--- Log opened Fri Jun 24 00:00:16 2011 | ||
@sonney2k | mlsec, blackburn was asking who mlsec is - that's it. | 00:00 |
---|---|---|
@mlsec | okay | 00:02 |
@mlsec | thx for the info. gotta slepp. 120% admin load each day at tu | 00:02 |
@sonney2k | mlsec, only 100% play load at daytime here | 00:02 |
@sonney2k | mlsec, see you soon old buddy! ETA 4.7! | 00:03 |
f-x | sonney2k: i still have to call a StreamingFeatures::function to set the vector readers from within StreamingSimpleFeatures right? | 00:28 |
f-x | because i can't call set_read_vector directly from the StreamingFeatures constructor | 00:28 |
@sonney2k | f-x, why not? | 00:28 |
f-x | because it is defined in StreamingSimpleFeatures.. the linker says my constructor cannot call a pure virtual function | 00:29 |
@sonney2k | f-x, hmmhh maybe that is a c++ limitation. if you don't make it pure virtual but add print statements in both classes - will the derived one be called? | 00:31 |
f-x | sonney2k: nope.. i don't think so | 00:31 |
f-x | because during construction the base class knows only about its own members, i guess | 00:32 |
@sonney2k | f-x, indeed - explanation here http://www.artima.com/cppsource/nevercall.html | 00:32 |
f-x | sonney2k: yeah.. exactly that.. | 00:32 |
f-x | so i can define a function say set_read_functions() in StreamingFeatures, which calls set_vector_reader and set_vector_and_label_reader of the derived class | 00:33 |
f-x | and call set_read_functions from streamingsimplefeatures | 00:33 |
@sonney2k | dangerous problem... | 00:33 |
@sonney2k | f-x, yes | 00:34 |
f-x | or some different version of init() itself, where we may later add more things | 00:34 |
f-x | as we generalize things more | 00:34 |
@sonney2k | yup | 00:35 |
f-x | okay, putting it under 'set_read_functions' for the time being | 00:36 |
f-x | sonney2k: req sent | 00:40 |
CIA-32 | shogun: Soeren Sonnenburg master * rec1eb8e / (7 files in 2 dirs): Merge branch 'streaming' of git://github.com/frx/shogun (+6 more commits...) - http://bit.ly/l0ovWs | 00:42 |
@sonney2k | thx | 00:50 |
f-x | sonney2k: you're welcome :) | 00:51 |
@sonney2k | f-x, now finish these features and get sgd done and feel happy :) | 00:52 |
@sonney2k | now == next steps | 00:52 |
f-x | sonney2k: yeah.. StringFeatures should be ready now | 00:53 |
f-x | but there's one thing: | 00:53 |
f-x | other than the basic get_* stuff, i'm including the remap thing | 00:53 |
f-x | the remapping is done (if desired) during each call to get_next_example() | 00:54 |
f-x | say remap is from alpha1->alpha2, how do we define 'alphabet' for the object at each step? | 00:54 |
f-x | should i be clearer? i guess i'm being vague... | 00:55 |
@sonney2k | alphabet needs to be set once beforehand (in constructor for example) | 00:56 |
f-x | yes, alpha1 | 00:56 |
f-x | right? | 00:57 |
f-x | say i need to remap from DNA to RAWDNA | 00:57 |
f-x | the alphabet set will be DNA | 00:57 |
f-x | and after every call to get_next_example, the histogram of the alphabet will be updated | 00:57 |
@sonney2k | yes | 00:58 |
@sonney2k | (or every char in example will be checkt | 00:59 |
@sonney2k | ed) | 00:59 |
f-x | and if i want to remap, there will be the RAWDNA alphabet whose histogram will be updated at the same time | 00:59 |
@sonney2k | no stop | 00:59 |
@sonney2k | the remap is not that flexible | 00:59 |
@sonney2k | it maps only from alpha -> raw | 00:59 |
@sonney2k | one cannot tranfer DNA to digits for example | 01:00 |
@sonney2k | or vice versa | 01:00 |
@sonney2k | so it is just 'A' -> 0 etc | 01:00 |
f-x | oh.. okk | 01:01 |
f-x | well i think things are becoming clearer as i ask.. is there any other function you want added in StreamingStringFeatures? | 01:02 |
@sonney2k | as I said only some get_string of the raw string :) | 01:05 |
@sonney2k | with optional translation / error checking | 01:05 |
f-x | converted to the required T type char-by-char? | 01:06 |
@sonney2k | f-x, for T types other than char like e.g. int it should read binary ints... | 01:06 |
@sonney2k | not a priority though now | 01:06 |
@sonney2k | I would just do it for chars for now | 01:07 |
@sonney2k | all the extras can come later | 01:07 |
f-x | okay, so just returning the raw string will do for now | 01:07 |
@sonney2k | yes | 01:07 |
-!- f-x [~user@117.192.214.250] has quit [Remote host closed the connection] | 01:33 | |
