--- Log opened Thu Jun 30 00:00:31 2011 | ||
f-x | sonney2k: available? | 00:03 |
---|---|---|
-!- f-x [~user@117.192.219.76] has quit [Read error: Connection reset by peer] | 00:33 | |
CIA-32 | shogun: Sergey Lisitsyn master * r7c7e3de / src/libshogun/preprocessor/Isomap.h : Fix for K-Isomap procedure - http://bit.ly/lYIcVj | 00:46 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 01:02 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 07:04 | |
-!- in3xes [~in3xes@59.163.196.121] has quit [Ping timeout: 240 seconds] | 07:07 | |
-!- in3xes1 [~in3xes@180.149.49.227] has joined #shogun | 08:24 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 08:28 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 08:37 | |
-!- in3xes1 [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 08:40 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 10:03 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 10:36 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 10:41 | |
-!- f-x [~user@117.192.217.184] has joined #shogun | 11:23 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 11:34 | |
-!- f-x [~user@117.192.217.184] has quit [Remote host closed the connection] | 11:38 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 11:38 | |
-!- f-x [~user@117.192.217.184] has joined #shogun | 11:38 | |
-!- heiko [~heiko@main.uni-duisburg.de] has joined #shogun | 12:12 | |
-!- in3xes1 [~in3xes@180.149.49.227] has joined #shogun | 12:19 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 263 seconds] | 12:23 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 12:31 | |
-!- in3xes1 [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 12:34 | |
-!- in3xes1 [~in3xes@59.163.196.121] has joined #shogun | 13:26 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 13:29 | |
blackburn | heiko: hey | 13:48 |
heiko | hi blackburn | 13:53 |
blackburn | heiko: how are you? | 13:53 |
heiko | I am fine, how are you? | 13:53 |
blackburn | fine too :) | 13:53 |
blackburn | there was a small issue with your code | 13:54 |
blackburn | have you seen? | 13:54 |
heiko | not yet, where in? | 13:54 |
blackburn | in string features | 13:54 |
blackburn | https://github.com/shogun-toolbox/shogun/commit/cc04bff4895f290acac3e492d925a8638e7425f6 | 13:54 |
blackburn | I didn't have any other idea how to get it compile | 13:55 |
blackburn | heiko: the only thing I really wonder - it compiled on your machine, right? | 13:55 |
heiko | yes it compiles here | 13:56 |
blackburn | really strange thing | 13:56 |
heiko | what is your error? | 13:56 |
heiko | that subset is protected/private? | 13:56 |
blackburn | the error is that protected | 13:56 |
blackburn | yes | 13:56 |
heiko | which compiler do you use? | 13:56 |
heiko | I use gcc version 4.5.2 (Ubuntu/Linaro 4.5.2-8ubuntu4) | 13:57 |
blackburn | gcc version 4.4.3 (Ubuntu 4.4.3-4ubuntu5) | 13:57 |
blackburn | I guess it is more swig related | 13:57 |
blackburn | because an error was while .cxx files are processed | 13:57 |
heiko | ah ok | 13:58 |
blackburn | I have SWIG Version 1.3.40 | 13:58 |
blackburn | what is you use? | 13:58 |
heiko | I do not compile for swig | 13:58 |
blackburn | hmm | 13:58 |
blackburn | you just compile it with interfaces=libshogun? | 13:58 |
heiko | well I tried yesterday but gave it up because i had some error | 13:58 |
heiko | yes | 13:58 |
blackburn | I see | 13:58 |
heiko | because my machine is quite slow | 13:58 |
heiko | well | 13:59 |
heiko | I will just implement this | 13:59 |
heiko | then the check can go | 13:59 |
blackburn | okay | 13:59 |
blackburn | now it is compiling | 14:00 |
heiko | but it is a strange bug anyway because it should be possible to access protected fields of instances of the same type | 14:01 |
heiko | or am I wring? | 14:01 |
heiko | wroing | 14:01 |
heiko | wrong | 14:01 |
blackburn | yes, it should be ok, but swig fails.. may be it is swig related | 14:01 |
heiko | ok | 14:01 |
heiko | well then, i will do this method now ... | 14:01 |
