--- Log opened Sun Jun 19 00:00:27 2011 | ||
--- Day changed Sun Jun 19 2011 | ||
@sonney2k | heh | 00:00 |
---|---|---|
blackburn1 | sonney2k: pull-requested | 00:01 |
CIA-32 | shogun: Sergey Lisitsyn master * r80f768e / (6 files in 2 dirs): Fixed C{ROC,PRC}Evaluation classes - http://bit.ly/l4GC27 | 00:01 |
CIA-32 | shogun: Soeren Sonnenburg master * rdfe5bdf / (6 files in 2 dirs): Merge branch 'master' of git://github.com/SergeyLisitsyn/shogun - http://bit.ly/jUAylV | 00:01 |
blackburn1 | answered baozeng | 00:03 |
blackburn1 | sonney2k: guys at scikits-learn doing manifold learning just like me | 00:09 |
blackburn1 | competition! :D | 00:10 |
@sonney2k | blackburn1, just do the 30+ methods from the link we found somewhere | 00:11 |
@sonney2k | then the definitive answer is given :) | 00:11 |
-!- f-x [~user@117.192.200.151] has quit [Quit: ERC Version 5.3 (IRC client for Emacs)] | 00:12 | |
blackburn1 | sure but now I'm focused on efficiency | 00:13 |
@sonney2k | blackburn1, you get your arpack | 00:14 |
@sonney2k | that should be very easy to do... | 00:14 |
blackburn1 | sonney2k: i'm thinking about more efficient routines | 00:14 |
blackburn1 | http://en.wikipedia.org/wiki/LOBPCG | 00:16 |
blackburn1 | for example | 00:16 |
@sonney2k | as you wish - simple and working things first - then speed and mem efficient stuff | 00:18 |
blackburn1 | I'm doing it in pace of my plan :) | 00:18 |
blackburn1 | hahaha | 00:19 |
blackburn1 | I lost link with 30+ methods | 00:19 |
blackburn1 | found in logs | 00:20 |
@sonney2k | btw scipy has this method implemented | 00:21 |
@sonney2k | so you can prolly even borrow the code | 00:21 |
blackburn1 | sonney2k: which? LOBPCG? | 00:22 |
@sonney2k | yes | 00:22 |
blackburn1 | yes, we could | 00:23 |
blackburn1 | wonder why dijkstra is used in LLE | 00:25 |
blackburn1 | in all implementations I have seen | 00:25 |
blackburn1 | for n vertices it is O(n^3) just as floyd-warshall | 00:26 |
blackburn1 | hahha this toolbox is not ideal | 00:28 |
blackburn1 | % NOTE: Classical scaling is identical to performing PCA, except the | 00:28 |
blackburn1 | % input data is different. Specifying pairwise similarity data is not | 00:28 |
blackburn1 | % yet supported by the toolbox. | 00:28 |
blackburn1 | mappedX = compute_mapping(X, 'PCA', no_dims); | 00:28 |
@sonney2k | 1:0 for russia | 00:29 |
blackburn1 | btw LLE is pretty fast now | 00:32 |
blackburn1 | cause it is optimized a little and cholesky solver is used | 00:33 |
blackburn1 | sonney2k: one 1rd year student is interested in shoguning a little :) | 00:37 |
@bettyboo | blackburn1, ;D | 00:37 |
blackburn1 | 1rd? I wrote 1rd? | 00:37 |
blackburn1 | omg I should sleep | 00:37 |
@sonney2k | ird? | 00:38 |
@sonney2k | 3rd? | 00:38 |
@sonney2k | or 1st? | 00:38 |
@sonney2k | blackburn1, do what you can :) | 00:38 |
blackburn1 | ahaha | 00:38 |
blackburn1 | 1st yes | 00:38 |
@sonney2k | I will have to sleep now too | 00:38 |
blackburn1 | good night :) | 00:39 |
@sonney2k | cu | 00:39 |
-!- blackburn1 [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 00:48 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 10:07 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 10:30 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 10:30 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 10:37 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 10:37 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 11:01 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 11:01 | |
blackburn | sonney2k: oh I have started refactoring preprocessor for SGMatrix usage.. sometimes code here sucks | 12:01 |
blackburn | пиздец.. | 12:18 |
blackburn | I have to refactor simple features too.. | 12:18 |