CIA-32 | shogun: Soeren Sonnenburg master * r16c3af6 / src/modular/Features.i : import preprocessors in features too - http://bit.ly/mrVfAR | 02:48 |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 10:37 | |
blackburn | sonney2k: hey I didn't | 10:42 |
blackburn | @sonney2k ml*sec, blackburn was asking who mlsec is - that's it. | 10:42 |
blackburn | ah I wrote his nickname, sorry Konrad :) | 10:54 |
-!- sploving [~root@124.16.139.196] has joined #shogun | 11:03 | |
sploving | anyone? | 11:04 |
blackburn | sploving: yes? | 11:06 |
sploving | when I test python, like python *.py, it shows could not find module shogun.Features | 11:07 |
sploving | blackburn, do you know how to set the lib path? | 11:07 |
blackburn | sploving: nope.. it fails with python_modular interface configured? | 11:08 |
sploving | blackburn, I use ./configure --interfaces=libshogun,python_modular | 11:08 |
sploving | it works | 11:08 |
sploving | then "make" | 11:08 |
sploving | do i need make install? | 11:09 |
blackburn | of course | 11:09 |
sploving | blackburn, in java, it does not need | 11:10 |
blackburn | hmm | 11:10 |
sploving | blackburn, it works! | 11:10 |
blackburn | heh | 11:10 |
sploving | it disturb me a long time | 11:11 |
sploving | then check.sh works most of "OK" | 11:12 |
sploving | I am glad to see the result! | 11:12 |
sploving | but some error | 11:13 |
blackburn | sploving: it is ok, we broke many things with ST* -> SGVector transition | 11:16 |
blackburn | three days ago there were ERROR for almost every test with some OK | 11:17 |
sploving | classifier_lda_modular.py OK | 11:17 |
sploving | blackburn, does that mean I install ATLAS well? | 11:18 |
blackburn | sploving: yes, it works only with atlas | 11:18 |
sploving | blackburn, but when ./configure, it shows ATLAS no. | 11:18 |
sploving | i think the configure need fix | 11:19 |
blackburn | well try to python preprocessor_locallylinearembedding_modular.py | 11:19 |
blackburn | it exactly requires atlas, I wrote it :) | 11:19 |
bettyboo | :^) | 11:19 |
sploving | it print "locallylinearembedding" | 11:20 |
sploving | does that mean well? | 11:20 |
blackburn | yes | 11:21 |
blackburn | it will fail if there was no atlas | 11:21 |
sploving | blackburn, now I try to play with tester.py generator.py | 11:21 |
sploving | I need write a similar java such as tester.java generator.java | 11:22 |
blackburn | I see | 11:22 |
sploving | what is generate_testdata.py used for? | 11:24 |
sploving | do I need run it before run tester.py? blackburn | 11:24 |
blackburn | sploving: I don't know yet | 11:25 |
blackburn | I never used it before | 11:25 |
sploving | blackburn, you never used testsuite? | 11:26 |
blackburn | yes :D | 11:26 |
bettyboo | :^) | 11:26 |
sploving | oh. | 11:26 |
blackburn | sploving: this testsuite is just some kind of 'wrapper' for test in examples | 11:27 |
blackburn | btw it is not in actual state now | 11:28 |
sploving | blackburn, you mean it does not work now? | 11:29 |
blackburn | yes, some things to be fixed here | 11:32 |
sploving | blackburn, No module named PreProc | 11:37 |
sploving | when I test_all.sh | 11:37 |
blackburn | it should be Preprocessor | 11:37 |
sploving | So these examples need updated? | 11:38 |
blackburn | I think it is in testsuite sources | 11:41 |
blackburn | examples are ok | 11:41 |
blackburn | something is wrong with serialization | 11:44 |
sploving | it is in test_one.py | 11:44 |
sploving | import preproc should be import preprocessor | 11:45 |
sploving | no it is in Kernel.py | 11:46 |
sploving | it said "from shogun.PreProc import *" | 11:47 |
blackburn | sploving: i'm fixing it now | 11:48 |
sploving | yeap | 11:48 |
sploving | one is kernel.py the other is test_one.py | 11:48 |
blackburn | sploving: I guess we need some supersonney here | 11:55 |
sploving | I agree | 11:56 |
blackburn | I will commit fixes for get it working but not everything is working now | 11:56 |
sploving | there always some error: not fond Preproc | 11:56 |
CIA-32 | shogun: Sergey Lisitsyn master * r8de57a2 / (8 files in 2 dirs): Fixes for python testsuite - http://bit.ly/mjrXrJ | 11:59 |
blackburn | sploving: should be ok now | 12:00 |
sploving | I will try. | 12:00 |
blackburn | now tests fails, not testsuite :D | 12:01 |
sploving | why? | 12:01 |
sploving | you should fix consider two sides not just onw | 12:01 |
sploving | one | 12:01 |
sploving | I gtg bye blackburn | 12:03 |
blackburn | bye | 12:03 |
sploving | looking forward to your news later! | 12:04 |
