| --- Log opened Sat May 14 00:00:58 2011 | ||
| -!- blackburn [~qdrgsm@188.168.3.72] has quit [Quit: Leaving.] | 01:44 | |
| -!- alesis-novik [~alesis@188.74.87.84] has quit [Quit: I'll be Bach] | 03:41 | |
| -!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has quit [Remote host closed the connection] | 03:47 | |
| -!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has joined #shogun | 04:27 | |
| -!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has quit [Remote host closed the connection] | 05:01 | |
| -!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 07:10 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 264 seconds] | 13:51 | |
| -!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has joined #shogun | 15:01 | |
| -!- sudarshan [~sudarshan@c-174-50-225-74.hsd1.tn.comcast.net] has joined #shogun | 15:28 | |
| -!- sudarshan [~sudarshan@c-174-50-225-74.hsd1.tn.comcast.net] has quit [Remote host closed the connection] | 15:36 | |
| -!- blackburn [~qdrgsm@188.122.225.203] has joined #shogun | 15:48 | |
| -!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has quit [Remote host closed the connection] | 17:46 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 21:04 | |
| CIA-31 | shogun: Soeren Sonnenburg master * rb9d67bb / (5 files in 4 dirs): add SGMatrix support in python typemaps - http://bit.ly/jB5H5I | 22:36 |
|---|---|---|
| @sonney2k | blackburn, I only now realized that we have another naming clash | 22:41 |
| @sonney2k | what I call sparse matrix now is actually only a sparse vector | 22:41 |
| @sonney2k | so I will have to do renamings to SGSparseVector etc | 22:42 |
| @sonney2k | and introduce SGSparseMatrix | 22:42 |
| @sonney2k | same for strings btw | 22:42 |
| @sonney2k | SGString | 22:42 |
| @sonney2k | will need to introduce SGStringList | 22:43 |
| blackburn | sonney2k: ehh, but do we have sparse vectors already? | 22:44 |
| @sonney2k | sure but called SGSparseFeatureMatrix or so | 22:45 |
| blackburn | ah | 22:45 |
| blackburn | sonney2k: I need some advice with LLE | 22:46 |
| blackburn | what is better - to allocate memory for all instance life or only within init()? | 22:46 |
| @sonney2k | blackburn, ask but I am not sure if mlsec/cwidmer could help better here | 22:47 |
| @sonney2k | blackburn, I don't understand | 22:47 |
| blackburn | sonney2k: not about algorithmic but about pure programming ;) | 22:47 |
| @bettyboo | rahaha!?! | 22:47 |
| @sonney2k | ok | 22:47 |
| blackburn | I mean I have to allocate some matrices | 22:47 |
| blackburn | should I allocate it within init() method | 22:47 |
| blackburn | or better allocate it with constructor? | 22:47 |
| @sonney2k | blackburn, do you know the sizes before calling init? | 22:48 |
| blackburn | :D | 22:49 |
| @sonney2k | do you use them 'often' | 22:49 |
| blackburn | sorry of course not | 22:49 |
| blackburn | oh how stupid I am | 22:49 |
| @sonney2k | ok | 22:49 |
| blackburn | sonney2k: another question about that preproc framework | 22:50 |
| blackburn | ehh. | 22:50 |
| blackburn | well, we have apply_to_feature_matrix and apply_to_feature_vector | 22:51 |
| blackburn | if to matrix I understand, it will call init, etc | 22:51 |
| blackburn | but what about vector? | 22:51 |
| blackburn | should I use already initialized features? | 22:51 |
| blackburn | sonney2k: https://github.com/SergeyLisitsyn/shogun/blob/274a2fefe4de3c82dc5c585e89ab3edd336a12b0/src/libshogun/preproc/LLE.cpp | 22:52 |
| @sonney2k | blackburn, you have to assume that your preproc is inited | 22:52 |
| blackburn | now all routines are in init | 22:52 |
| blackburn | I'm stucked what to do with apply_* methods | 22:53 |
| @sonney2k | so that you can apply it to single vectors (when the feature matrix is not avalable in memory) | 22:53 |
| @sonney2k | blackburn, btw we have a CMath::display_matrix() | 22:54 |
| blackburn | sonney2k: ok, will know | 22:54 |
| blackburn | it just some crap not cleaned up | 22:54 |
| blackburn | of course there wouldn't be any sg_print :) | 22:54 |