@sonney2k | blackburn, yeah many functions now need refactoring - basically all that get a float64_t* x, int32_t len argument and ** x ... | 12:29 |
blackburn | sonney2k: I'll make some temporary hacks now | 12:29 |
blackburn | because now I'm focused on Preprocessor refactoring :) | 12:29 |
blackburn | sonney2k: would you fix it a bit later? | 12:29 |
blackburn | now I'm going to fix all the preprocessors.. this will take a time | 12:31 |
@sonney2k | thanks | 12:31 |
blackburn | sonney2k: also I'm going to introduce new class - DimReductionPreprocessor | 12:32 |
blackburn | it contains target dimension - all of my algorithms works with it | 12:32 |
blackburn | it could re-dup some code | 12:33 |
@sonney2k | DimensionReductionPreprocessor | 12:34 |
@sonney2k | ok | 12:34 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 13:07 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 13:07 | |
CIA-32 | shogun: Heiko Strathmann master * r05ae96b / src/libshogun/features/Subset.h : added comment (+5 more commits...) - http://bit.ly/mypxuZ | 13:16 |
blackburn | sonney2k: that will be a HUGE pull request | 14:18 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 14:20 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 14:20 | |
blackburn | sonney2k: NormDerivativeLem3 is not implemented, may be we will just KILL it? | 15:07 |
blackburn | I want to kill somebody after 3 hrs of refactoring | 15:08 |
blackburn | sonney2k: sorry for such a large pull request | 16:03 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Ping timeout: 255 seconds] | 16:23 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 16:24 | |
@sonney2k | blackburn, feel free to kill it yes | 17:04 |
@sonney2k | but please don't kill me :D | 17:04 |
blackburn | sonney2k: you too :) you probably would want to kill me for these commits :D | 17:04 |
blackburn | many-many lines | 17:05 |
@sonney2k | blackburn, yes...... | 17:09 |
@sonney2k | does apply_to_feature_matrix still work in-place? | 17:10 |
@sonney2k | or what does it do? | 17:10 |
blackburn | ehh | 17:10 |
blackburn | what do you mean? | 17:10 |
@sonney2k | I really wonder why it returns anything | 17:10 |
blackburn | well it returns SGMatrix - feature matrix | 17:10 |
blackburn | before it was returning feature matrix too | 17:10 |
blackburn | but not of SGMatrix, just float64_t* | 17:11 |
@sonney2k | then I did it wrong already before... | 17:11 |
blackburn | you prefer not return anything? | 17:12 |
@sonney2k | I think I wanted it to be applied in-place (at least that is what I remember | 17:12 |
@sonney2k | for now I would just merge what you did but one should think about it | 17:12 |
blackburn | sonney2k: well it is doing preprocessing on features, then returns feature matrix | 17:15 |
@sonney2k | blackburn, yeah it applies things in-place but then basically calls get-feature_matrix on the preproc'd features | 17:15 |
blackburn | I didn't it this way | 17:16 |
blackburn | I set different feature matrix in my algorithms | 17:16 |
@sonney2k | blackburn, hmmhh it was intended this way - though of course deleting the old matrix / setting a new one can be the only option | 17:17 |
CIA-32 | shogun: Sergey Lisitsyn master * rf8cd650 / (22 files): SGMatrix and SGVector usage in Preprocessor - http://bit.ly/imXP6K | 17:18 |
CIA-32 | shogun: Sergey Lisitsyn master * r8a0e983 / (7 files in 2 dirs): Fixed naming {preproc}->{preprocessor} - http://bit.ly/mr8oJj | 17:18 |
CIA-32 | shogun: Sergey Lisitsyn master * r6b70388 / (18 files): Various fixes for examples - http://bit.ly/kiGzgQ | 17:18 |
CIA-32 | shogun: Sergey Lisitsyn master * r0c14232 / (2 files): Rearranged modular config for Preprocessor - http://bit.ly/l7Bhau | 17:18 |
@sonney2k | as long as the test suite still runs through - it is all good | 17:18 |