-!- sploving [~root@124.16.139.196] has left #shogun [] | 12:04 | |
blackburn | hehe | 12:04 |
@sonney2k | blackburn, around? | 12:34 |
blackburn | yeap | 12:34 |
@sonney2k | blackburn, I have a problem doing the dotfeatures T** transition | 12:34 |
@sonney2k | it fails in gaussiannaivebayes now | 12:35 |
@sonney2k | and I am not sure how to properly fix it | 12:35 |
@sonney2k | should I just comment this part and commit and you do the fix? | 12:35 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Read error: Connection reset by peer] | 12:36 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 12:36 | |
blackburn | sorry got disconnected | 12:36 |
@sonney2k | <sonney2k> blackburn, I have a problem doing the dotfeatures T** transition | 12:37 |
@sonney2k | <sonney2k> it fails in gaussiannaivebayes now | 12:37 |
@sonney2k | <sonney2k> and I am not sure how to properly fix it | 12:37 |
@sonney2k | <sonney2k> should I just comment this part and commit and you do the fix? | 12:37 |
blackburn | sonney2k: I will fix it in a minute | 12:37 |
@sonney2k | blackburn, wait you cannot | 12:37 |
blackburn | why? | 12:37 |
@sonney2k | I will comment and commit such that you know | 12:37 |
blackburn | ah | 12:38 |
blackburn | well there are some things to be fixed here | 12:38 |
blackburn | actually it is for simple features | 12:38 |
blackburn | sonney2k: you are talking about if at line 54 GaussianNaiveBayes.cpp, rgiht? | 12:39 |
@sonney2k | blackburn, please wait | 12:39 |
blackburn | ok | 12:39 |
@sonney2k | alesis-novik, I need to touch GMM too | 12:43 |
@sonney2k | and Gaussian | 12:43 |
@sonney2k | blackburn, btw I still like alesis-novik idea of having that do free flag set appropriately at all times quite a lot :) | 12:46 |
bettyboo | :^) | 12:46 |
blackburn | sonney2k: I have no idea about that so just do it like you want :) | 12:50 |
blackburn | sonney2k: ah I didn't tell you: new LMDS is really gorgeous | 12:54 |
CIA-32 | shogun: Soeren Sonnenburg master * r9d78a61 / (8 files in 6 dirs): T** transition for DotFeatures' get_feature_vector and matrix - http://bit.ly/lX3cvz | 12:54 |
CIA-32 | shogun: Soeren Sonnenburg master * r9e02b38 / (8 files in 2 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/lmqpBK | 12:54 |
blackburn | 15000 3d vectors was embedded to 2d space with 13s taken | 12:54 |
@sonney2k | blackburn, :) | 12:55 |
blackburn | with arpack and when I will write it better it could be even faster with factor of ~1.5 | 12:55 |
@sonney2k | very nice | 12:56 |
blackburn | actually I don't know why we need ClassicMDS - it will never finish on 15K examples | 12:58 |
blackburn | just like I will never finish reading Alberts' 1725 pages Molecular Biology of The Cell book | 13:01 |
blackburn | sonney2k: hey it is not compiling now | 13:11 |
blackburn | Distance_wrap.cxx: In function ‘void* _p_shogun__CPruneVarSubMeanTo_p_shogun__CSGObject(void*, int*)’: | 13:11 |
blackburn | Distance_wrap.cxx:55762: error: ‘CPruneVarSubMean’ is not a member of ‘shogun’ | 13:11 |
blackburn | and so on | 13:11 |
blackburn | for every preprocessor | 13:11 |
CIA-32 | shogun: Sergey Lisitsyn master * rdd0ff30 / (2 files): Removed ignored = operator at RandomFourierGaussPreproc - http://bit.ly/kr93Jy | 13:14 |
blackburn | sonney2k: ! | 14:01 |
CIA-32 | shogun: Sergey Lisitsyn master * r1b78b3a / (2 files): Fixes for GaussianNaiveBayes - http://bit.ly/msTjKd | 14:11 |
CIA-32 | shogun: Sergey Lisitsyn master * r88edc41 / (7 files in 2 dirs): Added LandmarkIsomap and fixed Isomap and LandmarkMDS - http://bit.ly/kEJP4g | 14:12 |
@sonney2k | blackburn, no idea | 14:29 |
@sonney2k | blackburn, seems like I can fix it by adding %include's in the modular if's | 14:34 |
@sonney2k | no idea why it is suddenly there though | 14:34 |
@sonney2k | I didn't do any changes there - at least I am not aware of any... | 14:34 |
blackburn | me too :( | 14:40 |
blackburn | sonney2k: strange thing about it is they all are including Features_includes.i | 14:42 |
blackburn | and preprocessors are included there | 14:43 |
CIA-32 | shogun: Soeren Sonnenburg master * r5a32d4d / src/libshogun/kernel/Kernel.cpp : add missing cases to kernels - http://bit.ly/l3iM74 | 14:46 |
CIA-32 | shogun: Soeren Sonnenburg master * re4b52f0 / (8 files): include preprocs in all modular interfaces - http://bit.ly/jWOWyT | 14:46 |
@sonney2k | blackburn, well swig does not include recursively | 14:46 |