blackburn | oh, I have realized that yesterday I removed my arpack.h with the whole shogun folder :D | 14:02 |
blackburn | writing it again hehe | 14:02 |
bettyboo | <:*) | 14:02 |
blackburn | I just re-cloned shogun but forgot that I haven't commited arpack.h | 14:03 |
blackburn | repititio est mater studiorum (I hope I remember it right) :) | 14:04 |
heiko | hehe :) | 14:08 |
heiko | oh no did you have to write it all again? | 14:08 |
blackburn | yeap | 14:10 |
blackburn | anyway it was 60% complete | 14:10 |
blackburn | I done it already | 14:10 |
heiko | Just sent a pull request | 14:15 |
blackburn | I'll push it right now | 14:16 |
blackburn | * but first will recall how to do it | 14:16 |
heiko | ok :) | 14:17 |
heiko | yes one has to be carefull | 14:17 |
CIA-32 | shogun: Heiko Strathmann master * r27b9479 / src/libshogun/features/StringFeatures.h : added remove subset in obtain_from_char_features - http://bit.ly/l0ZpQ2 | 14:18 |
CIA-32 | shogun: Heiko Strathmann master * rd10eaf3 / src/libshogun/features/StringFeatures.h : added subset support for obtain_from_char fixing sergey's fixme - http://bit.ly/j8KYb1 | 14:18 |
heiko | something similar happened to me a while ago, but luckily I only wrote interface up to that point :) | 14:18 |
heiko | I always leave my editor open also | 14:18 |
heiko | so I may save in case of such an accident | 14:18 |
blackburn | I'm now trying to use vim :) | 14:18 |
blackburn | what editor do you use? | 14:18 |
heiko | I am lazy: eclipse | 14:18 |
heiko | :) | 14:18 |
bettyboo | crazy | 14:18 |
blackburn | I was using eclipse too | 14:19 |
blackburn | trying to get rid of it now | 14:19 |
heiko | why? | 14:19 |
blackburn | it is pretty have and anyway I don't use it's features | 14:19 |
heiko | I use the code browsing a lot | 14:20 |
heiko | and automatic build/run stuff | 14:20 |
heiko | but thats probably also possible with vim | 14:20 |
heiko | I am just familiar with eclipse because I did a lot of java stuff with it | 14:20 |
blackburn | I was using netbeans for java | 14:21 |
heiko | I think eclipse really rocks for java | 14:21 |
blackburn | I heard idea rocks :) | 14:21 |
blackburn | I have to get some convenient ide for java to september: got an java developer job starting from september | 14:24 |
heiko | ok | 14:24 |
heiko | where? | 14:25 |
blackburn | netcracker corp. | 14:25 |
heiko | what will you do? | 14:25 |
heiko | what do they do? | 14:25 |
blackburn | operational support systems | 14:25 |
blackburn | for telecom. | 14:26 |
blackburn | I don't know exactly | 14:26 |
blackburn | some Java EE stuff | 14:26 |
blackburn | uh.. some roc bug reported | 14:28 |
heiko | afk sorry | 14:30 |
heiko | I will be gone for 30 minutes, see you | 14:33 |
blackburn | see ya | 14:33 |
-!- heiko [~heiko@main.uni-duisburg.de] has quit [Ping timeout: 258 seconds] | 14:37 | |
-!- f-x` [~user@117.192.217.184] has joined #shogun | 14:38 | |
-!- f-x [~user@117.192.217.184] has quit [Remote host closed the connection] | 14:38 | |
CIA-32 | shogun: Sergey Lisitsyn master * r336e167 / (3 files in 2 dirs): Fix for auROC and auPRC - http://bit.ly/mzzG3v | 15:20 |
blackburn | whoaa | 15:20 |
blackburn | fixed | 15:20 |
-!- heiko [~heiko@main.uni-duisburg.de] has joined #shogun | 15:24 | |
CIA-32 | shogun: Heiko Strathmann master * rded723b / (2 files in 2 dirs): | 15:44 |
CIA-32 | shogun: LibLinear now does now not SG_REF labels in appy anymore | 15:44 |
CIA-32 | shogun: (in fashion with other machines) - http://bit.ly/mz4YPM | 15:44 |
CIA-32 | shogun: Sergey Lisitsyn master * r715f2a2 / (2 files in 2 dirs): Merge branch 'master' of git://github.com/karlnapf/shogun - http://bit.ly/jkLwOF | 15:44 |
blackburn | HeikoMachine | 16:01 |
heiko | ? :) | 16:04 |
blackburn | hmm | 16:04 |
blackburn | it is not compiling | 16:04 |
blackburn | make[1]: *** [Clustering_wrap.cxx] Error 1 | 16:05 |
blackburn | ../shogun/features/SparseFeatures.h:585: Error: Syntax error in input(3). | 16:05 |
-!- f-x`` [~user@117.192.217.184] has joined #shogun | 16:05 | |
heiko | eh | 16:05 |
