--- Log opened Fri Jul 15 00:00:21 2011 | ||
-!- blackburn [~blackburn@188.122.238.99] has quit [Quit: Leaving.] | 00:07 | |
-!- f-x [~user@117.192.194.58] has quit [Ping timeout: 260 seconds] | 00:34 | |
-!- in3xes_ [~in3xes@180.149.49.230] has joined #shogun | 07:00 | |
-!- in3xes_ [~in3xes@180.149.49.230] has quit [Quit: Leaving] | 07:38 | |
[1]warpy | morning | 08:31 |
---|---|---|
[1]warpy | anyone here yet ? | 08:31 |
-!- gsomix [~gsomix@85.26.232.4] has joined #shogun | 08:41 | |
@sonney2k | [1]warpy, yes sure :) | 08:45 |
@sonney2k | saw your email - it is a good starting point | 08:46 |
CIA-32 | shogun: Baozeng Ding master * ra615b6e / (8 files): add remaining kernel examples - http://bit.ly/n9YeIT | 08:57 |
CIA-32 | shogun: Soeren Sonnenburg master * rf511a42 / (8 files): | 08:57 |
CIA-32 | shogun: Merge pull request #199 from sploving/master | 08:57 |
CIA-32 | shogun: add remaining kernel java examples - http://bit.ly/oh4mLl | 08:57 |
[1]warpy | thanks | 09:01 |
[1]warpy | email me back if you need a followup | 09:01 |
[1]warpy | also daniel is doing some work this weekend | 09:01 |
@sonney2k | [1]warpy, I don't have much time today - so it would be great if daniel could just get the vector / array stuff to work. matrices and all that are then probably much more easy. | 09:06 |
-!- sploving1 [~sploving@124.16.139.134] has joined #shogun | 09:10 | |
sploving1 | hi sonney2k | 09:10 |
@sonney2k | hi sploving1 | 09:10 |
@sonney2k | do the examples work for you? | 09:10 |
@sonney2k | work as in compile at least | 09:11 |
CIA-32 | shogun: Soeren Sonnenburg master * r98d5021 / (2 files): fix compile errors due to *kernel SGVector changes - http://bit.ly/rd31RG | 09:11 |
sploving1 | javac works | 09:11 |
sploving1 | some examples, java not work. | 09:12 |
@sonney2k | which don't work? | 09:12 |
@sonney2k | and did you now translate all or which are still missing? | 09:12 |
sploving1 | still working. maybe miss about some. | 09:13 |
sploving1 | for exmaple, structure_dynprog_modular.py | 09:14 |
sploving1 | I do not know how to translate it. as it use scipy a lot | 09:14 |
@sonney2k | sploving1, yeah ignore that | 09:14 |
sploving1 | another thing, maybe other people recently add some examples, I have not add them in time now | 09:15 |
sploving1 | there are 13 sparse examples also not added | 09:16 |
@sonney2k | but these you can add | 09:16 |
sploving1 | no. java does not support it now | 09:17 |
@sonney2k | you can write sth like | 09:17 |
@sonney2k | x=SparseRealFeatures() | 09:17 |
@sonney2k | x.obtain_from_dense(y) | 09:17 |
@sonney2k | doesn't matter | 09:17 |
@sonney2k | you can just write | 09:18 |
@sonney2k | feat=RealFeatures(fm_test_real); | 09:18 |
@sonney2k | feats_test=SparseRealFeatures(); | 09:18 |
@sonney2k | feats_test.obtain_from_simple(feat); | 09:18 |
sploving1 | oh. I should add others first. | 09:18 |
@sonney2k | which others? | 09:19 |
sploving1 | some missing. I am still working now | 09:19 |
-!- gsomix [~gsomix@85.26.232.4] has quit [Ping timeout: 255 seconds] | 09:20 | |
sploving1 | serialization_svmlight_modular like this | 09:20 |
@sonney2k | yeah please do so | 09:21 |
-!- sploving1 [~sploving@124.16.139.134] has left #shogun [] | 09:22 | |
-!- gsomix [~gsomix@85.26.232.4] has joined #shogun | 09:22 | |
-!- sploving1 [~sploving@124.16.139.134] has joined #shogun | 09:22 | |
sploving1 | sonney2k, SerializableHdf5File has the same meaning as pickle.dump? | 09:30 |
[1]warpy | bbl | 09:30 |
-!- [2]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has joined #shogun | 09:31 | |
-!- [1]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has quit [Ping timeout: 255 seconds] | 09:35 | |
-!- [2]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has quit [Ping timeout: 276 seconds] | 09:37 | |
CIA-32 | shogun: Baozeng Ding master * r899890c / examples/undocumented/java_modular/serialization_svmlight_modular.java : add serialization_svmlight_modular - http://bit.ly/roMVm5 | 09:41 |
