--- Log opened Mon May 14 00:00:40 2012 | ||
-!- av3ngr [av3ngr@nat/redhat/x-crzsxpnlvzddpsxs] has joined #shogun | 02:17 | |
-!- abn_ [av3ngr@nat/redhat/x-fpjupbrdnhavmnjp] has joined #shogun | 03:30 | |
-!- av3ngr [av3ngr@nat/redhat/x-crzsxpnlvzddpsxs] has quit [Ping timeout: 252 seconds] | 03:31 | |
-!- abn_ [av3ngr@nat/redhat/x-fpjupbrdnhavmnjp] has quit [Client Quit] | 03:33 | |
-!- vikram360 [~vikram360@117.192.165.186] has quit [Ping timeout: 244 seconds] | 05:46 | |
-!- vikram360 [~vikram360@117.192.161.59] has joined #shogun | 05:46 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 06:27 | |
-!- vikram360 [~vikram360@117.192.161.59] has quit [Quit: Leaving] | 06:38 | |
-!- uricamic [~uricamic@2001:718:2:1634:8cea:f88e:2be4:5ab] has joined #shogun | 09:00 | |
-!- emrecelikten [~emrecelik@213.74.82.26] has joined #shogun | 09:02 | |
emrecelikten | Hi all | 09:06 |
---|---|---|
n4nd0 | hey emrecelikten | 09:15 |
emrecelikten | n4nd0: How are you today? | 09:16 |
n4nd0 | emrecelikten: I am fine, what about you? | 09:16 |
emrecelikten | n4nd0: Fine, had a GSoC all-nighter. "Converging" into zombie mode with each passing hour :D | 09:16 |
n4nd0 | :D | 09:17 |
n4nd0 | brb | 09:17 |
CIA-113 | shogun: Soeren Sonnenburg master * r73b9eea / src/interfaces/lua_modular/swig_typemaps.i : remove left over %enddef - http://git.io/wt_gnQ | 09:22 |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 09:38 | |
eric_ | hi there | 09:38 |
n4nd0 | sonne|work: good morning! around?? | 09:45 |
n4nd0 | Nico and I discussed the other day that it would be good to think of how doing KernelSOMachine even if it is *very very* slow | 09:46 |
n4nd0 | in order to introduce it in shogun I came up with two possibilities for the class hierarchy | 09:47 |
n4nd0 | either: CMachine <---- CSOMachine and CSOMachine has two children, CLinearSOMachine and CKernelMachine | 09:47 |
n4nd0 | or: CLinearMachine <---- CLinearSOMachine and CKernelMachine <---- CKernelSOMachine | 09:47 |
n4nd0 | I like better the first possibility, sonne|work what do you think? | 09:48 |
sonne|work | n4nd0: yes first one | 09:48 |
sonne|work | it matches what we do with multiclassmachines | 09:48 |
n4nd0 | sonne|work: good, thank you! | 09:49 |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 10:11 | |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Quit: Leaving.] | 11:17 | |
-!- emrecelikten is now known as emre-away | 11:29 | |
-!- cronor [~cronor@fb.ml.tu-berlin.de] has joined #shogun | 11:52 | |
cronor | is it intended that in python you have to import CSVMLightOneClass? everything else is without the C | 11:53 |
n4nd0 | cronor: there might probably be a line missing in interfaces/python_modular/Classifier.i or similar file | 11:55 |
cronor | n4nd0: i'll check it | 11:55 |
-!- emre-away is now known as emrecelikten | 12:23 | |
CIA-113 | shogun: Soeren Sonnenburg master * reed11fa / src/interfaces/modular/Classifier.i : remove C prefix from CSVMLightOneClass in modular interfaces - http://git.io/TYSPwg | 13:22 |
sonne|work | cronor: thx | 13:52 |
n4nd0 | cronor: you fixed it, cool :) | 13:54 |
cronor | n4nd0 no, soeren fixed it. i wasn't sure if i should commit a one-line fix | 13:55 |
n4nd0 | oh, ahyhow, it got fixed | 13:56 |
cronor | although i would like to contribute to the project. are there any open issues which can be done without too much cpp knowledge? i looked through the github issues and they seem all kind of big | 13:57 |
n4nd0 | cronor: if you would like to avoid cpp you could for example work on some examples in python or any other language we support of your preference | 13:58 |