-!- f-x` [~user@117.192.217.184] has quit [Remote host closed the connection] | 16:05 | |
heiko | Ill check | 16:05 |
heiko | mine does | 16:06 |
heiko | but again, no swig here | 16:06 |
blackburn | yes it is swig | 16:06 |
blackburn | will try to find error now | 16:07 |
blackburn | line 585 is a comment hehe | 16:08 |
blackburn | wtf | 16:08 |
heiko | strange thing | 16:09 |
heiko | "syntax error" is also strange | 16:09 |
blackburn | btw it seems your eclipse is using spaces but not tabs | 16:15 |
heiko | no, acutally not. which file? | 16:18 |
blackburn | SparseFeatures.h | 16:19 |
heiko | oh well, i re-installed the computer ..perhaps thats the reason... | 16:19 |
heiko | which line? | 16:19 |
heiko | mostly tabs here | 16:20 |
blackburn | forgot :) may be <585 | 16:20 |
blackburn | I have glanced over that region with vim | 16:20 |
blackburn | there was spaces | 16:20 |
heiko | just made eclipse show white space characters, but its settings are to use tabs | 16:24 |
-!- vojta [~quassel@gw-101.scnet.cz] has joined #shogun | 17:06 | |
-!- vojta [~quassel@gw-101.scnet.cz] has quit [Read error: Connection reset by peer] | 17:06 | |
-!- VojtechFranc [~quassel@gw-101.scnet.cz] has joined #shogun | 17:07 | |
-!- VojtechFranc [~quassel@gw-101.scnet.cz] has quit [Read error: Connection reset by peer] | 17:17 | |
-!- VojtechFranc [~quassel@217.117.216.101] has joined #shogun | 18:45 | |
-!- heiko [~heiko@main.uni-duisburg.de] has quit [Ping timeout: 258 seconds] | 18:46 | |
-!- VojtechFranc [~quassel@217.117.216.101] has quit [Remote host closed the connection] | 19:03 | |
blackburn | ARPACK IS IN! | 19:18 |
serialhex | YAY!!! :P | 19:22 |
bettyboo | hihi | 19:22 |
serialhex | hihi bettyboo! how are you? | 19:22 |
bettyboo | serialhex: hey how are you? | 19:22 |
serialhex | i'm good thnks! | 19:22 |
blackburn | serialhex: how are you? | 19:22 |
serialhex | i'm doing well! | 19:22 |
serialhex | how about yourself?? | 19:23 |
blackburn | I'm happy I almost finished that wrapper issues | 19:23 |
serialhex | sweet!! | 19:23 |
blackburn | yesterday I deleted arpack.h by mistake :D | 19:23 |
serialhex | ooh! oww!!! | 19:23 |
serialhex | that's never good! | 19:23 |
serialhex | what else is new with you blackburn? | 19:25 |
blackburn | serialhex: ah nothing new :) | 19:25 |
-!- in3xes1 is now known as in3xes | 19:26 | |
blackburn | 3000 examples with 1.12s passed! | 19:47 |
CIA-32 | shogun: Sergey Lisitsyn master * re097863 / (src/libshogun/lib/arpack.cpp src/libshogun/lib/arpack.h): Added wrapper for ARPACK - http://bit.ly/mHNPMG | 19:56 |
CIA-32 | shogun: Sergey Lisitsyn master * r3d6a5b5 / src/libshogun/preprocessor/ClassicMDS.cpp : Added ARPACK support for ClassicMDS - http://bit.ly/kTiiv8 | 19:56 |
blackburn | I fucking like it | 19:56 |
blackburn | 3 seconds for 5000 | 19:57 |
blackburn | now isomap is kinda bottleneck | 19:59 |
blackburn | I guess I have to write some dijkstra with fibonacci heap | 20:02 |
-!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 20:07 | |
CIA-32 | shogun: Sergey Lisitsyn master * r964dcbd / src/libshogun/preprocessor/ClassicMDS.cpp : Changed eigenpairs order in ClassicMDS with ARPACK - http://bit.ly/lL3KaR | 20:13 |
blackburn | I have to have some vodka! | 20:14 |
-!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Quit: Page closed] | 20:35 | |
CIA-32 | shogun: Sergey Lisitsyn master * r523a175 / (6 files): Merge branch 'master' of git://github.com/sploving/shogun - http://bit.ly/klyplP | 20:52 |
-!- f-x`` [~user@117.192.217.184] has quit [Ping timeout: 260 seconds] | 21:00 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 21:03 | |
-!- in3xes [~in3xes@59.163.196.121] has quit [Ping timeout: 240 seconds] | 21:06 | |
-!- in3xes_ is now known as in3xes | 21:09 | |
-!- f-x [~user@117.192.194.239] has joined #shogun | 21:36 | |
CIA-32 | shogun: Soeren Sonnenburg master * r884b32b / (6 files in 3 dirs): | 21:52 |
CIA-32 | shogun: Merge pull request #158 from karlnapf/master | 21:52 |
CIA-32 | shogun: subset support for sparse features - http://bit.ly/iuZjvl | 21:52 |