CIA-32 | shogun: Soeren Sonnenburg master * r150cddb / examples/undocumented/java_modular/serialization_svmlight_modular.java : | 09:41 |
CIA-32 | shogun: Merge pull request #200 from sploving/master | 09:41 |
CIA-32 | shogun: add serialization example - http://bit.ly/mQhAEA | 09:41 |
-!- heiko [~heiko@main.uni-duisburg.de] has joined #shogun | 09:42 | |
@sonney2k | hi heiko :) | 09:44 |
heiko | hi sonney2k :) | 09:44 |
heiko | no sleeping at all? :) | 09:44 |
@sonney2k | sleep-what? | 09:45 |
heiko | hehe :) | 09:45 |
bettyboo | interesting | 09:45 |
heiko | indeed bettyboo | 09:46 |
bettyboo | heiko: in the classifier_averaged_perceptron_modular.py example, why parameter_list = ((traindat,testdat,label_traindat,1.,1000,1),(traindat,testdat,label_traindat,1.,1000,1)) | 09:46 |
@sonney2k | heiko, now that you example works - what are your next plans? | 09:46 |
heiko | oh, one little exaple works | 09:46 |
heiko | but i still have to put together the other ones | 09:46 |
heiko | especially the grid-search | 09:47 |
heiko | but this should be straight-forward | 09:47 |
heiko | so, as for future plans | 09:47 |
heiko | I still have the list: more subset classes, kernel machine sv stuff, ModelSelection Parameters generic | 09:47 |
@sonney2k | ok and what else? | 09:48 |
@sonney2k | I mean anything else you can think of? | 09:48 |
-!- warpy2 [~androirc@bzq-79-181-19-104.red.bezeqint.net] has quit [Ping timeout: 252 seconds] | 09:48 | |
heiko | one more thing is the seralization version stuff, but the gsoc project is ready then :) | 09:49 |
heiko | do you have anything in mind? | 09:49 |
heiko | I mean, still possible: | 09:49 |
sploving1 | sonney2k, fstream = SerializableHdf5File("blaah.h5", "w"). no such .java class? | 09:49 |
sploving1 | why? | 09:50 |
heiko | more modelSelection algos, gradient descent, bisection | 09:50 |
@sonney2k | heiko, no I am just asking what you woudl be interested in doing later | 09:50 |
heiko | more splitting strategies | 09:50 |
@sonney2k | sploving1, you probably dont' have hdf5 installed | 09:50 |
@sonney2k | heiko, ok... | 09:50 |
@sonney2k | heiko, one thing that comes to my mind is of course easy illustrative examples | 09:50 |
@sonney2k | heiko, and even a more compact way to write things at least for the python side | 09:51 |
sploving1 | sonney2k, anotherthing, SerializableAsciiFile fstream = new SerializableAsciiFile("blaah.asc", 'w'); svm.save_serializable(fstream); but class SerializableAsciiFile have no method save_serializable | 09:51 |
heiko | yes more real-life examples would be nie | 09:51 |
heiko | what do you mean with a more compact way? | 09:51 |
sploving1 | GMNPSVM have no method save_serializable | 09:52 |
@sonney2k | heiko, the example you have is long and complex already - hard to digest for outsiders, so sth like { 'C' : [1,2,3,], 'kernel' : [PowerKernel ...] would be easier to understand | 09:53 |
@sonney2k | I guess that can be done with python dictionaries and lists | 09:53 |
heiko | ah ok, know what you mean | 09:53 |
@sonney2k | and so one could write a typemap that translates this dictionary/list into ModelSelectionParameters | 09:54 |
heiko | yes, this would be quite cool | 09:54 |
@sonney2k | heiko, this is actually not even so difficult once one has a way to formulate it with dictionaries/lists | 09:54 |
heiko | ok, sounds good | 09:55 |
heiko | probably makes sense | 09:55 |
heiko | the examples are very long indeed | 09:55 |
@sonney2k | heiko, I guess when you do just an example where you select the C of an SVm it would be much easier to digest | 09:56 |
heiko | yes, but sonney2k, the current example is only an example for a complex parameter tree | 09:56 |
heiko | i am building one where only the C is selected, this is much shorter an easier | 09:57 |
heiko | have a look at the create_param_tree in libshogun/modelselection_grid_search_simple.cpp example | 09:57 |
-!- alesis-novik [~alesis@188.74.87.206] has joined #shogun | 09:58 | |