CIA-32 | shogun: Soeren Sonnenburg master * ref42f49 / (11 files in 3 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/iz7ulZ | 14:46 |
@sonney2k | blackburn, it can with the includeall directive | 14:46 |
blackburn | strange.. | 14:46 |
@sonney2k | but this generates other problems I wasn't able to resolve | 14:46 |
blackburn | I don't understand why it was emerged | 14:47 |
@sonney2k | anyway I put the include for preprocs everywhere and now it compiles | 14:47 |
@sonney2k | me neither | 14:47 |
@sonney2k | me fiddling with dotfeatures* certainly shouldn't | 14:47 |
blackburn | exactly | 14:47 |
blackburn | https://github.com/shogun-toolbox/shogun/commit/16c3af6ff7d46104d4a50f61e3d09638d67fbb2e | 14:48 |
blackburn | sonney2k: I guess this caused | 14:48 |
@sonney2k | blackburn, nope | 14:48 |
@sonney2k | I removed that and it still didn't compile here | 14:48 |
blackburn | well in this case I completely have no idea | 14:49 |
@sonney2k | I even did git clean again to check | 14:49 |
@sonney2k | anyway we need that include there for splovings stuff | 14:49 |
blackburn | sonney2k: btw sploving have commit rights too now? | 14:50 |
@sonney2k | blackburn, no | 14:50 |
@sonney2k | why? | 14:50 |
blackburn | I was wondering wtih | 14:50 |
blackburn | sploving merged commit 16c3af6 into shogun-toolbox:master from sploving:master about 5 hours ago | 14:50 |
@sonney2k | blackburn, he is probably not using rebase | 14:52 |
blackburn | I see | 14:53 |
blackburn | I need to invent (n^2) dijkstra analogue heheh | 14:58 |
blackburn | *not dijkstra, floyd-warshall | 14:58 |
@sonney2k | seems like sparse features work nicely now | 15:02 |
@sonney2k | blackburn, the only big issue I have no idea why it appeared is to get the multi-class classifiers back to work | 15:02 |
@sonney2k | I really wonder why the apply() function is not visible... | 15:03 |
blackburn | yes, I remind | 15:03 |
blackburn | lapack was not working | 15:03 |
blackburn | errrr | 15:03 |
blackburn | larank | 15:03 |
blackburn | hehe | 15:03 |
bettyboo | :> | 15:03 |
@sonney2k | blackburn, btw sploving still has no lapack/atlas | 15:03 |
blackburn | sonney2k: he was running tests today | 15:04 |
@sonney2k | his atlas is not compiled with -lm and libs are missing | 15:04 |
blackburn | lda was working | 15:04 |
blackburn | and my LLE too | 15:04 |
@sonney2k | not tests - examples and yeah these fail gracefully IIRC when not installed | 15:04 |
blackburn | it won't even import without lapack | 15:04 |
@sonney2k | are you sure? | 15:04 |
blackburn | examples, yes | 15:04 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 15:04 | |
blackburn | sonney2k: yes, LDA is not working without lapack | 15:05 |
blackburn | and I'm sure LLE too because it uses dsymv :) | 15:05 |
@sonney2k | blackburn, I know not working - but the check* stuff just runs or? | 15:05 |
blackburn | in test for LLE lapack routines will be called exactly | 15:06 |
blackburn | I don't know how can it say there are no atlas and works at the same time | 15:06 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 276 seconds] | 15:07 | |
@sonney2k | blackburn, works != python check.sh doesn't crash | 15:07 |
@sonney2k | *sigh* java modular does no longer compile - wtf?! | 15:07 |
blackburn | no idea :) | 15:07 |
@sonney2k | I wish github had some build bot | 15:08 |
blackburn | sonney2k: won't it crash if there are no lapack and dsymv routine is called?! | 15:08 |
@sonney2k | blackburn, it is all ifdef'd remember? | 15:08 |
blackburn | sonney2k: all class LocallyLinearEmbedding is ifdef'd | 15:09 |
@sonney2k | exactly | 15:09 |
blackburn | it won't be imported | 15:09 |
@sonney2k | blackburn, I mean just try | 15:09 |
blackburn | ah ok | 15:09 |
@sonney2k | --disable-lapack | 15:09 |
@sonney2k | blackburn, his configure with --enable-lapack failed for a good reason | 15:09 |
@sonney2k | at least | 15:09 |
blackburn | sonney2k: compiling.. I believe it will fail :) | 15:14 |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 15:14 | |
blackburn | on import | 15:14 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 15:14 | |
@sonney2k | yeah it should | 15:16 |
@sonney2k | but check.sh - will that just say OK? | 15:17 |
blackburn | should be error I guess | 15:17 |
-!- in3xes_ is now known as in3xes | 15:19 | |
@sonney2k | blackburn, well let me know :D | 15:20 |
* sonney2k is f*** with this java crap | 15:20 | |
blackburn | oh damn tropic tolyatti hehe rainy-sunny-rainy-sunny-rainy-sunny | 15:23 |