blackburn | Soeren in da house | 21:53 |
blackburn | sonney2k: you want arpack - you got arpack :D | 21:54 |
@sonney2k | blackburn, heh | 22:07 |
@sonney2k | thanks | 22:07 |
@sonney2k | I hope it still works with lame lapack too :) | 22:07 |
blackburn | sonney2k: yes, classic MDS works with any of them | 22:07 |
blackburn | i'll add similar thing to LLE tomorrow | 22:07 |
blackburn | PCA is going to be changed too I guess | 22:08 |
@sonney2k | nice nice | 22:08 |
blackburn | it is really fast as I said above | 22:08 |
blackburn | LMDS for 5000 examples and 200 landmarks did it by 5s | 22:08 |
@sonney2k | I just read the logs | 22:10 |
@sonney2k | sounds nice | 22:10 |
@sonney2k | blackburn, I hope your 'side' job will still leave you a few shogun minutes though - but hey congrats | 22:10 |
blackburn | sonney2k: which side job? | 22:10 |
blackburn | ah | 22:11 |
blackburn | that one I got from september? | 22:11 |
@sonney2k | yes :) | 22:11 |
blackburn | it is half-time, ~4hrs a day, I really hope I will have much time for shogun too | 22:11 |
@sonney2k | blackburn, ...and your studies! | 22:11 |
blackburn | yes studies too :) | 22:12 |
* sonney2k why do I always have to remind you of this | 22:12 | |
blackburn | I don't know | 22:12 |
@sonney2k | ok, so lets see | 22:12 |
@sonney2k | you already did most of my job today | 22:12 |
blackburn | hehe | 22:12 |
@sonney2k | so I can fully concentrate on one patch now | 22:12 |
blackburn | some problem with heiko patch | 22:12 |
@sonney2k | blackburn, still? | 22:13 |
blackburn | another one | 22:13 |
@sonney2k | blackburn, ohh I just merged - so I should probably check for that and fix it | 22:13 |
@sonney2k | blackburn, what is happening? | 22:13 |
blackburn | the issues are appearing because he is compiling it without any modular | 22:14 |
blackburn | swig tells about some syntax error (sic!) | 22:14 |
@sonney2k | syntax error? where? | 22:14 |
blackburn | eh.. let me recall | 22:14 |
blackburn | SparseFeatures | 22:14 |
blackburn | some 583 line or so | 22:16 |
@sonney2k | hmmhh void static? | 22:17 |
@sonney2k | probably static void | 22:17 |
@sonney2k | recompiling | 22:17 |
blackburn | AH! | 22:17 |
blackburn | I really was staring at this line | 22:18 |
blackburn | and seen nothing bad at all | 22:18 |
blackburn | ah yes | 22:18 |
blackburn | let me guess - it is because I do java too | 22:19 |
@sonney2k | now compiles | 22:20 |
CIA-32 | shogun: Soeren Sonnenburg master * rb54d7be / src/libshogun/features/SparseFeatures.h : Fix typo - void static -> should be static void - http://bit.ly/kimWCP | 22:21 |
blackburn | nice | 22:21 |
blackburn | today have glanced over twitter of some scikits developer | 22:31 |
blackburn | he was wondering shogun have no projects with kernels | 22:32 |
f-x | sonney2k: still there? | 22:37 |
@sonney2k | blackburn, no kernels? | 22:48 |
@sonney2k | f-x, yes | 22:48 |
blackburn | sonney2k: kernel-based methods, he was disappointed we do an another general ML library :) | 22:48 |
f-x | sonney2k: i'm facing a bit of a problem with that thing we discussed | 22:50 |
@sonney2k | blackburn, well we are mostly kernel methods right - and I guess the *kernel methods* toolbox | 22:51 |
blackburn | but we have no kernel methods project | 22:51 |
@sonney2k | blackburn, of course we also have large focus | 22:51 |
@sonney2k | and need other infrastructure too | 22:51 |
@sonney2k | yes it is more - world domination :) | 22:52 |
blackburn | I hope next year gsoc we will pass through too | 22:52 |
@sonney2k | f-x, ok please say what the problems are | 22:52 |
blackburn | with more students | 22:52 |
@sonney2k | blackburn, and 5 clones of myself | 22:52 |
f-x | sonney2k: ok. to begin: what object will be used by the parser? | 22:52 |
f-x | as the input source | 22:52 |
blackburn | sonney2k: i'll help you a bit ;) | 22:53 |
@sonney2k | f-x, could you please repeat what the parser does? | 22:53 |