blackburn | there are some errors but it wasn't related to my commits | 17:18 |
CIA-32 | shogun: Soeren Sonnenburg master * ra26d599 / (3 files in 2 dirs): remove dysfunctional/not-implemented preproc - http://bit.ly/kPg4d5 | 17:19 |
@sonney2k | blackburn, I mean the preproc tests | 17:19 |
blackburn | ah I see | 17:19 |
@sonney2k | there are probably more errors still :) | 17:20 |
blackburn | sonney2k: but how can we resize feature matrix? | 17:20 |
blackburn | I mean I can work on given feature matrix but how can I free it partially after | 17:21 |
blackburn | e.g. 1000d feature matrix -> 2d feature matrix, 998 unused rows | 17:21 |
@sonney2k | blackburn, I usually didn't - I just left the bigger memory chunk lying around and set new feature matrix sizes | 17:21 |
blackburn | don't you think it is bad? | 17:22 |
@sonney2k | I know not optimal but sometimes it was impossible to store the preproc'd data because it was just too big | 17:22 |
@sonney2k | say feature matrix twice | 17:22 |
@sonney2k | (think of norm one or log(1+x) transform) | 17:23 |
blackburn | no problem for preprocessor not changing dimensionality | 17:23 |
blackburn | I think about my algorithms :) | 17:23 |
blackburn | I can't see any way to reduce matrix size.. | 17:23 |
blackburn | sonney2k: do you know any? | 17:25 |
blackburn | it is ok if it will broke matrix elements, I just want to reduce its size | 17:25 |
@sonney2k | blackburn, then you will have to delete[] the features feature_matrix and set a new one (or jsut set a new one - I hope that should take care of the delete[]) | 17:26 |
@sonney2k | blackburn, it can still happen that you run out of memory | 17:26 |
blackburn | bad | 17:27 |
@sonney2k | the only way I see is to use SG_MALLOC / SG_REALLOC | 17:27 |
blackburn | I think about realloc | 17:27 |
blackburn | some SGMatrix::resize that doing realloc | 17:27 |
@sonney2k | however that will reduce memory on many systems but not OSX (!) | 17:27 |
blackburn | what's wrong with osx? | 17:28 |
@sonney2k | for this to work we would need to get only SG_MALLOC'd matrices - no longer new[]'d ones | 17:28 |
blackburn | is there some problems with new? | 17:29 |
@sonney2k | blackburn, one cannot do REALLOC on new'd memory | 17:32 |
blackburn | ah | 17:32 |
blackburn | I see | 17:32 |
blackburn | sonney2k: anyway there are (number_of_examples)^2 matrix in both LLE and MDS | 17:34 |
blackburn | so it is very memory consumpting at all | 17:35 |
blackburn | with realloc or without | 17:35 |
@sonney2k | heh :) | 17:35 |
@sonney2k | I guess then it does not really matter that much | 17:35 |
blackburn | and there is no way to not use such large matrices | 17:36 |
@sonney2k | sounds like precomputed kernel matrices | 17:38 |
@sonney2k | I guess there always is a way but it is becoming computationally demanding | 17:38 |
blackburn | the only way is to use another method :D | 17:39 |
blackburn | it is symmetrical | 17:39 |
blackburn | I could use some fancy sym matrix class of course | 17:40 |
blackburn | but lapack doesn't work with it | 17:40 |
blackburn | make[1]: *** No rule to make target `../shogun/preprocessor/NormDerivativeLem3.h', needed by `Preprocessor_wrap.cxx'. Stop. | 17:43 |
blackburn | I've already made distclean! | 17:43 |
blackburn | sonney2k: does it compiles on your machine? | 17:45 |
blackburn | sonney2k: Line 13, modular/Preprocessor_includes.i | 17:47 |
blackburn | sonney2k: seems I should use arpack++ not arpack.. | 18:20 |
blackburn | I would rather use arpack but there is no wrappers to fortran code, I don't sure I can do it :) | 18:21 |
blackburn | may be some 'Writing wrappers for Fortran 77 for dummies' book exists? | 18:23 |