bettyboo | :*) | 15:23 |
blackburn | running classifier_lda_modular.py .. ERROR | 15:24 |
@sonney2k | blackburn, samer here | 15:24 |
@sonney2k | weather wise | 15:25 |
blackburn | strange summer | 15:25 |
blackburn | last summer was so warm that 30% of forest there burned | 15:25 |
blackburn | sonney2k: my preprocessors fails, lda too | 15:26 |
@sonney2k | blackburn, then I have no idea what sploving sent | 15:26 |
@sonney2k | his configure failed badly | 15:26 |
blackburn | sonney2k: this morning he tried and it was working on his machine | 15:26 |
@sonney2k | with all sorts of errors | 15:26 |
@sonney2k | and with these it just cannot work | 15:27 |
@sonney2k | no idea | 15:27 |
blackburn | and he wants me to fix testsuite hehe | 15:27 |
blackburn | sonney2k: | 15:27 |
blackburn | (02:01:03 PM) blackburn: now tests fails, not testsuite :D | 15:27 |
blackburn | (02:01:15 PM) sploving: why? | 15:27 |
blackburn | (02:01:45 PM) sploving: you should fix consider two sides not just onw | 15:27 |
@sonney2k | I cannot parse this language | 15:28 |
blackburn | I'm feared he will abuse me if I won't fix tests :( | 15:28 |
@sonney2k | java is really broken now - and I don't have the slightest clue why | 15:30 |
blackburn | in fact shogun is really broken now :D | 15:30 |
@sonney2k | blackburn, python_modular is not or? | 15:31 |
blackburn | it compiles yes | 15:31 |
blackburn | libshogunui too | 15:31 |
blackburn | I mean we will fix it for a long :) | 15:31 |
blackburn | with ST* transition and other things | 15:31 |
@sonney2k | blackburn, it not only compiles but almost all examples run | 15:32 |
blackburn | well are you sure examples spread over all the functionality? | 15:32 |
@sonney2k | blackburn, yes | 15:33 |
blackburn | hmm then it is ok | 15:34 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 15:35 | |
@sonney2k | blackburn, well no - we need to get the examples running first | 15:35 |
@sonney2k | then run the actual test suite | 15:35 |
@sonney2k | if that passes - with valgrind running and no memmory leaks etc - then it is all good | 15:35 |
blackburn | sonney2k: seems only multiclass SVMs is failing now | 15:35 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 15:36 | |
@sonney2k | blackburn, and we don't know why it fails | 15:36 |
@sonney2k | classify is in kernel machine | 15:36 |
@sonney2k | err apply | 15:36 |
@sonney2k | but it for some reason is not seen in the derived class | 15:36 |
@sonney2k | s/seen/visible/ | 15:37 |
@sonney2k | here also the serialization tests fail | 15:37 |
blackburn | serialization is not working | 15:37 |
@sonney2k | because of add_preproc :) | 15:37 |
blackburn | ah | 15:37 |
blackburn | this morning I fixed some lines at testsuite | 15:38 |
@sonney2k | ahh and because of multiclass | 15:38 |
blackburn | tried to run, and there was some serialization issues | 15:38 |
@sonney2k | (it uses GMNPSVM -> multiclass) | 15:38 |
blackburn | ahaha | 15:40 |
blackburn | found in some work's references: | 15:40 |
blackburn | Anonymous Author. Anonymous title | 15:40 |
@sonney2k | blackburn, serialization for labels is broken | 15:40 |
blackburn | why it is? ST*? | 15:40 |
@sonney2k | blackburn, I guess heikos subset patch | 15:40 |
* sonney2k checks | 15:41 | |
blackburn | ah | 15:41 |
blackburn | in the case won't try to fix it - don't understand what heiko is doing :D | 15:41 |
@sonney2k | blackburn, unfortunately heiko's computer died | 15:41 |
blackburn | bad | 15:41 |
* sonney2k is looking at it | 15:41 | |
-!- heiko [865b7525@gateway/web/freenode/ip.134.91.117.37] has joined #shogun | 15:41 | |
blackburn | just in time, Heiko :) | 15:42 |
heiko | hi | 15:42 |
heiko | :) | 15:42 |
bettyboo | :> | 15:42 |
heiko | whats up? | 15:42 |
blackburn | heiko: we just noticed that labels serialization is not working | 15:42 |
heiko | oh, ok | 15:42 |
blackburn | heiko: what's with your computer? | 15:43 |
heiko | I will have a look on this | 15:43 |
@sonney2k | I see what the problem is | 15:43 |
@sonney2k | heiko, serialization support in labels is broken | 15:43 |
@sonney2k | heiko, I guess due to the num_labels_total / num_labels thingy | 15:43 |
heiko | well, its old and the harddisc had an eror in the partition table or something, I am currently re-installing | 15:43 |
heiko | ok | 15:43 |
heiko | I will check it as soon as possible | 15:44 |
@sonney2k | heiko, shouldn't m_parameters->add_vector(&labels, &num_labels_total, "labels" | 15:44 |