f-x | sonney2k: in short, take input from somewhere, get that input converted into vectors and supply those vectors to whoever needs them (the StreamingFeatures objects) | 22:54 |
f-x | in the previous implementation, the 'input' was through a StreamingFile object | 22:54 |
f-x | sonney2k: it was the job of the StreamingFeatures objects | 22:55 |
f-x | to set what functions the parser should use to read from the StreamingFile | 22:55 |
@sonney2k | f-x, ok | 22:55 |
f-x | (since the reading should be done differently for say SimpleFeatures and StringFeatures) | 22:55 |
f-x | and since StreamingFile was the base class for all Streaming*Files, it was possible to just have a StreamingFile* as a member | 22:56 |
@sonney2k | I totally swapped out the planned structure | 22:57 |
@sonney2k | have SimpleStream, StringStream etc right? | 22:57 |
f-x | sonney2k: yeah, that's what we planned | 22:58 |
f-x | sonney2k: and we didn't specify one common base class for all these *Stream objects (which would replace the StreamingFile stuff) | 22:58 |
@sonney2k | f-x, ok so since these will already return parsed vectors etc these need the parser(s) | 22:58 |
@sonney2k | f-x, yeah but that was the feature | 22:59 |
@sonney2k | I mean one common base class would need these many methods implemented | 22:59 |
f-x | sonney2k: then the structure of the parser gets changed in many places | 23:00 |
f-x | how should i replace that StreamingFile object? | 23:00 |
f-x | sonney2k: and with a common base class, we could just implement the dummy versions of the functions in the base class, and reimplement only the ones which apply to our case in the derived classes | 23:01 |
f-x | and whenever someone calls the function from our derived class, the functions defined in the derived class are called | 23:01 |
@sonney2k | f-x, but how does havng a single base class help? | 23:03 |
@sonney2k | I mean you need only few functions for the respective feature types | 23:03 |
@sonney2k | so you only need to implement those for that particular feature type interface | 23:04 |
@sonney2k | and you also need a parser just for this feature type | 23:04 |
@sonney2k | it doesn't help to have it for all types there | 23:04 |
@sonney2k | this means streamingfile is removed | 23:05 |
@sonney2k | with no replacement | 23:05 |
@sonney2k | and instead you have SimpleStream etc | 23:05 |
f-x | sonney2k: wait.. i guess it | 23:05 |
f-x | will be helpful for me if you could just open lib/InputParser.h | 23:06 |
@sonney2k | done | 23:06 |
f-x | line ~247 | 23:06 |
f-x | near the protected member declarations | 23:06 |
f-x | sonney2k: i should throw away that input_source object? | 23:07 |
@sonney2k | ok | 23:07 |
@sonney2k | f-x, I think I start to see your point now | 23:07 |
@sonney2k | StreamingFile is the stream-centered view | 23:08 |
f-x | sonney2k: exactly, not type-centered | 23:08 |
@sonney2k | so you could have a file or network socket or whatever stream | 23:08 |
f-x | correct | 23:08 |
@sonney2k | and based on which is used it will stream from there | 23:08 |
f-x | right | 23:08 |
f-x | StreamingFile is at the top of all other classes | 23:08 |
f-x | Below that, we may have Streaming{Ascii,Binary}File, | 23:09 |
f-x | StreamingSocket or whatever | 23:09 |
f-x | sonney2k: and check out the definitions of the set_read_vector function just below the class definition | 23:09 |
f-x | line ~275 | 23:10 |
f-x | the StreamingFeatures object sets the "read_vector" function to be any member function of the StreamingFile class... | 23:10 |
f-x | that function pointer is class-dependent and parameter-dependent | 23:10 |
f-x | i mean, it will work only with a function that comes directly or indirectly from a StreamingFile class | 23:11 |
@sonney2k | I understand - stream centric it should be done exactly like it is now | 23:13 |
f-x | sonney2k: here, we implement all the required vector reading functions (which read from an ascii file) in StreamingAsciiFile, and let the Features object choose which one it wants to use | 23:15 |
@sonney2k | f-x, yes | 23:16 |
