | no problem with some sed magic | 12:21 |
heiko | I see patches from you | 12:21 |
@sonney2k | ? | 12:21 |
heiko | my current pull request | 12:21 |
heiko | is it possible to merge it? | 12:21 |
@sonney2k | no | 12:21 |
heiko | just made a puosh | 12:22 |
heiko | push | 12:22 |
@sonney2k | we didn't talk about it so I didn't merge | 12:22 |
heiko | I had some trouble with git, and I think i made a mistake with all the rebasing | 12:23 |
heiko | well it introduces support for multiple data types in model selection | 12:23 |
heiko | sonney2k, is it technically possible to merge it? | 12:36 |
heiko | because if not i have to reset my fork again, which is kind of annoying | 12:36 |
heiko | (I have to check in all the changes again) | 12:36 |
@sonney2k | you just have to go to your master branch | 12:37 |
@sonney2k | then do git pull there | 12:37 |
@sonney2k | then go to your feature branch | 12:37 |
@sonney2k | and then do git rebase master | 12:37 |
@sonney2k | if there are conflicts - fix them | 12:37 |
@sonney2k | and then go to master branch and git merge feature | 12:37 |
@sonney2k | and git push | 12:37 |
-!- f-x [~user@117.192.210.213] has joined #shogun | 12:45 | |
heiko | yes, I know, but this is not working, because of some merge problems :( | 12:50 |
heiko | tried to solve it, but it does not work | 12:50 |
heiko | and my commit list is growing ... | 12:51 |
heiko | will reset fork | 12:51 |
blackburn | bad way | 12:53 |
heiko | yes | 12:53 |
heiko | but I have no idea how to solve this | 12:54 |
@sonney2k | heiko, OK wait let me try to do it from your request | 12:58 |
heiko | no | 12:59 |
heiko | i think its even worse now | 12:59 |
heiko | but i have it correct locally | 12:59 |
heiko | but cannot push | 12:59 |
heiko | sais i have to pull before | 12:59 |
heiko | but a pull destory my local changes, it does not ask to merge :( | 12:59 |
@sonney2k | the pull worked without problems | 13:00 |
@sonney2k | pushing | 13:00 |
heiko | think i got it now, the pull request is clean(er) now | 13:00 |
CIA-87 | shogun: Soeren Sonnenburg master * r440c5ef / (10 files in 4 dirs): Merge branch 'master' of git://github.com/karlnapf/shogun (+19 more commits...) - https://github.com/shogun-toolbox/shogun/commit/440c5efb2787654db9db583a44514cef20d8fc81 | 13:00 |
heiko | i just used push -f to overwrite changes in my fork | 13:01 |
heiko | argh -- | 13:01 |
heiko | a HEAD is in the diff | 13:01 |
@sonney2k | https://github.com/shogun-toolbox/shogun/pull/233/files ? | 13:01 |
blackburn | :D | 13:02 |
@bettyboo | he blackburn | 13:02 |
blackburn | lol | 13:02 |
heiko | git push | 13:02 |
heiko | ah sorry | 13:02 |
@sonney2k | heiko, ?! | 13:02 |
@sonney2k | is master correct now or not | 13:03 |
@sonney2k | ? | 13:03 |
heiko | mmmh | 13:03 |
heiko | I will check | 13:03 |
@sonney2k | to me it seems it included your original patch | 13:03 |
@sonney2k | if it does I would suggest to do some git reset --hard <lastworkingversion> | 13:03 |
heiko | yes | 13:04 |
heiko | shogun-toolbox is ok now | 13:04 |
@sonney2k | and push -f to your branch and then just merge from master | 13:04 |
@sonney2k | heiko, master is not compiling though | 13:04 |
heiko | oh no :( | 13:04 |
heiko | i will pull and fix | 13:05 |
@sonney2k | I don't really get what the problem is thoguh | 13:07 |
@sonney2k | ohh I see | 13:08 |
heiko | perhaps current pull request solves? | 13:10 |
heiko | only one line change but many commits | 13:10 |
heiko | if not i will do a hard reset | 13:10 |
heiko | libshogun compiles here btw | 13:10 |
blackburn | thogun | 13:10 |