cronor | it's not about avoiding cpp, i just don't have very little experience and don't feel able to do a restructuring issue | 14:00 |
n4nd0 | all right | 14:00 |
cronor | *just have little experience | 14:00 |
n4nd0 | I know that sonne|work was interested in expanding LDA to support multiclass | 14:00 |
sonne|work | yup | 14:01 |
n4nd0 | there is already code implemented in python to do that in scikits | 14:01 |
eric_ | hi all | 14:01 |
n4nd0 | I ported their QDA into shogun | 14:02 |
n4nd0 | so I guess that the plan here would be to substitute our current LDA by something similar to what they have there, working for multiclass directly | 14:02 |
n4nd0 | sonne|work: right? | 14:02 |
eric_ | which string kernel could you advice me to use for string features of different sizes ? and using a quite large alphabel (RAWBYTE). Thanks in advance for any hints! | 14:02 |
n4nd0 | eric_: hey! I am not an expert on string kernels but, isn't the normal thing that string features are of different lengths?? | 14:03 |
eric_ | n4nd0: no in shogun most of the string kernel are implemented to compare strings of the same length | 14:04 |
n4nd0 | eric_: oh, my bad then :( | 14:05 |
eric_ | n4nd0: only few of them (maybe I am wrong..) are compatible with fetures of different size, and since I have a quite big alphabet I dont know which kernel could do the work ?? | 14:05 |
n4nd0 | eric_: I am sorry I cannot help you, I don't know that much about different string kernels | 14:06 |
cronor | n4nd0: i'll look into it tonight and see | 14:06 |
n4nd0 | cronor: all right, feel free to ask me around here if you need some help | 14:06 |
eric_ | n4nd0: do you know who implemented the available string kernels in shogun ? | 14:07 |
n4nd0 | eric_: I am not sure ... I'd say sonne|work is the most likely option | 14:07 |
n4nd0 | eric_: anyway, have you just tried testing different of them and analyse which one gives to you better results? | 14:08 |
eric_ | n4nd0: allright, thx, I hope he will read the logs | 14:08 |
n4nd0 | eric_: I guess you could do an example using your test data and just plug in one or other StringKernel and see which one performs best | 14:08 |
sonne|work | n4nd0: re MC-LDA yes or even just add a multiclass variant | 14:09 |
n4nd0 | sonne|work: I have a bit of trouble here https://gist.github.com/2634487 | 14:09 |
sonne|work | ? | 14:10 |
n4nd0 | sonne|work: you see for example in CResultTest or in CStructuredApplication::get_joint_feature_representation | 14:10 |
n4nd0 | the return type | 14:10 |
n4nd0 | the one that appears there as vector | 14:10 |
n4nd0 | I am not sure what to use in shogun since it must be something like an SGVector | 14:10 |
n4nd0 | but it may work if the features are Dense, Sparse or String | 14:11 |
n4nd0 | sonne|work: do you know what I mean? | 14:11 |
sonne|work | no | 14:13 |
sonne|work | n4nd0: can't you use sth like http://shogun-toolbox.org/doc/en/current/classshogun_1_1CDotFeatures.html | 14:14 |
sonne|work | I mean never ever explicitly use the feature representation but just define some operations that are needed? | 14:15 |
eric_ | sonne|work: Hi, hopping you have time to respond: what family of string kernels should I focus if I use string features of different size with a quite big alphabet ? | 14:16 |
n4nd0 | sonne|work: but I don't think that the joint feature representation should return something like CDotFeatures, it should return just a feature vector | 14:17 |
n4nd0 | sonne|work: so as I see it is, is the joint space is represented with DenseFeatures it should return a SGVector, if it is with SparseFeatures a SGSparseVector, and so on | 14:18 |
sonne|work | eric_: some n-gram thing I would say... probably hashed | 14:18 |