@sonney2k | so if you do streamingfilefromfeatures it will only define the ones that make sense | 23:18 |
@sonney2k | or simplefeatures even | 23:18 |
@sonney2k | streamingfilefromsimplefeatures | 23:19 |
f-x | sonney2k: StreamingFileFromSimpleFeatures? | 23:19 |
f-x | yes | 23:19 |
@sonney2k | yes | 23:19 |
f-x | only say get_simple_int_vector() and all those things | 23:19 |
f-x | then StreamingSimpleFeatures | 23:19 |
f-x | sets the read function (for the parser) to be get_simple_*_vector() | 23:19 |
f-x | depending on type T in StreamingSimpleFeatures<T> | 23:19 |
@sonney2k | f-x, yes makes sense | 23:20 |
f-x | sonney2k: so if done in this manner, the user will never use say get_string_*_vector() for any StreamingSimpleFeatures object, thus never encountering those unimplemented (dummy) functions | 23:20 |
blackburn | mid-term is veryyyyy sooon | 23:20 |
@sonney2k | so it makes a lot of sense to not have StreamingFile with all pure virtual functions but all of them just return SG_ERROR("not supported by interface") or so like you do | 23:21 |
@sonney2k | err stream | 23:21 |
f-x | sonney2k: exactly. that's what i thought.. | 23:22 |
@sonney2k | nto supported by stream | 23:22 |
@sonney2k | not | 23:22 |
@sonney2k | blackburn, fasten your seatbelts | 23:22 |
f-x | yes, yes! now i've conveyed my point :) | 23:22 |
blackburn | in fact I've done everything :D | 23:22 |
@sonney2k | f-x, so it doesn't make sense to have SimpleStream etc but just StreaminSImpleFeatures etc | 23:23 |
@sonney2k | f-x, so fine so good | 23:23 |
f-x | sonney2k: for the time being, otherwise we have to make drastic changes | 23:24 |
@sonney2k | then I can only agree to this design | 23:24 |
blackburn | today jonas behr was very surprised with our shogun™ troubleshooting service© :D | 23:24 |
@sonney2k | blackburn, it has never been better ;-) | 23:24 |
blackburn | sonney2k: I forgot to change area_under_curve for column-major | 23:25 |
blackburn | and he got 0.45 auROC for some task :D | 23:25 |
f-x | sonney2k: so i just proceed with whatever i'd done before? | 23:25 |
@sonney2k | f-x, I recall that we wanted to avoid some #define'd code with that other design - anyway | 23:25 |
blackburn | auROC=0.45 is a pretty bad random decision I guess, right? | 23:25 |
@sonney2k | f-x, what is your current pull request about? | 23:25 |
f-x | sonney2k: the StreamingAsciiFile stuff; StreamingFileFromSimpleFeatures etc | 23:26 |
@sonney2k | f-x, I mean the streamfromstimplefeatures is still new right? | 23:26 |
f-x | yes, new as in not yet in the master repository | 23:26 |
@sonney2k | f-x, ok - I have a few comments before I can merge this | 23:26 |
f-x | sonney2k: go ahead | 23:27 |
@sonney2k | blackburn, yeah - I read the emails. looks like anti learning :) | 23:27 |
@sonney2k | f-x, please use SG_PRINT (or SG_SPRINT if not within SG_OBJECT) to do the printf / SG_WARNING/SG_ERROR respectively) | 23:27 |
@sonney2k | f-x, then you still have void foo(void) | 23:28 |
@sonney2k | please use void foo(); | 23:28 |
f-x | sonney2k: yeah.. i haven't updated them.. that void thing was there in one of the standard shogun files.. any reason for it over there? | 23:28 |
@sonney2k | f-x, I am fixing the void foo(void) thing whereever I find it | 23:29 |
f-x | sonney2k: ok. i'll do it in mine too | 23:29 |
@sonney2k | feel free to do so too | 23:29 |
f-x | sonney2k: sure | 23:29 |
@sonney2k | f-x, then the multi-line string | 23:30 |
@sonney2k | inseat of your | 23:30 |
@sonney2k | + SG_INFO("Call to unimplemented vector read function!\n"); \ | 23:30 |
@sonney2k | 53 | 23:30 |
@sonney2k | + SG_INFO("This means this function is not appropriate "); \ | 23:30 |
@sonney2k | 54 | 23:30 |
@sonney2k | + SG_INFO("for the type of feature you are working with,"); \ | 23:30 |
@sonney2k | 55 | 23:30 |
@sonney2k | + SG_INFO("Or the corresponding reader isn't implemented.\n"); \ | 23:30 |
@sonney2k | 1 | 23:30 |
@sonney2k | please use SG_ERROR("foo" | 23:30 |
@sonney2k | "bar" | 23:30 |
@sonney2k | "bla"); | 23:30 |
f-x | sonney2k: much better | 23:30 |