-!- gsomix [~gsomix@85.26.232.4] has quit [Ping timeout: 252 seconds] | 10:15 | |
sploving1 | Exception in thread "main" java.lang.UnsatisfiedLinkError: org.shogun.PreprocessorJNI.new_SortUlongString()J | 10:20 |
sploving1 | at org.shogun.PreprocessorJNI.new_SortUlongString(Native Method) | 10:20 |
sploving1 | at org.shogun.SortUlongString.<init>(Unknown Source) | 10:20 |
sploving1 | at preprocessor_sortwordstring_modular.run(preprocessor_sortwordstring_modular.java:36) | 10:20 |
sploving1 | at preprocessor_sortwordstring_modular.main(preprocessor_sortwordstring_modular.java:58) | 10:20 |
sploving1 | when I compile "java preprocessor_sortwordstring_modular " | 10:20 |
heiko | sonney2k, how to checkout shogun-data again? forgot it :) | 10:23 |
bettyboo | rotfl | 10:23 |
sploving1 | oh. I forgot loadLibrary("Preprocessor") | 10:25 |
-!- gsomix [~gsomix@85.26.235.71] has joined #shogun | 10:28 | |
heiko | sonney2k, ok forgot it, i found it in the mailing list :) | 10:30 |
-!- sploving1 [~sploving@124.16.139.134] has left #shogun [] | 10:37 | |
-!- warpy2 [~androirc@bzq-218-206-2.red.bezeqint.net] has joined #shogun | 10:53 | |
warpy2 | back | 10:55 |
warpy2 | anyone here? | 10:55 |
@sonney2k | heiko, OK | 11:05 |
heiko | sonney2k, grid-search example runs :) | 11:05 |
@sonney2k | very good | 11:10 |
heiko | is there any real-life data in data? | 11:11 |
heiko | because accuracy is 1.-0 for fm_train_real | 11:11 |
@sonney2k | heiko, you could generate some overlapping gaussians | 11:12 |
@sonney2k | actually some example is already doing this | 11:12 |
heiko | another qeustion: | 11:12 |
@sonney2k | classifier_libsvm_minimal_modular.py | 11:12 |
@sonney2k | does this | 11:12 |
heiko | most of the examples define a function which takes parameters | 11:12 |
heiko | do i have to do this? | 11:12 |
heiko | or can I just construct a script? | 11:12 |
warpy2 | soeren do i need the data repo for running examples? | 11:13 |
heiko | warpy2 yes you do | 11:16 |
heiko | for most of them | 11:16 |
-!- warpy2 [~androirc@bzq-218-206-2.red.bezeqint.net] has quit [Remote host closed the connection] | 11:17 | |
heiko | sonney2k, just thought of another features: really cool would be using multiple threads for the grid search, | 11:17 |
@sonney2k | heiko, please do it in the function way - these will be automagic tests | 11:21 |
heiko | oh ok | 11:21 |
heiko | then wait with the merge of my pull request :) | 11:21 |
heiko | also with this parameter list? | 11:22 |
@sonney2k | heiko, yes | 11:24 |
heiko | sonney2k, pushed | 11:29 |
CIA-32 | shogun: Soeren Sonnenburg master * r8567b11 / (10 files in 4 dirs): | 11:33 |
CIA-32 | shogun: Merge pull request #201 from karlnapf/master | 11:33 |
CIA-32 | shogun: working grid-search example in python_modular (+9 more commits...) - http://bit.ly/nJs0U6 | 11:33 |
@sonney2k | heiko, looks great | 11:34 |
heiko | sonney2k, thanks :) finally it runs :) | 11:34 |
CIA-32 | shogun: Baozeng Ding master * rffb86ed / (11 files): add remaining examples - http://bit.ly/p1HlKd | 11:48 |
CIA-32 | shogun: Soeren Sonnenburg master * r747718f / (11 files): | 11:48 |
CIA-32 | shogun: Merge pull request #202 from sploving/master | 11:48 |
CIA-32 | shogun: add all the remaining examples - http://bit.ly/qLg7kk | 11:48 |
-!- gsomix [~gsomix@85.26.235.71] has quit [Ping timeout: 255 seconds] | 12:13 | |
-!- gsomix [~gsomix@85.26.234.14] has joined #shogun | 12:25 | |
-!- Sabrina [~sabrina@124.16.139.134] has joined #shogun | 12:29 | |
-!- Sabrina [~sabrina@124.16.139.134] has left #shogun [] | 12:29 | |
-!- gsomix [~gsomix@85.26.234.14] has quit [Ping timeout: 260 seconds] | 12:53 | |
CIA-32 | shogun: Heiko Strathmann master * rb454101 / (2 files): some work/interface changes towards computing confidence intervals - http://bit.ly/ovbPTz | 12:56 |
CIA-32 | shogun: Heiko Strathmann master * rae9e30f / (2 files in 2 dirs): apllied interface changes in other cross-validation class - http://bit.ly/pqUe7U | 12:56 |