@sonney2k | err stop | 15:44 |
@sonney2k | I think we should also serialzie num_labels then - or is that no longer needed (known from subset?!) | 15:45 |
blackburn | heiko: better not use old hard drives - check it's S.M.A.R.T for bad blocks | 15:45 |
blackburn | if <>0 - it will increase for sure, better buy new :) | 15:46 |
heiko | well yes I am planing to buy a new laptop some day when I have money ;) | 15:46 |
heiko | the num_abels thing | 15:47 |
heiko | I think num_labels_total has to be used since we want to serialize the complete array | 15:47 |
blackburn | I guess you will have some at 17-19th of July :D | 15:47 |
@sonney2k | heiko, agreed | 15:47 |
@sonney2k | but what about the other variables later? | 15:47 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 15:47 | |
@sonney2k | they are not serialized and so all 0 when labels is loaded | 15:47 |
@sonney2k | e.g. num_labels = 0 then | 15:48 |
heiko | blackburn, yes :) but I will have to keep the money to pay study fees for my master | 15:48 |
heiko | sonney2k, yes you are right that is the poblem | 15:48 |
blackburn | heiko: I see, well just buy new hard drive for your laptop - it is cheap | 15:48 |
heiko | blackburn, yes, I will think of it :) | 15:48 |
@sonney2k | heiko, can't we do this via the subset classsss | 15:49 |
blackburn | losing data is FFUUUU | 15:49 |
blackburn | :D | 15:49 |
@sonney2k | I mean never having two num_labels things? | 15:49 |
heiko | sonney2k, yes, the subset class serializes its array and size anyways | 15:49 |
@sonney2k | heiko, so I can remove num_labels right? | 15:49 |
heiko | you mean, only have one variable for the size? | 15:50 |
@sonney2k | yes | 15:50 |
heiko | ah ok, yes this is possible | 15:50 |
@sonney2k | because subset will give the subset size | 15:50 |
heiko | rue | 15:50 |
heiko | true | 15:50 |
heiko | some methods might have to be changed then but I think this might work | 15:50 |
@sonney2k | ok I am checking for num_labels* stuff | 15:50 |
heiko | blackburn, I do regular backups :) | 15:50 |
heiko | ok, alright | 15:50 |
heiko | perhaps remove num_labels_total | 15:51 |
heiko | because this was the one that was added by me | 15:51 |
heiko | same thing for string features by the way | 15:51 |
heiko | do we have somehing like tests for serialization? | 15:52 |
alesis-novik | sonney2k, what do you need to do with Gaussian and GMM? | 15:52 |
@sonney2k | alesis-novik, I did already - look at the diff | 15:52 |
@sonney2k | some SGVector changes | 15:52 |
@sonney2k | hmmhh | 15:55 |
@sonney2k | heiko, I think you need to fix the subset business for get labels | 15:55 |
@sonney2k | I will just make this test work without caring too much about subsets | 15:55 |
heiko | ok, I will have a look at all the suset stuff | 15:56 |
alesis-novik | sonney2k, so what would happen if I made a pull request now? | 15:56 |
@sonney2k | I'd like to see all tests pass for python_modular now | 15:56 |
@sonney2k | alesis-novik, make sure you do git pull --rebase | 15:57 |
-!- blackburn [~blackburn@31.28.40.202] has quit [Quit: Leaving.] | 16:01 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 16:01 | |
-!- blackburn [~blackburn@31.28.40.202] has quit [Client Quit] | 16:01 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 16:01 | |
CIA-32 | shogun: Soeren Sonnenburg master * r8ca4620 / (4 files in 3 dirs): fix the serialization examples and Labels' serialization - http://bit.ly/iyJXxZ | 16:12 |
@sonney2k | heiko, I think you only have to modify get_num_labels() - I would write labels like this: if you have a subset object set it uses this number / subsetting | 16:13 |
@sonney2k | if not then it uses things directly | 16:13 |
@sonney2k | the other option is to always have a subset object that is just the whole set - that would work too but might be confusing | 16:14 |
heiko | I would prefer the first version | 16:14 |
@sonney2k | me too | 16:15 |
@sonney2k | in addition these confidence things in labels can go | 16:15 |
@sonney2k | no one uses these so far so no need to make things overly complex | 16:15 |
heiko | ok, alright | 16:15 |
alesis-novik | sonney2k, I hope you are up for some commenting on formatting and other problems :) | 16:18 |
bettyboo | ;) | 16:19 |
@sonney2k | alesis-novik, totally totally ;D | 16:19 |
@sonney2k | alrighty | 16:19 |
@sonney2k | all examples run through (python_modular) again :)) | 16:19 |