@sonney2k | yeah but not the rest | 13:10 |
@sonney2k | heiko, https://github.com/shogun-toolbox/shogun/pull/241/files ? | 13:11 |
heiko | canceled | 13:12 |
heiko | i will fix compile errors now | 13:12 |
heiko | sorry for the trouble, git makes me mad | 13:13 |
@sonney2k | it compiles now | 13:13 |
@sonney2k | heiko, relax | 13:13 |
CIA-87 | shogun: Soeren Sonnenburg master * r9df895b / src/shogun/lib/DataType.h : fix mistake in string data type - https://github.com/shogun-toolbox/shogun/commit/9df895b9b474c21f6d2486c547a3f5262e8316d5 | 13:14 |
@sonney2k | heiko, you should check though if you have used any new[] or delete[] | 13:14 |
@sonney2k | ^ in your last patch I mean | 13:14 |
* sonney2k cries | 13:14 | |
heiko | i did, will change it | 13:14 |
heiko | whats happening? | 13:15 |
@sonney2k | my 16GB memory won't work in this mainboard | 13:15 |
@sonney2k | it just beeps :((( | 13:15 |
heiko | oh no :( | 13:15 |
@sonney2k | bye bye build bot | 13:15 |
@bettyboo | sonney2k, sleep well | 13:15 |
@sonney2k | heiko, anyway everything built ok now... | 13:18 |
@sonney2k | so please just adjust your fork to it | 13:18 |
@sonney2k | heiko, btw the only intrusive change now is this subset business right? | 13:18 |
@sonney2k | then it is basically all working or? | 13:19 |
* heiko will adjust his fork | 13:19 | |
heiko | yes, basically | 13:19 |
heiko | and model storage for other machines than kernel machine | 13:19 |
@sonney2k | heiko, there is only kernel / distance machine | 13:22 |
@sonney2k | both work the same way so it is copy & paste :) | 13:22 |
heiko | sonney2k, I thought about just adding this stuff to machine | 13:23 |
heiko | or dont you think that is a good idea? | 13:24 |
@sonney2k | heiko, which stuff? | 13:24 |
heiko | the split train/train_kernel_machine and the setter to tell a machine to store its model | 13:25 |
@sonney2k | but a machine has no idea about a kernel... | 13:25 |
heiko | true, i mean the name could change, | 13:26 |
heiko | I meant the call of store_model that is done automatically if the flag is set | 13:26 |
-!- gsomix [~gsomix@109.169.154.188] has quit [Quit: Ухожу я от вас (xchat 2.4.5 или старше)] | 13:35 | |
-!- f-x [~user@117.192.210.213] has quit [Ping timeout: 260 seconds] | 13:38 | |
-!- f-x [~user@117.192.219.5] has joined #shogun | 13:40 | |
@sonney2k | heiko, sorry I got distracted ordering different memory again... | 13:49 |
heiko | np | 13:50 |
@sonney2k | heiko, I think you need that do you? | 13:50 |
@sonney2k | I mean how could you tell the machine to store it in memory otherwise? | 13:51 |
heiko | ehm what? | 13:51 |
heiko | ah ok | 13:51 |
heiko | no i meant | 13:51 |
heiko | should the possibility to tell a machine to store its model be part of CMachine instead of CKernelMachine? | 13:51 |
@sonney2k | I mean otherwise you have to cast to kernelmachine | 13:51 |
@sonney2k | etc | 13:51 |
@sonney2k | so you need that option in machine for you model selection business | 13:52 |
heiko | well yes, that is another problem | 13:52 |
heiko | the cross-validation class only sees a CMachine | 13:52 |
heiko | I added the flag in the example when the SVM was created | 13:52 |
heiko | but that is no-good | 13:52 |
heiko | ok, I will change this then. | 13:52 |
blackburn | sonney2k: I didn't kernel PCA ;) | 13:53 |
@sonney2k | blackburn, but you will do :) | 13:53 |
blackburn | yeah, now I have to | 13:53 |
@sonney2k | heiko, btw | 13:54 |
@sonney2k | $ grep 'new .*(' shogun -r | wc -l | 13:54 |
@sonney2k | 1023 | 13:54 |
@sonney2k | blackburn, btw what do you think about doing the same thing we did for new[] for new() ? | 13:55 |