-!- blackburn1 [~blackburn@31.28.40.202] has joined #shogun | 18:38 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Read error: Connection reset by peer] | 18:39 | |
serialhex | blackburn1: fortran?? | 19:40 |
blackburn1 | serialhex: I'd like to use arpack :) it is written in fortran 77 | 19:42 |
blackburn1 | lapack too btw | 19:42 |
serialhex | really?? hmm *is kind of afraid of old code* | 19:43 |
blackburn1 | serialhex: it is simply faster :) | 19:43 |
serialhex | that's cool, but still, who uses fortran anymore??? or fortran 77 for that matter!?!? | 19:43 |
serialhex | didn't that die out with the last of the dodo's?? | 19:44 |
blackburn1 | hey but why should it die? | 19:44 |
blackburn1 | :) | 19:44 |
serialhex | i dunno... i'm just wondwering | 19:45 |
serialhex | so whats going on with you? | 19:45 |
blackburn1 | fortran compilers are optimized, safety, blabla | 19:45 |
blackburn1 | they are doing it for 40+, it is almost bug-free and so on | 19:46 |
blackburn1 | moreover some loops vectorized automatically and there are a lot of opportunities to make code parallelized | 19:46 |
blackburn1 | today have been doing some refactoring, it was awful | 19:47 |
serialhex | cool cool... my objection was mostly in jest, at least it's not using cobol :P | 19:47 |
serialhex | ooh, refactoring pains... | 19:47 |
blackburn1 | had to introduce some new class and some naming issues | 19:48 |
blackburn1 | and float64_t* -> SGVector | 19:49 |
serialhex | hmm, sounds like fun! | 19:49 |
blackburn1 | serialhex: have you seen pictures in my weekly report? :) | 19:50 |
serialhex | yeah, the spooky faces & the roll :P | 19:50 |
blackburn1 | that's fun - refactoring not :D | 19:51 |
@bettyboo | ha ha | 19:51 |
CIA-32 | shogun: Soeren Sonnenburg master * rbbd6767 / src/modular/Preprocessor_includes.i : really remove all occurences of NormDerivativeLem3 - http://bit.ly/loslKS | 19:51 |
blackburn1 | sonney2k: thanks | 19:52 |
blackburn1 | I didn't want to make a pull request with 1-liner :) | 19:52 |
@sonney2k | blackburn1, thanks for pointing it out | 19:52 |
blackburn1 | thankfulness on shogun | 19:52 |
@sonney2k | blackburn1, getting arpack wrappers should be easy | 19:52 |
@sonney2k | what are the funcitons you need? | 19:52 |
blackburn1 | sonney2k: yes? how? | 19:53 |
blackburn1 | I don't know exactly | 19:53 |
blackburn1 | yet | 19:53 |
blackburn1 | there are no headers - that's a problem. or not? | 19:53 |
@sonney2k | blackburn1, so? just create the header like in lib/lapack.h | 19:54 |
@sonney2k | blackburn1, btw I guess your refactoring also broke the static interfaces (received complaints...) | 19:54 |
* sonney2k testing | 19:55 | |
blackburn1 | just preproc - preprocessing issue I guess | 19:55 |
@sonney2k | blackburn1, look at the fortran function signatures | 19:55 |
@sonney2k | I guess so too :) | 19:55 |
@bettyboo | hihi | 19:55 |
blackburn1 | I just don't use static interfaces :) | 19:56 |
blackburn1 | how it works?! | 19:56 |
blackburn1 | we can just call fortran function? | 19:56 |
@sonney2k | blackburn1, the same way as linking to any dynamic library works - we just need a .so file with the function entry point and matching signature | 19:59 |
@sonney2k | blackburn1, I think I fixed the compile errrors | 19:59 |
blackburn1 | wow, it is just like I had enquired how child is born :D | 20:00 |
blackburn1 | surprise for me, never knew how it works | 20:00 |
blackburn1 | sonney2k: preproc? | 20:01 |
@sonney2k | yup | 20:01 |
blackburn1 | but who uses up-to-date static shogun?? | 20:01 |
@sonney2k | hmmhh no news from alesis :( | 20:01 |
@sonney2k | blackburn1, not me :) | 20:01 |
CIA-32 | shogun: Soeren Sonnenburg master * rbee38a5 / (2 files): fix compilation errors in libshogunui caused by preproc refactoring - http://bit.ly/jQnzoO | 20:02 |