@sonney2k | heiko, btw please use SG_UNREF() / SG_REF for the subset - this way you can even just return the whole subset object | 16:41 |
@sonney2k | no need for wrappers then | 16:41 |
-!- f-x [~user@223.190.80.31] has joined #shogun | 16:53 | |
@sonney2k | alesis-novik, a monster patch! | 16:53 |
@sonney2k | it is soo big that I cannot really check it | 16:54 |
@sonney2k | alesis-novik, I will just merge - do the minor change please | 17:11 |
@sonney2k | btw it would be great if you could fully use SGVector & friends | 17:11 |
@sonney2k | instead of these ptr / size thingies | 17:11 |
@sonney2k | then for serialization you can just call m_parameters->add(&vec) etc | 17:12 |
CIA-32 | shogun: Soeren Sonnenburg master * r8ace2c2 / (6 files in 3 dirs): Merge branch 'gmm' of git://github.com/alesis/shogun (+8 more commits...) - http://bit.ly/mJe9mA | 17:13 |
alesis-novik | sonney2k, that's the plan. | 17:13 |
@sonney2k | ok great | 17:13 |
@sonney2k | I've merged as is - but please do the overhaul then :) | 17:13 |
alesis-novik | and the n_ thing was because I saw it somewhere else :) | 17:13 |
bettyboo | <:*) | 17:13 |
alesis-novik | sonney2k, overhaul as in the SG* thing? | 17:14 |
@sonney2k | n_ thingy is just awful wherever it occurs | 17:14 |
@sonney2k | alesis-novik, with overhaul - yes | 17:15 |
heiko | sonney2k, I have to go now, but compuer is ready, I will impement all the stuff we talked about and send a patch tommorrow ok? | 17:15 |
@sonney2k | heiko, yes thanks! | 17:16 |
heiko | alright, then see you! | 17:16 |
@sonney2k | cu! | 17:16 |
-!- heiko [865b7525@gateway/web/freenode/ip.134.91.117.37] has quit [Quit: Page closed] | 17:16 | |
-!- f-x [~user@223.190.80.31] has quit [Quit: ERC Version 5.3 (IRC client for Emacs)] | 17:19 | |
CIA-32 | shogun: Soeren Sonnenburg master * r267bed1 / (4 files in 2 dirs): Merge branch 'streaming' of git://github.com/frx/shogun - http://bit.ly/jSAIDt | 17:20 |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 17:52 | |
blackburn | fuck.. | 17:55 |
blackburn | my keyoard =( | 17:59 |
-!- alesis-novik [~alesis@188.74.87.206] has quit [Quit: biab] | 18:18 | |
-!- f-x [~user@27.57.156.173] has joined #shogun | 19:19 | |
@sonney2k | blackburn, it works! | 20:04 |
blackburn | what works? | 20:05 |
@sonney2k | examples | 20:05 |
blackburn | all? | 20:05 |
@sonney2k | yes | 20:05 |
blackburn | nice | 20:05 |
@sonney2k | I will now valgrind / check the static interfaces | 20:06 |
blackburn | I have ()roked my key()oard :D | 20:07 |
blackburn | and now I cant type a letter | 20:08 |
blackburn | hmm I will type it like 6 | 20:08 |
blackburn | 6aseclass lol | 20:13 |
-!- f-x [~user@27.57.156.173] has quit [Read error: Connection reset by peer] | 22:14 | |
-!- f-x [~user@27.61.146.235] has joined #shogun | 22:15 | |
@sonney2k | blackburn, around? | 23:17 |
blackburn | sonney2k: yes | 23:17 |
@sonney2k | blackburn, I think I now know why java_modular doesn't compile | 23:17 |
blackburn | why | 23:18 |
blackburn | ? | 23:18 |
@sonney2k | problem is that we have DotFeatures and there a SGVector<float64_t> get_feature_vector() is defined | 23:18 |
@sonney2k | (not virtual so local to that class) | 23:18 |
blackburn | in simple features too? | 23:18 |
@sonney2k | now we have the same function in simplefeautres but with SGVector<ST> as return arg | 23:19 |
@sonney2k | -> clash | 23:19 |
@sonney2k | in java for some reason | 23:19 |
blackburn | I see | 23:19 |
blackburn | do we really need it in DotFeatures? | 23:19 |
@sonney2k | blackburn, the issue is that some dotfeatures don't provide a get_feature_vector | 23:19 |
@sonney2k | and indeed it is not needed as one can just call add_to_dense_vec and that will give one the feature vector | 23:20 |
blackburn | may 6e just do it virtual? | 23:20 |
@sonney2k | blackburn, this won't work for sure | 23:21 |
blackburn | damn it is pretty 6ad to type without 6 | 23:21 |
blackburn | stupid me | 23:21 |
@sonney2k | I mean then we definitely have e.g. SGVector<float64_t> in ther and SGVector<bool> get_feature_vector(int) | 23:21 |
blackburn | ah I see | 23:21 |
@sonney2k | so we need a rename | 23:21 |
blackburn | SGVector<float64_t> get_feature_vector() | 23:22 |
blackburn | what it returns? | 23:22 |
@sonney2k | how about compute_feature_vector? | 23:22 |
@sonney2k | int | 23:22 |
blackburn | with int? | 23:22 |
blackburn | aha | 23:22 |
@sonney2k | I mean you give it the vector number | 23:22 |
@sonney2k | and it returns the vector | 23:22 |