blackburn | e.g? | 13:55 |
@sonney2k | new -> SG_NEW | 13:56 |
blackburn | SG_NEW(CFibonacciHeap)? | 13:56 |
@sonney2k | with the argument that we can enable debugging etc | 13:56 |
blackburn | how to pass params? | 13:56 |
heiko | sonney2k, do you think this works straight forward with sed? | 13:56 |
@sonney2k | new KernelMachine(17) -> SG_NEW(KernelMachine(17)) | 13:57 |
@sonney2k | or alternatively | 13:57 |
@sonney2k | SG_NEW KernelMachine(17); | 13:57 |
blackburn | SG_NEW CLASS looks more attractive for me | 13:58 |
@sonney2k | hmhh not sure if that will work sufficiently well | 13:58 |
@sonney2k | heiko, I see a couple of problems currently | 13:58 |
@sonney2k | when we do SG_NEW CKernel() | 13:58 |
@sonney2k | we don't have access to the parameters | 13:59 |
@sonney2k | so we cannot alloc the object and store meta data about it | 13:59 |
@sonney2k | so we would need SG_NEW(CKernel(17,3,4,5)) | 14:00 |
@sonney2k | but then the macro parser will think that it gets 4 arguments instead of 1 | 14:00 |
CIA-87 | shogun: Sergey Lisitsyn master * r34dfe53 / doc/pages/FAQ.mainpage : Removed note about octave crash from FAQ - https://github.com/shogun-toolbox/shogun/commit/34dfe53d8f3ec7eebbfd8cf6cd84aa5317a3ecc2 | 14:01 |
blackburn | hehe | 14:01 |
blackburn | sonney2k: I can't start translating doc to ru - it is OLD! | 14:01 |
@sonney2k | blackburn, ohh so you will have to update the old doc too :D | 14:01 |
@bettyboo | ^_^ | 14:01 |
blackburn | sonney2k: chinese too? | 14:02 |
@sonney2k | of course | 14:02 |
blackburn | :D | 14:02 |
blackburn | I guess we should add some questions to FAQ | 14:03 |
heiko | sonney2k, mmh, for debugging the parameter trees, I simply added something to the CSGObject ocnstructor | 14:04 |
heiko | but that then only works for shogun objects | 14:04 |
@sonney2k | blackburn, no idea which ones though | 14:05 |
@sonney2k | heiko, I mean one more hack is to overload new() and delete() | 14:05 |
@sonney2k | (we do this already) | 14:05 |
heiko | mmh | 14:05 |
@sonney2k | and then try a dynamic cast to CSGObject | 14:06 |
heiko | anything against this? | 14:06 |
heiko | oh | 14:06 |
@sonney2k | it is a global thing | 14:06 |
@sonney2k | so bad when someone else relies on that too | 14:06 |
heiko | true | 14:06 |
-!- John_Langford [6c152263@gateway/web/freenode/ip.108.21.34.99] has joined #shogun | 14:06 | |
blackburn | wow | 14:06 |
heiko | but, mmh, also an advantage that you dont have to change your programming habits | 14:06 |
@sonney2k | heiko, macros won't work for nicely with comma's | 14:08 |
@sonney2k | http://ingomueller.net/node/1203 | 14:08 |
@sonney2k | horrible... | 14:08 |
heiko | indeed | 14:09 |
heiko | so then the new operator | 14:09 |
heiko | but instead of a dynamic cast i would prefer adding something to the constructors | 14:09 |
@sonney2k | heiko, the trace malloc business is already there | 14:09 |
-!- sploving1 [~sploving@210.77.14.219] has joined #shogun | 14:09 | |
@sonney2k | but for some reason it is no longer working (crashing) | 14:09 |
sploving1 | serialhex, are you around? | 14:10 |
serialhex | yes | 14:10 |
sploving1 | do you know ruby or narray has function that read numbers from a file? | 14:10 |
sploving1 | like fromfile/ loadtxt in numpy? | 14:10 |
sploving1 | I need write a load.rb, | 14:11 |
serialhex | is each number on a different line? | 14:11 |
@sonney2k | heiko, I suspect that there is an unwanted recursion going on... | 14:11 |
@sonney2k | heiko, like to trace the malloc, we need another malloc which needs another or so | 14:12 |