@sonney2k | mlsec, bettyboo is no longer an op? | 23:30 |
bettyboo | sonney2k: I saw. maybe my swig_type_map.i. I write the ujmp typemap. so it is still have the error. i need have a check my typemap | 23:31 |
* sonney2k is happy to not be kicked due too flood | 23:31 | |
f-x | sonney2k: and before i forget | 23:31 |
f-x | (unless you had further suggestions about my pull request first) | 23:31 |
blackburn | bettyboo: would you marry serialhex? | 23:32 |
bettyboo | blackburn: will you marry | 23:32 |
bettyboo | never mind | 23:32 |
f-x | I think an OnlineMachine has to be made for all these online algorithms | 23:32 |
@sonney2k | f-x, no that's it | 23:32 |
serialhex | WTF!?!?!? | 23:32 |
serialhex | you trying to get me married off blackburn??? | 23:32 |
blackburn | serialhex: sorry, but she is nice! | 23:32 |
@sonney2k | serialhex, if knives won't do - woman will | 23:32 |
@sonney2k | f-x, yes | 23:33 |
@sonney2k | makes sense | 23:33 |
serialhex | she is nice! and knives will never be as nice as a woman!!! | 23:33 |
@sonney2k | f-x, but note that some of them could still do batch training | 23:33 |
f-x | sonney2k: either directly derived from CMachine or maybe even independent | 23:33 |
blackburn | I suggest to create HeikoMachine for some fancy things | 23:33 |
@sonney2k | f-x, you just need to introduce a new function to the streamingfile class - reset_stream (or so to seek to the beginning) | 23:34 |
f-x | sonney2k: that would be nice, but i'm finding myself needing to create the online SGD as another class altogether | 23:34 |
@sonney2k | blackburn, VodkaMachine is missing too | 23:34 |
@sonney2k | f-x, why? | 23:34 |
blackburn | not VodkaMachine, AlcoMachine | 23:34 |
@sonney2k | f-x, the calibrate thing cannot be the issue | 23:34 |
f-x | ah! yes, i remember that.. asked john for some suggestions regarding that too | 23:34 |
blackburn | VodkaMachine is a simple structured learning base class | 23:34 |
@sonney2k | f-x, I mean jus use the few N examples to do calibration and then start learning | 23:35 |
f-x | sonney2k: but that won't work always, i think | 23:35 |
f-x | so we'd be skipping the examples we calibrate from? (for training) | 23:35 |
CIA-32 | shogun: Sergey Lisitsyn master * r35c6ef6 / (2 files): Fix uppercase naming of MDSs - http://bit.ly/ignxoX | 23:36 |
@sonney2k | f-x, sure - but that is how the batch SGD works - and it is an assumption that the examples come in random order | 23:36 |
@sonney2k | f-x, or buffer them | 23:36 |
@sonney2k | f-x, but calibration is done on a very small portion of examples usually | 23:37 |
f-x | sonney2k: in the batch SGD, the calibration is done over the whole set | 23:37 |
@sonney2k | say 1000 | 23:37 |
f-x | (or until some condition is matched) | 23:37 |
f-x | okay.. c=0 in the beginning | 23:38 |
@sonney2k | so it doesn't matter if we skip them | 23:39 |
@sonney2k | f-x, what would an online machine do? | 23:39 |
@sonney2k | I mean which interfaces would it provide? | 23:39 |
f-x | sonney2k: probably the same stuff, but there'd be one difference | 23:40 |
f-x | CLabels* may or may not be removed; that's secondary | 23:40 |
f-x | but in say CLinearMachine | 23:40 |
@sonney2k | I currently don't see which extra features are needed | 23:40 |
f-x | which operates on CDotFeatures* | 23:40 |
@sonney2k | yes? | 23:40 |
f-x | we'll need it to work with StreamingDotFeatures | 23:40 |
f-x | which isn't derived from DotFeatures | 23:41 |
f-x | it only has the FP_DOT property which relates them | 23:41 |
@sonney2k | f-x, yeah that is a problem | 23:41 |
f-x | sonney2k: we could have a workaround | 23:42 |
@sonney2k | so one would need Machine -> OnlineMachine -> OnlineLinearMachine,OnlineKernelMachine or so | 23:42 |
f-x | sonney2k: that would do | 23:42 |
@sonney2k | f-x, which? | 23:42 |
f-x | hmm one minute | 23:42 |
f-x | sonney2k: no sorry, that assumed we had a separate class for online sgd | 23:43 |
f-x | forget about that | 23:43 |
f-x | sonney2k: and if we make the Online* classes that way, what would CSVMSGD derive from? | 23:44 |