sonne|work | n4nd0: what I meant is - why is it necessary at all? | 14:18 |
sonne|work | otherwise I think it is just a SGVector<float64_t> | 14:18 |
sonne|work | but a huuuuuge one | 14:18 |
sonne|work | (potentially) | 14:19 |
n4nd0 | sonne|work: mm what do you mean with why is it necessary? | 14:19 |
sonne|work | n4nd0: for example in SVMs you never need access to the examples x | 14:21 |
sonne|work | or Phi(x) | 14:21 |
sonne|work | all you need is the operations defined in dotfeatures | 14:21 |
sonne|work | like w <- w+ alpha*Phi(x) | 14:22 |
eric_ | sonne|work: does shogun permits such ngram hashing ? | 14:22 |
sonne|work | eric_: indeed migt not for n-gram kernel... | 14:23 |
eric_ | sonne|work: tx. another dummy question: I have a alphabet of size=100, how basically can I do the "mapping" to match the CStringFeatures<char> to use it in the implemented string kernels in shogun ? | 14:29 |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun | 16:29 | |
-!- gsomix [~gsomix@188.168.2.14] has joined #shogun | 16:39 | |
gsomix | #????? ! | 16:39 |
gsomix | water, water everywhere | 16:39 |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun [] | 16:41 | |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 16:44 | |
-!- emrecelikten [~emrecelik@213.74.82.26] has quit [Quit: Leaving.] | 16:48 | |
gsomix | blackburn, ????? | 16:50 |
blackburn | heh\ | 16:51 |
n4nd0 | hey gsomix! | 17:08 |
n4nd0 | you have been working lately with SGVector and SGSparseVector right? | 17:08 |
blackburn | gsomix: have you finished with your array conversion? | 17:09 |
-!- uricamic [~uricamic@2001:718:2:1634:8cea:f88e:2be4:5ab] has quit [Quit: Leaving.] | 17:18 | |
CIA-113 | shogun: Soeren Sonnenburg master * rea2e2f2 / src/interfaces/lua_modular/swig_typemaps.i : fix valgrind error in lua typemap - http://git.io/kqOp3Q | 17:21 |
blackburn | oo | 17:22 |
-!- karlnapf [~heiko@host86-174-150-108.range86-174.btcentralplus.com] has joined #shogun | 17:32 | |
gsomix | n4nd0, a little. sonney2k working with SGVector/SGMatrix/SGSparseVector now. | 17:38 |
gsomix | blackburn, nope. | 17:38 |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Quit: Leaving.] | 17:38 | |
n4nd0 | gsomix: nico and I are wondering if we could have a type that could behave either as a SGVector or as a SGSparseVector | 17:40 |
n4nd0 | you know to put a method like | 17:40 |
n4nd0 | vector f() {} | 17:41 |
n4nd0 | and some implementations of f return SGVector and others SGSparseVector | 17:41 |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 17:45 | |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 17:52 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 17:54 | |
gsomix | n4nd0, it sounds cool. but I think you should talk with sonney2k about it. sorry | 17:54 |
blackburn | n4nd0: why do you need it? | 17:55 |
n4nd0 | gsomix: ok, I'll ask him | 17:57 |
n4nd0 | blackburn: https://gist.github.com/2634487 | 17:57 |
n4nd0 | look at line 101 | 17:58 |
n4nd0 | for the joint feature vectors | 17:58 |
n4nd0 | i.e. the feature vectors that one do with information of the training data and the labels | 17:58 |
blackburn | n4nd0: can it be sparse? | 17:58 |
n4nd0 | something like psi(xi, yi) | 17:58 |
n4nd0 | blackburn: nico said that normally dense representation is used | 17:59 |
n4nd0 | but that he would like to use sparse for high dimensional spaces | 17:59 |
n4nd0 | he seemed interested in that sparse vector provided here | 18:00 |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Quit: Page closed] | 18:02 | |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun | 18:02 | |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 18:04 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 18:13 | |