sploving1 | no, every line have two numbers, serialhex | 14:12 |
@sonney2k | serialhex, space separated | 14:12 |
-!- heiko [~heiko@134.91.54.253] has quit [Read error: Connection reset by peer] | 14:12 | |
sploving1 | I use IO.readlines, it return "1 3\n" | 14:12 |
serialhex | so it'd be a 2D matrix? is that how you want it? | 14:12 |
-!- heiko [~heiko@134.91.54.253] has joined #shogun | 14:13 | |
sploving1 | yeap. for instance, 1 3 | 14:13 |
serialhex | so that would return [[1], [3]] | 14:13 |
serialhex | ok... | 14:13 |
sploving1 | serialhex, I want every line is a row of the matrix | 14:16 |
sploving1 | and the columes is 2 | 14:16 |
* heiko gets crazy with git | 14:16 | |
@sonney2k | heiko, so the downside of 'just' overloading new | 14:17 |
@sonney2k | is that we cannot say '*in which line*' the object got allocated | 14:17 |
@sonney2k | but we can print refcount and ptr | 14:17 |
@sonney2k | and name | 14:17 |
heiko | well, ok | 14:17 |
sploving1 | for instance, 1 3, the seconde line is 5.3 7, the third line is 8.8 6.6, then I want to return: [[1, 3], [5.3, 7], [8.8, 6.6]] | 14:18 |
@sonney2k | sploving1, btw orico did some example translation - draft - for c# here https://github.com/shogun-toolbox/shogun/pull/238/files | 14:21 |
@sonney2k | maybe you can use this as a baseline later | 14:21 |
CIA-87 | shogun: Sergey Lisitsyn master * rec69179 / examples/undocumented/python_modular/graphical/dimreduct/swissroll.py : Added HLLE and LTSA to swissroll graphical example - https://github.com/shogun-toolbox/shogun/commit/ec69179210b7d8f4587de129f3ebb726daeeaf01 | 14:21 |
-!- gsomix [~gsomix@109.169.154.188] has joined #shogun | 14:22 | |
sploving1 | I know that. but they use jblas, not working now. I have no idea about C# libriary, that similiar to C# | 14:23 |
sploving1 | similiar to jblas. | 14:23 |
heiko | sonney2k, ok sounds good | 14:26 |
@sonney2k | sploving1, I think we have no option but to just use nested arrays . so use their files as inspirations | 14:27 |
serialhex | sploving1: i've got to get ready for work so i'll try and debug this later, butit's prettymuch along these lines: https://gist.github.com/1109251 | 14:27 |
serialhex | so far it dosn't return floats, just strings... i'll see what ... hd on | 14:28 |
sploving1 | serialhex, thx. I will have a try | 14:28 |
serialhex | sploving1: ok i figured out a (non-pretty) way to do it... i'll see if i can find something better later but i updated the gist and it works now | 14:29 |
serialhex | if you like you can change the n.to_f to n.to_i for integers or whatever you want... maybe make a wrapper to catch each case or something | 14:30 |
sploving1 | serialhex, you give me is one row matrix | 14:31 |
sploving1 | not 2D matrix? | 14:31 |
sploving1 | updated? I saw that | 14:32 |
serialhex | yeah, it's 2d... or should be, it will mimmic the layout of the file | 14:37 |
-!- vojta [~quassel@2001:718:2:1634:221:6aff:fe58:151a] has joined #shogun | 14:38 | |
-!- vojta is now known as VojtechFranc | 14:38 | |
serialhex | i updated it again with my input file & my output... | 14:39 |
sploving1 | great serialhex, thx. I will try | 14:40 |
sploving1 | serialhex, do you know how to fix the bug in the example | 14:41 |
sploving1 | classifier_libsvm_minimal_modular | 14:41 |
sploving1 | .rb | 14:41 |
sploving1 | what does class_eval mean in ruby? | 14:42 |
serialhex | class _eval isa metaprogramming thing, it lets you open up the class and add new methods or whatever. | 14:43 |
sploving1 | serialhex, it said undefined method `sign' for #<NArray:0xb5538210> (NoMethodError) | 14:43 |