blackburn1 | I thought libshogunui is dead completely :) | 20:03 |
@sonney2k | blackburn1, so which arpack functions do you need? | 20:08 |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 20:11 | |
-!- blackburn1 [~blackburn@31.28.40.202] has quit [Read error: Connection reset by peer] | 20:11 | |
@sonney2k | blackburn, so which arpack functions do you need? | 20:11 |
blackburn | 10 minutes :) | 20:14 |
blackburn | dsaupd I guess | 20:17 |
blackburn | hmm. they have bad documentation | 20:18 |
@sonney2k | blackburn, I guess that will help you http://www-heller.harvard.edu/people/shaw/programs/dsaupd.h | 20:20 |
@sonney2k | look at the extern "C" line | 20:20 |
blackburn | ah! | 20:21 |
blackburn | I see | 20:21 |
blackburn | so I should add this to arpack.h in lib/? | 20:21 |
blackburn | nice | 20:22 |
@sonney2k | yes | 20:23 |
blackburn | how simple | 20:23 |
@sonney2k | and some minimal code to call that function | 20:23 |
blackburn | for .configure? | 20:23 |
@sonney2k | well that dsaupd function has a trillion arguments... | 20:23 |
@sonney2k | yes | 20:23 |
blackburn | I see | 20:23 |
@sonney2k | so non-trivial to get it working | 20:23 |
blackburn | yes, many arguments, just like in some lapack routines | 20:23 |
@sonney2k | well lapack has really a lot less... | 20:24 |
@sonney2k | 16 arguments! | 20:24 |
* sonney2k counted | 20:24 | |
blackburn | where are drivers | 20:24 |
blackburn | and three 'work' arrays | 20:24 |
blackburn | and info | 20:24 |
blackburn | and some params and strategies | 20:25 |
@sonney2k | blackburn, did I understand this right - dsaupd is for sparse matrices? | 20:26 |
blackburn | seems no | 20:27 |
blackburn | dseupd for eigenvalues | 20:28 |
blackburn | it is some postprocessing routine, I don't know for sure | 20:28 |
blackburn | ah | 20:28 |
blackburn | yes | 20:28 |
blackburn | right | 20:28 |
serialhex | wow... shogun has a total of 2,135,434 loc :P | 20:31 |
blackburn | loc? | 20:31 |
serialhex | 1,753,324 is in C++ | 20:31 |
serialhex | lines of code | 20:31 |
blackburn | ah | 20:31 |
blackburn | I see | 20:31 |
serialhex | yeah | 20:31 |
blackburn | wait, how? | 20:32 |
blackburn | it has 130+ | 20:32 |
blackburn | K | 20:32 |
serialhex | 1,753 kloc :P | 20:33 |
@bettyboo | hehe!?! | 20:33 |
serialhex | yes bettyboo! insane!!! | 20:33 |
@bettyboo | serialhex: eh.. what you mean? | 20:33 |
blackburn | ehh? it cant be true :) | 20:33 |
blackburn | http://www.ohloh.net/p/shogun/analyses/latest | 20:34 |
serialhex | https://gist.github.com/1034570 thats the results i got from running http://cloc.sourceforge.net/ | 20:34 |
serialhex | or, more specifically: cloc --by-file-by-lang --exclude-dir=.git . | 20:35 |
blackburn | something is wrong, it can't be true for sure | 20:36 |
serialhex | maybe it included the .git dir accidentially? | 20:36 |
blackburn | may be, or shogun-data :) | 20:36 |
blackburn | because linux kernel have ~14M | 20:37 |
serialhex | maybe | 20:37 |
blackburn | if shogun was ~2M lines of code sonney2k would be happy :D | 20:38 |
@bettyboo | <:*) | 20:38 |
@sonney2k | blackburn, I don't think so | 20:38 |
@sonney2k | then it would never compile | 20:38 |
@sonney2k | serialhex, I guess you should do git clean -dfx before doing so | 20:39 |
serialhex | oh, der that might help :P | 20:39 |
serialhex | ahh yeah, much more reasionable! | 20:40 |
serialhex | https://gist.github.com/1034570 | 20:40 |
serialhex | 74,644 loc in C++ | 20:42 |
serialhex | with 134,648 loc total | 20:43 |
serialhex | much more sane!! :D | 20:43 |
blackburn | sonney2k: I imagine I we had 2M loc and doing ST* -> SGVector refactoring :D | 20:44 |
serialhex | hmm... though this gives me a different number than the one you sent me to blackburn... idk, different was of counting i guess :P | 20:44 |