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Quit: Page closed] | 18:23 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 18:44 | |
-!- cronor [~cronor@fb.ml.tu-berlin.de] has quit [Ping timeout: 248 seconds] | 18:51 | |
@sonney2k | karlnapf, around and have a bit of time? | 19:22 |
@sonney2k | n4nd0 - this is the wrong(tm) approach | 19:22 |
@sonney2k | karlnapf, if so | 19:22 |
@sonney2k | karlnapf, please run make check-valgrind in the libshogun dir | 19:23 |
karlnapf | sonney2k, hi there, I have a few minutes unfortunately :( | 19:24 |
karlnapf | yes I know that the examples fail | 19:24 |
karlnapf | I already had a look but couldnt find the error after some time | 19:24 |
karlnapf | I dont really get it, for the SGVector transition, just de-activating the reference counts worked | 19:24 |
karlnapf | but for the matrices, it doesnt work | 19:25 |
@sonney2k | karlnapf, I think it is due to the way we serialize sgvector stuff | 19:25 |
karlnapf | yes | 19:25 |
karlnapf | definately | 19:25 |
@sonney2k | we need a separate way... | 19:25 |
@sonney2k | basically we have to store the refcount for these too | 19:25 |
CIA-113 | shogun: Heiko Strathmann master * r33e2e37 / (7 files): -some interface changes after talked to Arthur - http://git.io/TyMy0w | 19:26 |
CIA-113 | shogun: Heiko Strathmann master * r2639838 / src/shogun/evaluation/CrossValidation.cpp : code cleanups - http://git.io/zn11kg | 19:26 |
CIA-113 | shogun: Heiko Strathmann master * rd2863d9 / (8 files in 2 dirs): Merge pull request #526 from karlnapf/master - http://git.io/9kZc4Q | 19:26 |
@sonney2k | otherwise we have a leak at some point /double free | 19:26 |
karlnapf | yes | 19:26 |
karlnapf | I already thought a bit about this | 19:26 |
@sonney2k | I even don't mind if you do an incompatible change here | 19:27 |
karlnapf | I am a bit afraid that this will cause trouble | 19:27 |
karlnapf | when you save there is a certain refcount | 19:27 |
karlnapf | then when you load from another situation | 19:27 |
@sonney2k | shogun 2.0 is very different from 1.0 - so we cannot really do anything about it | 19:27 |
karlnapf | its not correct anymore | 19:27 |
karlnapf | well, that solves at least the migration problem :) | 19:28 |
@sonney2k | karlnapf, ok this will happen if we have external objects that are not serialized pointing to things | 19:28 |
karlnapf | sonney2k, ok, lets save the refcount into the vector then | 19:29 |
karlnapf | however, this still doesnt solve the mem-leak problem | 19:29 |
@sonney2k | but still much better than leaks in the general case | 19:29 |
@sonney2k | why not? | 19:29 |
karlnapf | at least I think | 19:29 |
karlnapf | because if you de-activate the ref-counting | 19:30 |
karlnapf | it still has to work | 19:30 |
@sonney2k | yes but w/ leaks | 19:30 |
karlnapf | why? there were no leaks before the transition | 19:30 |
@sonney2k | exactly | 19:31 |
@sonney2k | *before* | 19:31 |
@sonney2k | now we never use SG_FREE(vec.vector) or destroy_vector() etc | 19:31 |
karlnapf | but if I now use the system with the ref-counting deactivated | 19:31 |
karlnapf | Oh, I already checked that | 19:31 |
karlnapf | I manually added SG_FREEs to the examples | 19:31 |
karlnapf | still leaking | 19:31 |
karlnapf | there is something more subtle going on | 19:31 |
@sonney2k | it is not the correct way anyway | 19:32 |
karlnapf | yes true | 19:32 |
karlnapf | mmh, I really fear touching all these examples again :) | 19:32 |
@sonney2k | karlnapf, so please ping me when you write out the refcount along with the vector | 19:32 |
karlnapf | but ok | 19:32 |
karlnapf | yes, will notify you. | 19:32 |