sploving1 | then you mean we can define a method sign for NArray? | 14:44 |
serialhex | yes we can! | 14:44 |
sploving1 | narray has a method mean, then we can remove it from the example! am I right? serialhex? | 14:46 |
serialhex | yep | 14:49 |
serialhex | and you can probably just put an "N" before Array.class_eval and everything should work | 14:49 |
sploving1 | okay. | 14:51 |
serialhex | sploving1: it should work if you make it like this: https://gist.github.com/1109300 | 14:52 |
@sonney2k | *lol* http://www.memo.tv/istevejobs | 14:56 |
-!- cwidmer [~quassel@connect.tuebingen.mpg.de] has joined #shogun | 14:56 | |
@sonney2k | shall we start? | 15:03 |
heiko | ok for me | 15:03 |
cwidmer | good to go | 15:04 |
-!- mikiobraun [~mikio@brln-4db90002.pool.mediaWays.net] has joined #shogun | 15:04 | |
@sonney2k | alright then | 15:04 |
@sonney2k | letst start | 15:04 |
mikiobraun | hello | 15:04 |
sploving1 | hello | 15:04 |
sploving1 | okay. start... | 15:04 |
@sonney2k | so hello everyone to thi s(now for real) last month of GSoC | 15:04 |
@sonney2k | we have slightly more than 2 weeks until google recommends pencils down so this is the final sprint for features: do the most drastic changes now, i.e. this and next week - afterwards only additions please... | 15:05 |
@sonney2k | to notify you about some bigger changes we did internally | 15:06 |
@sonney2k | ewer again use new[] / delete[] but instead SG_MALLOC/SG_CALLOC/SG_REALLOC/SG_FREE - this will enable us to interoperate more easily with libraries and also to trace memory allocations | 15:06 |
@sonney2k | we also did get rid of the multi-module swig interfaces but link everything into one big file 'modshogun' now - this resolved all the issues we had with the interfaces failing more or less randomly | 15:07 |
@sonney2k | apart from that everyone should really focus on doing the intrusive things until end of next week | 15:08 |
@sonney2k | additions are OK | 15:08 |
@sonney2k | later | 15:08 |
@sonney2k | and polishing too of course | 15:08 |
@sonney2k | does anyone have any concerns regarding this? | 15:08 |
@sonney2k | btw VojtechFranc where is alesis? | 15:08 |
cwidmer | just to clarify, so does this mean no more submodules, like shogun.Features etc? | 15:09 |
VojtechFranc | no, idea I sent him email two days ago but no reply yet | 15:09 |
@sonney2k | cwidmer, yes | 15:09 |
cwidmer | finally | 15:09 |
cwidmer | :) | 15:09 |
@sonney2k | cwidmer, it would be from modshogun import xxx | 15:09 |
cwidmer | that has caused me quite some pain in the past | 15:09 |
blackburn | sonney2k: why "from shogun.Preprocessor import LandmarkMDS" works? :D | 15:09 |
@sonney2k | however to not screw users I added some compat wrappers emulating the old behavior | 15:09 |
@bettyboo | ho? | 15:09 |
blackburn | ah | 15:10 |
@sonney2k | ^ | 15:10 |
blackburn | okay | 15:10 |
cwidmer | sonney2k, way to go | 15:10 |
blackburn | I have a question about my preprocessors | 15:10 |
@sonney2k | blackburn, shoot. | 15:10 |
blackburn | most of them doesn't support any out-of-sample extension for applying to some feature vector | 15:10 |
blackburn | e.g. it is not possible with classic mds | 15:10 |
blackburn | it is possible with landmark mds but there is a need to store landmarks | 15:11 |
blackburn | now all of my preprocessors have SG_NOTIMPLEMENTED for apply_to_feature_vector | 15:12 |
blackburn | I guess it is bad for release, right? | 15:12 |
@sonney2k | blackburn, well just add a more meaningful SG_ERROR msg | 15:13 |
@sonney2k | OK? | 15:13 |
@sonney2k | anyone else? | 15:13 |