| blackburn | sonney2k: so, when apply_to_feature_matrix is called I should re-init (or init) with given features, right? | 22:55 |
| @sonney2k | blackburn, never re-init | 22:56 |
| @sonney2k | always assume that you are inited | 22:56 |
| blackburn | I'm confused what exactly I should do in init() | 22:56 |
| blackburn | sonney2k: if it is inited already why CFeatures instance is passed to apply_to_feature_matrix? | 22:59 |
| @sonney2k | blackburn, I don't understand - it should apply ti to the CFeatures object that is passed... | 23:01 |
| blackburn | but what should init(CFeatures* f) do? | 23:01 |
| @sonney2k | example: | 23:02 |
| blackburn | I can't see any relation between two different features.. | 23:02 |
| @sonney2k | you want to remove features which have small variance | 23:02 |
| @sonney2k | then you measure variance on init() and recall which dims should be removed | 23:02 |
| @sonney2k | then when you call apply* you use that list | 23:03 |
| blackburn | ah! | 23:03 |
| blackburn | sonney2k: but what about dim reduction techniques like LLE? it is absolutely determined to feature set | 23:05 |
| blackburn | we can't init on one feature set and process other | 23:05 |
| blackburn | do you understand my miscomprehension? | 23:06 |
| @sonney2k | no | 23:07 |
| blackburn | eeeh | 23:07 |
| blackburn | nevermind :) | 23:07 |
| @bettyboo | cool | 23:07 |
| blackburn | will try to understand how it is done with pca | 23:07 |
| blackburn | ah don't understand how to do it with LLE | 23:14 |
| blackburn | sonney2k: will you implement addition of two SGVectors? | 23:20 |
| blackburn | and I'm interested with scalar multiplication too | 23:20 |
| CIA-31 | shogun: Soeren Sonnenburg master * r7b958bf / (57 files in 14 dirs): | 23:24 |
| CIA-31 | shogun: rename sparse matrix to sparse vector and introduce sparse matrix | 23:24 |
| CIA-31 | shogun: structure and string list structure - http://bit.ly/j1Sl11 | 23:24 |
| @sonney2k | blackburn, I won't - but feel free to do so | 23:25 |
| @sonney2k | btw, have a look at DataType.h and tell me if that is OK | 23:25 |
| blackburn | I don't understand why you are asking me for that (in fact I can't grasp preproc) :) | 23:26 |
| blackburn | oh nice I switched branch and lost LLE :D | 23:29 |
| @sonney2k | hard to believe | 23:29 |
| @sonney2k | switch back! | 23:29 |
| blackburn | aha | 23:29 |
| blackburn | will get from github | 23:29 |
| blackburn | but I was surprised | 23:30 |
| blackburn | EHRRR i'm useless | 23:31 |
| @sonney2k | blackburn, bah get some sleep or eat more chocolate ;-) | 23:32 |
| @bettyboo | 8) | 23:32 |
| blackburn | sonney2k: should ops on vectors be implemented as its methods? | 23:34 |
| @sonney2k | blackburn, yes makes most sense - at some point this does no longer belong in datatype.h though - but for now yes | 23:35 |
| blackburn | sonney2k: SGVector.add(a,b)? | 23:36 |
| @sonney2k | no only add(v) | 23:36 |
| blackburn | it is less readable, don't you think so/ | 23:36 |
| blackburn | ? | 23:36 |
| blackburn | ah | 23:38 |
| blackburn | I see | 23:38 |
| @sonney2k | i think it is more readable - it adds v to the current object | 23:44 |
| blackburn | moreover it doesn't copy vector | 23:45 |
| blackburn | but what if we want x,y and x+y both? | 23:45 |
| @sonney2k | sorry what? | 23:47 |
| @sonney2k | ahh I understand | 23:47 |
| @sonney2k | well then feel free to add another function for this | 23:48 |
| blackburn | sonney2k: what about zero vector constructor? | 23:49 |
| @sonney2k | blackburn, implement what you need... | 23:50 |
| blackburn | ok. | 23:53 |
| @sonney2k | blackburn, you probably can take much of what is in CMath* now | 23:54 |
| @sonney2k | e.g. dot, fill vector range_fill vector etct etc | 23:55 |
| blackburn | anyway my addition doesn't compile yet :D | 23:56 |
| -!- warpy [~warpy@bzq-79-183-44-64.red.bezeqint.net] has quit [Ping timeout: 252 seconds] | 23:56 | |
| --- Log closed Sun May 15 00:00:00 2011 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!