karlnapf | sorry for being not so much available currently, but theres another exam tomorrow | 19:33 |
@sonney2k | oh | 19:33 |
@sonney2k | ok | 19:33 |
karlnapf | and two next week | 19:33 |
@sonney2k | no worries | 19:33 |
karlnapf | but then I almost got them all, the last two are more relaxed :) | 19:33 |
blackburn | 9th exam - 678 to go but heiko is relaxed now :D | 19:34 |
karlnapf | well, whats the alternative? freaking out all day is quite exhausting :D | 19:35 |
blackburn | I rather joke on # of exams :) | 19:35 |
karlnapf | k :) | 19:36 |
karlnapf | too many | 19:36 |
karlnapf | my brain feels so saturated | 19:36 |
karlnapf | ok, guys, gotta go, take care sonney2k, blackburn, bye | 19:36 |
@sonney2k | blackburn, does the lua_modular stuff die on your machine too | 19:36 |
@sonney2k | karlnapf, cu! and thanks | 19:36 |
blackburn | sonney2k: I have never never ever tried | 19:37 |
-!- karlnapf [~heiko@host86-174-150-108.range86-174.btcentralplus.com] has quit [Quit: Leaving.] | 19:37 | |
@sonney2k | blackburn, then please try features_string_char_modular.lua | 19:39 |
blackburn | sonney2k: ok need to switch system | 19:40 |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 19:40 | |
@sonney2k | hmmhh | 19:43 |
@sonney2k | look at this | 19:43 |
@sonney2k | #0 0x00007ffff5f84e7c in shogun::CStringFeatures<char>::set_feature_vector (this=0x0, vector=..., num=0) at features/StringFeatures.cpp:230 | 19:43 |
@sonney2k | #1 0x00007ffff67c2f8b in _wrap_StringCharFeatures_set_feature_vector (L=0x62b010) at modshogun_wrap.cxx:172721 | 19:43 |
@sonney2k | this === 0x0 | 19:43 |
@sonney2k | gsomix, done with dynamicobjectarray? | 19:44 |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 19:44 | |
gsomix | sonney2k, nope. I'm little busy with optics now. | 19:45 |
@sonney2k | blackburn, <sonney2k> #0 0x00007ffff5f84e7c in shogun::CStringFeatures<char>::set_feature_vector (this=0x0, vector=..., num=0) at features/StringFeatures.cpp:230 | 19:51 |
@sonney2k | <sonney2k> #1 0x00007ffff67c2f8b in _wrap_StringCharFeatures_set_feature_vector (L=0x62b010) at | 19:51 |
@sonney2k | crazy! | 19:51 |
@sonney2k | seems like the object is dead | 19:51 |
blackburn | sonney2k: cool :D | 19:52 |
blackburn | sonney2k: modshogun_wrap.cxx:742:17: fatal error: lua.h: No such file or directory | 19:52 |
blackburn | sonney2k: which pkg? | 19:52 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 19:56 | |
n4nd0 | sonney2k: hi! tell me, what is the good approximation then :)? | 19:56 |
@sonney2k | n4nd0, as I said | 19:57 |
@sonney2k | n4nd0, don't use the feature representation | 19:57 |
@sonney2k | figure out which operations you need on Phi(x,y) and define only these | 19:57 |
@sonney2k | e.g <Phi(x,y),w> | 19:58 |
@sonney2k | w <- w+ alpha Phi(x,y) | 19:58 |
@sonney2k | etc | 19:58 |
@sonney2k | so you will never need to use Phi(x,y) *explicitly* | 19:58 |
n4nd0 | sonney2k: ok, I understand that | 19:59 |
n4nd0 | sonney2k: but what I don't see is how to considere it that way is going to save me somehow the need of defining a return type for the function | 19:59 |
@sonney2k | n4nd0, that trick is very powerful... we use it in linear svms to train in million dim feature spaces w/ millions of examples w/o every computing Phi(x) explicitly | 20:00 |
@sonney2k | you don't need that function at all | 20:00 |
blackburn | COFFIN! | 20:00 |
@sonney2k | yeah | 20:00 |
n4nd0 | aham, so no function | 20:00 |
@sonney2k | n4nd0, you can provide a default one that returns a SGVector by computing w <- 0 + Phi(x,y) following the example above | 20:02 |
n4nd0 | sonney2k: blackburn , can you put me an example how to apply this trick in a simple case? | 20:05 |