blackburn | so it is okay to be gay^W^W just write it is not supported? | 15:13 |
@sonney2k | sure it is ok to say that this particular function is not implemented because it is not possible or would be inefficient or... | 15:14 |
@sonney2k | just hint people what they should do | 15:14 |
blackburn | everyone fell asleep | 15:14 |
heiko | nono :) | 15:14 |
cwidmer | sonney2k, ok with me | 15:15 |
serialhex | ZZzzzzzzZZZZZzzzzzzzZZZZZzzzzzzzz.... | 15:15 |
cwidmer | heh | 15:15 |
@sonney2k | to continue: so to be able to have a release end of august we need to do bug hunts / cleanups and we need everyones help. | 15:15 |
@sonney2k | so for example clean up tasks could be | 15:15 |
@sonney2k | everyone compile with doxygen enabled and fix the documentation! | 15:15 |
@sonney2k | doxygen will issue warnings | 15:16 |
@sonney2k | and these should *all* be gone before the release | 15:16 |
@sonney2k | blackburn - would be great to have the russian documentation ready before release :) | 15:16 |
blackburn | sonney2k: just after we fix english one | 15:16 |
@sonney2k | then please everyone check the NEWS file and add new features (I did add a few items - please check for more) | 15:16 |
@sonney2k | then we will need some help to port examples to all supported languages | 15:16 |
@sonney2k | and of course if you encounter compile warnings - fix them | 15:17 |
@sonney2k | BTW, I saw some inefficient code in SGVector allocation appearing more often than just once | 15:18 |
heiko | (we could treat warnings as errors ) | 15:18 |
@sonney2k | SGVector can be used like SGVector(length), SGMatrix(rows, cols) - will alloc vector / matrix. They need to be manually destroyed (matrix.destroy()) | 15:18 |
@sonney2k | heiko, every compiler has a different set of warnings - so it is dangerous to do this on the long run | 15:18 |
heiko | now meant just for one time | 15:18 |
@sonney2k | heiko, once we are down to 0 - it makes sense | 15:19 |
@sonney2k | currently there are jsut too many and then no one would be able to do work | 15:19 |
heiko | ok :) | 15:19 |
@sonney2k | one more thing: it would be great if we internally (in object member variables) use SGVector/ SGMatrix etc whenever sth. needs to be serialized. | 15:20 |
@sonney2k | and also return/pass SGVector etc only not the double*, len thing we did before (if possible) | 15:20 |
@sonney2k | I know everyone is busy for the next 1.5 weeks but when you encounter such issues please attempt to fix them | 15:21 |
@sonney2k | and of course - if you can test things, be it installing shogun on a mac / cygwin / different linux distributions with whatever language you are interested in - please do | 15:22 |
@sonney2k | I suspect that we will have many small issues... | 15:23 |
@sonney2k | so most important, test, test, test! | 15:23 |
@sonney2k | That's it from my side. Any comments, questions, concerns? | 15:23 |
heiko | not from me | 15:24 |
heiko | oh | 15:24 |
heiko | the CSGObject trace | 15:24 |
@sonney2k | yes? | 15:24 |
heiko | I will try to add it in the next week | 15:24 |
heiko | so if anyone has debug problems: soon there will be a list of all allocated CSGObjects | 15:25 |
@sonney2k | heiko, I think I should better do this - and you concentrate on the other 2 intrusive changes you need to do | 15:25 |
blackburn | this week I'll make a try getting all the java examples to work | 15:25 |
heiko | or this way | 15:25 |
heiko | just wanted to point out that this stuff will be there soon, because its useful | 15:25 |
@sonney2k | heiko, it used to work already maybe a year back or so :) | 15:25 |