blackburn | may be it is not actual at ohloh | 20:46 |
blackburn | Objective C 6 23 0 385 | 20:47 |
blackburn | hahah | 20:47 |
blackburn | shogun for iPhone™ | 20:47 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 21:01 | |
-!- blackburn1 [~blackburn@31.28.40.202] has joined #shogun | 21:01 | |
-!- blackburn1 [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 21:13 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 21:13 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Read error: Connection reset by peer] | 21:15 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 21:16 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Ping timeout: 240 seconds] | 21:21 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 21:22 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Ping timeout: 255 seconds] | 21:28 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 21:29 | |
@sonney2k | blackburn, it did run on Gunnar's iphone btw... | 21:38 |
blackburn | shogun on iphone? :D | 21:38 |
blackburn | for what?? :) | 21:39 |
@bettyboo | 8) | 21:39 |
blackburn | sonney2k: I was surprised gunnar answered with some praise about my progress :) | 21:39 |
@sonney2k | me too | 21:39 |
@sonney2k | I thought he has -inf time and doesn't read mail :) | 21:40 |
@bettyboo | <:*) | 21:40 |
@sonney2k | maybe he is on holidays :D | 21:40 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Ping timeout: 255 seconds] | 21:44 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 21:47 | |
blackburn | damn I'm tired with this wi-fi | 21:51 |
blackburn | I guess something is wrong with driver for adapter on other machine | 21:51 |
blackburn | sonney2k: do we plan to integrate some 'online' for preprocessors too? | 21:56 |
blackburn | I have some ideas how to make my algos 'online' but not sure | 21:58 |
@sonney2k | not a top priority | 22:00 |
blackburn | sonney2k: what about apply_to_feature_matrix? | 22:11 |
blackburn | should it return nothing? | 22:12 |
@sonney2k | waht about it? | 22:12 |
blackburn | you said it is wrong it returns anything | 22:12 |
@sonney2k | yeah | 22:30 |
@sonney2k | you should check if this is used though | 22:30 |
blackburn | actually nothing bad with it but may be it should return pointer to given features | 22:35 |
blackburn | i.e. it is _modifying_ function and it should be emphasized somehow | 22:36 |
@sonney2k | blackburn, up to you... | 22:37 |
blackburn | will think about it | 22:38 |
blackburn | btw naming sucks | 22:38 |
blackburn | apply_to_feature_matrix but features is passed :) | 22:38 |
@sonney2k | you mean apply() only? | 22:39 |
blackburn | i don't know but naming is wrong - apply_to_features(features) is ok, but apply_to_feature_matrix(features) looks strange | 22:39 |
blackburn | may be apply(SGMatrix or SGVector) should be better | 22:40 |
blackburn | just like in ->machine transition :) | 22:40 |
@sonney2k | but that does not make sense for preproc | 22:40 |
@sonney2k | because things can be strings / sparse matrix etc | 22:40 |
blackburn | well I understand why it is called apply_to_feature_matrix but it could be confusing | 22:41 |
@sonney2k | we had the apply_to_example or so function to be able to apply the preproc to just one example | 22:41 |
@sonney2k | and matrix to the whole set of examples | 22:41 |
@sonney2k | blackburn, you are now the preproc master - so naming is up to you | 22:43 |
blackburn | yeap :) | 22:44 |
@sonney2k | I am thinking of removing all the legacy typemaps now | 22:44 |
@sonney2k | otherwise it is too hard to see where examples fail | 22:44 |
@sonney2k | dammed I will just do it | 22:45 |
@sonney2k | this reminds me that I still have to fix r_modular / octave_modular typemaps. | 22:45 |