n4nd0 | let's say we have to do <Phi(x,y), w> | 20:05 |
n4nd0 | how do we compute without computing Phi(x,y) explicitily | 20:06 |
blackburn | n4nd0: if you need <Phi(x,y), w> then just provide a function that does <Phi(x,y), w> | 20:06 |
n4nd0 | haha | 20:06 |
blackburn | hmm it is not said that it is always possible | 20:06 |
blackburn | not explicitly | 20:06 |
n4nd0 | eeeh can you elaborate there a bit? :) | 20:06 |
blackburn | n4nd0: okay for example we have poly features | 20:07 |
blackburn | n4nd0: in dot features we have only two required operations | 20:07 |
blackburn | dot and add | 20:08 |
blackburn | so if we want poly features we don't need to construct it explicitly | 20:08 |
@sonney2k | n4nd0, please have a look at the link I gave you with the operations in there | 20:09 |
n4nd0 | sonney2k: blackburn I will look at it after dinner and get back with any doubt | 20:09 |
n4nd0 | thank you! | 20:09 |
blackburn | sonney2k: hmm what should be done to make lua work? | 20:11 |
blackburn | lua: features_dense_real_modular.lua:1: module 'modshogun' not found: | 20:13 |
blackburn | no field package.preload['modshogun'] | 20:13 |
blackburn | no file './modshogun.lua' | 20:13 |
@sonney2k | p SWIG_Lua_ConvertPtr(L,1,(void**)&arg1,swig_types[878],0) | 20:14 |
@sonney2k | 0x0 | 20:14 |
@sonney2k | !!! | 20:14 |
@sonney2k | wtf | 20:14 |
@sonney2k | seems like the thing is NULL - so no wonder it dies | 20:14 |
@sonney2k | blackburn, look at the beginning of check.sh | 20:14 |
blackburn | ok | 20:15 |
-!- blackburn [~qdrgsm@188.122.250.167] has quit [Quit: Leaving.] | 20:28 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 20:36 | |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 20:57 | |
-!- blackburn [~blackburn@188.122.250.167] has joined #shogun | 20:57 | |
CIA-113 | shogun: Soeren Sonnenburg master * ra4985db / examples/undocumented/lua_modular/features_string_char_modular.lua : disable set_feature_vector from lua for now - http://git.io/pnkbsw | 20:59 |
CIA-113 | shogun: Soeren Sonnenburg master * rcb50eab / src/interfaces/lua_modular/swig_typemaps.i : simplify lua typemaps (use sgvector & co) - http://git.io/XlIz8w | 20:59 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 20:59 | |
-!- ckwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has joined #shogun | 21:00 | |
-!- blackburn [~blackburn@188.122.250.167] has quit [Quit: Leaving.] | 21:45 | |
-!- blackburn [~qdrgsm@188.122.250.167] has joined #shogun | 21:47 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 21:54 | |
-!- gsomix_ [~gsomix@188.168.14.11] has joined #shogun | 21:58 | |
-!- gsomix [~gsomix@188.168.2.14] has quit [Ping timeout: 244 seconds] | 22:00 | |
-!- gsomix_ [~gsomix@188.168.14.11] has quit [Read error: Operation timed out] | 22:21 | |
n4nd0 | blackburn: hey | 22:34 |
blackburn | n4nd0: hi | 22:35 |
n4nd0 | blackburn: so about the features issue | 22:39 |
n4nd0 | I have got the idea that what I should do then is to make a new class that inherits from CDotFeatures | 22:39 |
n4nd0 | I think that the operations that we need are basically those that are defined there | 22:40 |
blackburn | yeah that is common case | 22:40 |
n4nd0 | and I think that we need a new class since the features here are computed from a feature vector of the input space and some kind of structured data | 22:40 |
n4nd0 | blackburn: do you think that is a good idea? | 22:40 |
blackburn | n4nd0: yes probably | 22:41 |
n4nd0 | ok | 22:41 |
n4nd0 | then I don't really think that the sparse or non sparse issue matter | 22:41 |
n4nd0 | matters* | 22:42 |
n4nd0 | I mean that there is no need to regard it separately | 22:42 |