blackburn | yes I understood | 23:22 |
@sonney2k | blackburn, such problem could not occur when we had functions like void get_feature_vector(some_type_here) | 23:23 |
@sonney2k | but returning different types is a problem | 23:23 |
@sonney2k | so what do you think? rename to compute_* | 23:24 |
blackburn | well just do it compute_* | 23:24 |
blackburn | yes | 23:24 |
@sonney2k | the only problem is that it is not so obvious that this will return something | 23:25 |
@sonney2k | get_* is a better name for that | 23:25 |
blackburn | get_dot_feature_vector? | 23:26 |
@sonney2k | but hey - on th eother side get_* sounds like it is already there | 23:26 |
@sonney2k | or get_computed_* | 23:26 |
blackburn | get_computed - like it | 23:26 |
@sonney2k | get_computed_dot_feature_vector | 23:26 |
* sonney2k is pondering | 23:27 | |
blackburn | get_computed_dot_feature_hello_mom_vector | 23:27 |
@sonney2k | I guess I will have to change 1 million classes | 23:27 |
@sonney2k | *_my_ass_* :) | 23:27 |
blackburn | why? is it used widely? | 23:27 |
@sonney2k | blackburn, for some reason some people use this (including yourself :) | 23:28 |
blackburn | sonney2k: I just wondering where it used | 23:28 |
blackburn | in SVMs? | 23:28 |
@sonney2k | no | 23:29 |
blackburn | I removed get_feature_vector from gaussiannaive6ayes | 23:29 |
blackburn | using feature_matrix now | 23:29 |
@sonney2k | but seriously I would call it get_computed_dot_feature_vector and get_computed_dot_feature_matrix | 23:29 |
blackburn | 6ayes haha | 23:30 |
-!- blackburn [~blackburn@31.28.40.202] has left #shogun [] | 23:30 | |
-!- blackburn [~blackburn@31.28.40.202] has joined #shogun | 23:30 | |
@sonney2k | blackburn, yes much better (well use get_feature_vector or get_feature_matrix -> but be aware that sometimes a feature matrix does not exist) | 23:30 |
blackburn | sonney2k: I hope no6ody will use it for such features | 23:31 |
@sonney2k | blackburn, I made the name disgusting enough now | 23:32 |
@sonney2k | so I dont' think anyone will | 23:32 |
blackburn | okay :) | 23:33 |
blackburn | hmm what to do next.. | 23:33 |
@sonney2k | gaussiannaivebayes uses it | 23:33 |
blackburn | sonney2k: uses what? | 23:33 |
@sonney2k | sgdqn | 23:34 |
@sonney2k | GMM | 23:34 |
@sonney2k | gaussian | 23:34 |
@sonney2k | that's it | 23:35 |
@sonney2k | compiling for all the 100 languages now | 23:35 |
blackburn | nice | 23:36 |
@sonney2k | did I say we need a build bot? | 23:36 |
@sonney2k | some machine with 64 cores or so | 23:36 |
blackburn | ahah | 23:36 |
blackburn | is it possile? | 23:36 |
@sonney2k | too bad that we don't get 5 per student as organization | 23:36 |
@sonney2k | then we could buy sth like that :) | 23:37 |
@sonney2k | java built | 23:37 |
blackburn | sonney2k: will you receive 2.5K for us? | 23:37 |
@sonney2k | the university will | 23:37 |
@sonney2k | not me | 23:38 |
blackburn | I mean you as organization | 23:38 |
@sonney2k | I will receive gray hair :D | 23:38 |
blackburn | hahah | 23:38 |
blackburn | sonney2k: which one university? | 23:39 |
@sonney2k | tu berlin | 23:39 |
blackburn | is it highly related to tu 6erlin? | 23:40 |
blackburn | sonney2k: 6tw do you do some teaching activities? | 23:40 |
@sonney2k | well our machine learnign department | 23:41 |
blackburn | *I guess I will receive new key6oard in 1-2 weeks* | 23:41 |
@sonney2k | so yes | 23:41 |
@sonney2k | teaching yes | 23:42 |
blackburn | what is do you teach? :) | 23:42 |
@sonney2k | some ML :) | 23:43 |
blackburn | I see ;) | 23:43 |
@sonney2k | large scale learning, optimization, bioinformatics, clustering, kernels ... | 23:43 |
@sonney2k | such stuff | 23:43 |
blackburn | you could encourage your students to improve our little shogunny | 23:43 |
@sonney2k | not my students though | 23:44 |
@sonney2k | the ones of our prof ... | 23:44 |
blackburn | well anyway you can suggest somethig :) | 23:44 |
blackburn | if you won't have time just suggest them to ask drunk russian for some tasks | 23:45 |
blackburn | :D | 23:45 |
blackburn | I'm lazy a6out writing 6alltree and spectral clustering | 23:45 |
blackburn | may 6e some other student will :D | 23:46 |
CIA-32 | shogun: Soeren Sonnenburg master * r65ac9bf / (13 files in 8 dirs): rename dot features' get_feature_vector/matrix functions to avoid naming clash - http://bit.ly/kiyyxz | 23:58 |
--- Log closed Sat Jun 25 00:00:18 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!