f-x | (if we wish to make it both batch- and online-compatible?) | 23:44 |
@sonney2k | OnlineLinearMacine | 23:44 |
@sonney2k | look it is easy | 23:44 |
@sonney2k | SGD could then check if the reset_stream_to_beginning() function is available | 23:45 |
@sonney2k | and do batch learning | 23:45 |
@sonney2k | (if enabled) | 23:45 |
f-x | then our features object would be of type CFeatures*, not CDotFeatures* ? | 23:45 |
@sonney2k | f-x, always Streaming*Features | 23:45 |
@sonney2k | for SGD and other online methods | 23:46 |
f-x | sonney2k: okay, so the user first makes the Streaming*Features object from the corresponding CDotFeatures object if he wants to use it | 23:46 |
f-x | or we have a constructor which enables that | 23:46 |
f-x | within SGD | 23:47 |
@sonney2k | f-x, no | 23:47 |
@sonney2k | the user uses StreamingFileFromSimpleFeatures etc | 23:47 |
f-x | sonney2k: exactly. | 23:47 |
f-x | or the SGD has a constructor taking a CDotFeatures as argument, which does all that by itself | 23:48 |
f-x | (the user won't have to instantiate a StreamingFileFromSimpleFeatures explicitly) | 23:48 |
@sonney2k | the other way round is not possible unfortunately, i.e. we cannot turn a streaming feature object into a batch one (number of vectors is just not known) | 23:48 |
@sonney2k | f-x, but that would be messy in SGD | 23:49 |
f-x | sonney2k: yeah, in general | 23:49 |
@sonney2k | because then you need to write SGD twice | 23:49 |
@sonney2k | once for dotfeatures and once for streamingdotfeatures | 23:49 |
@sonney2k | then better have SGD(CSimpleFeatures<float64_t>* feat) | 23:50 |
blackburn | sonney2k: what is 'sg' analogue for shogun-data? | 23:50 |
blackburn | ssh://git@shogun-toolbox.org/shogun-data? | 23:50 |
@sonney2k | or can we translate general dot features into StreamingF*Features? | 23:50 |
f-x | sonney2k: not yet | 23:50 |
f-x | but i think it may be possible | 23:50 |
f-x | but not something to depend on right now | 23:51 |
@sonney2k | f-x, hmmhh I guess that won't work. I mean you need a concrete feature vector right | 23:51 |
@sonney2k | f-x, and dot features can be strings underneath and returning the (dense) feature representation would be a waste | 23:52 |
@sonney2k | f-x, so better later do it | 23:52 |
f-x | sonney2k: seems like there isn't a 'general' way to do it | 23:52 |
f-x | at least now | 23:52 |
@sonney2k | blackburn, yes | 23:53 |
@sonney2k | f-x, wait there is - if you just map the dotfeature operations to streamingdotfeatures | 23:54 |
f-x | typecast? | 23:54 |
f-x | sonney2k: sorry i'll need an example | 23:54 |
@sonney2k | f-x, no - a new class that does the same as streamingfilefromsimplefeatures | 23:54 |
@sonney2k | but without ever returning a concrete vectro | 23:55 |
@sonney2k | vector | 23:55 |
f-x | sonney2k: StreamingFileFromDotFeatures? | 23:55 |
@sonney2k | it just does these add_to_dense_vec etc stuff | 23:55 |
@sonney2k | yes | 23:55 |
CIA-32 | shogun: Sergey Lisitsyn master * rd79f50b / examples/undocumented/python_modular/graphical/swissroll.py : Added swissroll python_modular graphical example - http://bit.ly/iiz35I | 23:55 |
@sonney2k | so it relies on the algorithm underneath to only ever use the supplied operations | 23:55 |
blackburn | git push does nothing in data.. | 23:56 |
@sonney2k | blackburn, did you git add ? | 23:56 |
f-x | sonney2k: that could just work | 23:56 |
blackburn | sonney2k: i did git add and git commit | 23:56 |
blackburn | but git push origin tells me 'everything is up-to-date' | 23:56 |
@sonney2k | f-x, hmmhh could StreamingFileFromSimpleFeatures go then? | 23:58 |
@sonney2k | blackburn, let me update | 23:58 |
@sonney2k | blackburn, hmmhh says up-to-date | 23:59 |
blackburn | git push does nothing hmm | 23:59 |
@sonney2k | blackburn, you do the commit form within data? | 23:59 |
blackburn | yes | 23:59 |
f-x | sonney2k: it could go, yes | 23:59 |
blackburn | it appears in git log | 23:59 |
--- Log closed Fri Jul 01 00:00:28 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!