blackburn | good thing about java_modular - I might add clojure and scala examples | 15:26 |
mikiobraun | You should know that I'll be on vacation soon, so any burning Java questions should be asked before Friday. | 15:26 |
blackburn | well I hadn't any ;) | 15:26 |
@sonney2k | blackburn, we have enough things to polish right now - and the plan is to release end of august | 15:26 |
mikiobraun | very good ;) | 15:27 |
@sonney2k | so better postpone these issues and do some tests/polishing | 15:27 |
blackburn | sonney2k: okay | 15:27 |
blackburn | it could be easy because both clojure and scala utilizes java things | 15:28 |
blackburn | so we don't have to do another typemaps | 15:28 |
@sonney2k | alright, I think everyone knows what to do? | 15:28 |
heiko | yes :) | 15:28 |
@sonney2k | then shall we meet again say August 10, 13:00 UTC? | 15:29 |
mikiobraun | blackburn: You can just use Java libraries from clojure/java, one could only think about some nice syntactic sugar | 15:29 |
mikiobraun | I'll not be there on Aug 10 | 15:29 |
@sonney2k | mikiobraun, when are you back | 15:29 |
@sonney2k | ? | 15:29 |
blackburn | mikiobraun: yes it is exactly what I'm talking about | 15:29 |
@sonney2k | well Aug16 is suggested pencils down date | 15:30 |
sploving1 | mikiobraun, enjoy your holiday. now I am working ruby, and next week I will try to support C# | 15:30 |
@bettyboo | ;> | 15:30 |
sploving1 | then the last week may fix some bugs | 15:30 |
mikiobraun | Well, at least it shouldn't be public ;) | 15:30 |
@bettyboo | :) | 15:30 |
@sonney2k | so better lets meet on Aug 10 if possible | 15:30 |
@sonney2k | is the date ok for the others? | 15:31 |
VojtechFranc | yes, ok with me | 15:31 |
f-x | it's good | 15:31 |
heiko | for me ok | 15:31 |
blackburn | oakya | 15:31 |
@sonney2k | John_Langford, VojtechFranc cwidmer f-x heiko blackburn serialhex ? | 15:31 |
cwidmer | k | 15:31 |
@sonney2k | ok | 15:31 |
sploving1 | k | 15:31 |
serialhex | sounds good... i'm gonna do what i can to help the ruby stuff along | 15:31 |
@sonney2k | serialhex, great! | 15:32 |
@sonney2k | then thanks everone for attending and please keep up the good work | 15:32 |
cwidmer | great, thanks sören | 15:32 |
@sonney2k | keep the patches coming and on Aug 10 we will discuss about who polishes what :) | 15:32 |
@sonney2k | ahh stop | 15:33 |
@sonney2k | any of the mentors wanting to go to the mentors event? | 15:33 |
@sonney2k | in october 21-23? | 15:33 |
@sonney2k | at google hq? | 15:33 |
@sonney2k | mikiobraun, cwidmer, John_Langford , VojtechFranc ? | 15:33 |
cwidmer | I'd be interested to go | 15:34 |
VojtechFranc | me not, sorry | 15:34 |
mikiobraun | Would be interesting, but I don't think I can go. | 15:34 |
blackburn | *it seems chris likes to travel ;) | 15:34 |
heiko | is this paid by google? | 15:34 |
@sonney2k | mostly yes | 15:34 |
@sonney2k | but it is for mentors only | 15:34 |
cwidmer | especially to california ;) | 15:35 |
heiko | sad thing :) | 15:35 |
@sonney2k | if any of you students want to be mentors next year - tell us :) | 15:35 |
sploving1 | beatiful place | 15:35 |
blackburn | I want to be student next year hehe | 15:35 |
heiko | would be cool :) | 15:35 |
heiko | but i am in london busy studying next year | 15:36 |
@sonney2k | alright that really is all from my side | 15:36 |
heiko | ok, see you all! | 15:36 |
cwidmer | ok, bye all! | 15:36 |
VojtechFranc | ok, cheers | 15:36 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 15:38 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 15:39 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!