blackburn | you mean all these float32_t** dst ..? | 22:47 |
@sonney2k | yes | 22:53 |
@sonney2k | die | 22:54 |
@sonney2k | floatxxx** bla | 22:54 |
@sonney2k | all these modular/*.i files that do apply -> die | 22:54 |
blackburn | hehe | 22:58 |
blackburn | let's make a funeral | 22:58 |
blackburn | tragic moment | 22:58 |
@sonney2k | blackburn, if you'd like to help go wild and over all the .h / .cpp files and replace these doulbe* or ** thingies with SGVector/SGMatrix,SGStringList, SGSparseMatrix / SparseVector SGNDArray | 23:13 |
@sonney2k | I erased all typemaps now and am now porting octave / r_modular to new datatypes | 23:14 |
blackburn | sonney2k: ok, will do tomorrow | 23:14 |
blackburn | ah, damn it, will do now | 23:23 |
@sonney2k | alright we have octave typemaps | 23:24 |
blackburn | sonney2k: oh hehe it is much code to change | 23:24 |
@sonney2k | yes | 23:25 |
@sonney2k | do file after file please | 23:25 |
@sonney2k | such that we can parallelize | 23:25 |
@sonney2k | I will notify you when I have the r_typemaps | 23:25 |
@sonney2k | then I join you | 23:25 |
blackburn | sonney2k: I will do classifier now | 23:25 |
blackburn | just after get landmarkmds prototype to work :) | 23:26 |
blackburn | sonney2k: just interface issues now, not guts of algos, right? | 23:33 |
@sonney2k | only interfaces for now | 23:34 |
blackburn | sonney2k: CKNN::classify_for_multiple_k | 23:35 |
blackburn | phew! | 23:36 |
@sonney2k | ? | 23:36 |
blackburn | do we need it? | 23:36 |
@sonney2k | sure | 23:36 |
blackburn | it works strange way | 23:36 |
@sonney2k | this is cool for knn model selection | 23:36 |
blackburn | not returning labels but some array | 23:36 |
@sonney2k | I mean you run knn once with large k and get the results for free for all | 23:37 |
@sonney2k | k | 23:37 |
@sonney2k | k < max_K | 23:37 |
blackburn | yes I understand | 23:37 |
blackburn | okay will adapt it | 23:37 |
blackburn | hm it is simple | 23:38 |
blackburn | hehe | 23:38 |
@sonney2k | compiling for libshogun,libshogunui,r,matlab,octave,python,r_modular,octave_modular,ruby_modular,csharp_modular,python_modular,java_modular | 23:44 |
blackburn | hahaha | 23:44 |
blackburn | better go sleep, at morning it will be done | 23:44 |
blackburn | knn done | 23:46 |
@sonney2k | blackburn, let me have a look at kernels in the meantime | 23:49 |
blackburn | sonney2k: could you please do it for distributions? | 23:49 |
@sonney2k | or distributions if you want | 23:49 |
blackburn | and classifier/PluginEstimate | 23:50 |
blackburn | because too much code for me, I think it is easier for you | 23:50 |
@sonney2k | hmmhh alesis didn't send any email - I wonder what happened.... | 23:51 |
@sonney2k | last time I met him on IRC he was pretty enthusiastic | 23:51 |
blackburn | bad | 23:51 |
blackburn | WTF is classifier/svm/Tron?? | 23:53 |
@sonney2k | some liblinear solver | 23:53 |
blackburn | is it used somewhere externally? | 23:53 |
@sonney2k | no | 23:54 |
@sonney2k | (I think) | 23:54 |
blackburn | I hope so | 23:54 |
blackburn | I've never seen such | 23:54 |
blackburn | oh | 23:54 |
@sonney2k | hmmhh alesis did his last commit exactly 1 week ago | 23:55 |
blackburn | i won't sleep this night :D | 23:55 |
blackburn | sonney2k: bad.. | 23:56 |
@sonney2k | blackburn, he might have had a good reason... hope he is well | 23:57 |
blackburn | sonney2k: his mentor is in trip, right? | 23:57 |
blackburn | sonney2k: just write him email, no any suggestion | 23:57 |
@sonney2k | yes - his mentor (vojtech) just returned | 23:57 |
@sonney2k | did already | 23:58 |
blackburn | oh this solvers make me mad | 23:59 |
--- Log closed Mon Jun 20 00:00:01 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!