-!- Marty28 [~marty@cable-158-181-77-81.cust.telecolumbus.net] has joined #shogun | 22:43 | |
n4nd0 | hey Marty28, how is it going? | 22:44 |
Marty28 | Hiho | 22:45 |
Marty28 | Fine | 22:45 |
Marty28 | Applying shogun to several datasets | 22:45 |
n4nd0 | good | 22:46 |
Marty28 | Has your google summer started? | 22:46 |
n4nd0 | nice results? | 22:46 |
Marty28 | Yes for easy cases | 22:46 |
n4nd0 | the oficial date has not come yet, but I think we are all hands on it already :D | 22:46 |
Marty28 | Cool | 22:47 |
Marty28 | I am currently playing hide and seek with shogun | 22:47 |
n4nd0 | why so? | 22:47 |
Marty28 | Creating artificial data and letting shogun identify the features | 22:47 |
n4nd0 | nice, what are you using to identify the features? | 22:48 |
Marty28 | If the field i am in is new i cannot rely on existinf experience | 22:48 |
Marty28 | G | 22:49 |
Marty28 | I have to make assumptions on what feature combonations my labels rely on | 22:50 |
Marty28 | Depend on | 22:50 |
Marty28 | E.g. Localized motifs combined with other numbers | 22:51 |
Marty28 | So first i have to make shogun depend on them | 22:52 |
Marty28 | Else later shogunwill not show me the importance of the real features | 22:54 |
n4nd0 | interesting | 22:55 |
Marty28 | I do not go for sensitivity but for the importance and usage of features as a result | 22:55 |
Marty28 | My boss does not like that | 22:55 |
Marty28 | Bioinformaticists want benchmarks | 22:56 |
-!- in3xes [~in3xes@106.78.49.114] has joined #shogun | 22:56 | |
Marty28 | Biologists want explanations | 22:56 |
n4nd0 | as we all do ;) | 22:57 |
n4nd0 | well ... ok, maybe not all | 22:57 |
Marty28 | So ideally i take features that have been shown to be important by biologists | 22:57 |
Marty28 | In SOME of my positive labels | 22:58 |
Marty28 | Then i optimize shogun for using this type of information | 22:59 |
Marty28 | F shogun finds my features of the biologists' examples | 22:59 |
Marty28 | If | 22:59 |
Marty28 | It will also find/use features that are like these cases | 23:00 |
Marty28 | Candidates for research for biologists | 23:00 |
shogun-buildbot | build #543 of lua_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/543 | 23:00 |
Marty28 | So i guess i have to be careful that the feature selection does not go for the trivial features masking the subtle ones | 23:08 |
Marty28 | Hmmm | 23:09 |
Marty28 | So i could use real data and hide noisy features in the positives | 23:10 |
Marty28 | The i remove the big features and see if the small ones pop up with different methods | 23:11 |
Marty28 | Hmmm | 23:11 |
n4nd0 | there may be also some of weighting for the features that could help here? | 23:11 |
n4nd0 | more importance to the subtle ones could make that they are not forgotten | 23:12 |
Marty28 | I know | 23:12 |
n4nd0 | all right ... | 23:12 |
Marty28 | Still i guess artificial data will help | 23:12 |
Marty28 | Also it gives me presentable results | 23:13 |
Marty28 | My real data is rather difficult | 23:13 |
Marty28 | Alsomi will check how methods will react on mixed phenomena | 23:14 |
Marty28 | I.e. When feature combinations a and b and c lead to +1 label | 23:15 |
Marty28 | E.g. 1,2,4 and 5,6,8 but not 1,6,8 | 23:16 |
Marty28 | I will see, just a master thesis | 23:19 |
n4nd0 | good night people | 23:24 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 23:25 | |
Marty28 | N8 | 23:32 |
-!- cronor [~cronor@e178176184.adsl.alicedsl.de] has joined #shogun | 23:42 | |
-!- Marty28 [~marty@cable-158-181-77-81.cust.telecolumbus.net] has quit [Quit: Colloquy for iPad - http://colloquy.mobi] | 23:43 | |
--- Log closed Tue May 15 00:00:40 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!