CIA-32 | shogun: Heiko Strathmann master * r11771cc / examples/undocumented/libshogun/modelselection_grid_search_simple.cpp : applied interface changes in cross-validation class - http://bit.ly/nUe4Sa | 12:56 |
CIA-32 | shogun: Soeren Sonnenburg master * raf60362 / (5 files in 4 dirs): | 12:56 |
CIA-32 | shogun: Merge pull request #203 from karlnapf/master | 12:56 |
CIA-32 | shogun: some work towards computing confidence intervals and more precise cross-validation - http://bit.ly/q3DhSZ | 12:56 |
-!- sploving1 [~sploving@124.16.139.134] has joined #shogun | 13:06 | |
sploving1 | features_string_file_modular: cannot apply org.shogun.BaseFeatures : load(org.shogun.SWIGTYPE_p_shogun__CFile) to (org.shogun.AsciiFile) | 13:07 |
sploving1 | sonney2k, can you fix it? | 13:07 |
@sonney2k | looking at it... | 13:09 |
sploving1 | sonney2k, another thing: public HistogramWordStringKernel(int size, SWIGTYPE_p_shogun__CPluginEstimate pie) | 13:10 |
sploving1 | it should be public HistogramWordStringKernel(int size, PluginEstimate pie) | 13:10 |
sploving1 | sonney2k, another thing, I think KernelPCACut(SWIGTYPE_p_shogun__CKernel k, double thresh) should be KernelPCACut(Kernel k, double thresh). | 13:21 |
sploving1 | whant does this error mean: number of symbols in observation (512) larger than M (64) | 13:39 |
@sonney2k | sploving1, does classifier_averaged_perceptron_modular.java compile for you? | 14:19 |
@sonney2k | mine here says | 14:19 |
@sonney2k | classifier_averaged_perceptron_modular.java:7: cannot find symbol | 14:19 |
@sonney2k | symbol: class test | 14:19 |
@sonney2k | public class classifier_averaged_perceptron_modular implements test { | 14:19 |
@sonney2k | ^ | 14:19 |
@sonney2k | kernel_histogram_word_string_modular.java:45: cannot find symbol | 14:19 |
-!- cwidmer [~quassel@connect.tuebingen.mpg.de] has joined #shogun | 14:22 | |
-!- f-x [~user@117.192.197.185] has joined #shogun | 14:37 | |
-!- blackburn [~blackburn@188.122.238.99] has joined #shogun | 14:37 | |
blackburn | have I missed anything? ;) | 14:37 |
@sonney2k | blackburn, yes - more includes! | 14:37 |
blackburn | more includes?? | 14:38 |
-!- in3xes_ [~in3xes@210.212.58.111] has joined #shogun | 14:39 | |
blackburn | one would better include everything | 14:39 |
CIA-32 | shogun: Soeren Sonnenburg master * rdb900c3 / (7 files): fix org.shogun.SWIGTYPE_p_shogun__CFile issue in Features - http://bit.ly/n9Y2OK | 14:40 |
CIA-32 | shogun: Soeren Sonnenburg master * r97083bb / examples/undocumented/java_modular/Load.java : fix warnings in Load.java - http://bit.ly/ouBAWR | 14:40 |
CIA-32 | shogun: Soeren Sonnenburg master * r1d5de60 / (5 files in 4 dirs): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/n1KkSn | 14:40 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 258 seconds] | 14:41 | |
@sonney2k | sploving1, what are all these Note: distribution_histogram_modular.java uses unchecked or unsafe operations. | 14:42 |
@sonney2k | Note: Recompile with -Xlint:unchecked for details. | 14:42 |
@sonney2k | errors? | 14:42 |
CIA-32 | shogun: Soeren Sonnenburg master * rc28e5bf / (3 files): | 14:45 |
CIA-32 | shogun: Merge pull request #204 from sploving/master | 14:45 |
CIA-32 | shogun: fix some bug - http://bit.ly/nlyDAM | 14:45 |
@sonney2k | blackburn, the downside of doing this is that it will take ages to recompile - as soon as you change something somewhere *everything* needs to be recompiled | 14:46 |
blackburn | phew! | 14:49 |
blackburn | huh shogun is kinda active today | 14:50 |
sploving1 | I have no idea. Note: Recompile with -Xlint:unchecked for details | 14:53 |
@sonney2k | sploving1, then please recompile with that option... | 14:53 |
-!- mikiobraun [~mikio@squid.ml.tu-berlin.de] has joined #shogun | 14:57 | |
mikiobraun | betty: hi | 14:57 |
bettyboo | mikiobraun: you modified something related to | 14:57 |
mikiobraun | hello all. Am I late ;) | 14:59 |
@sonney2k | alesis-novik, do you know where Vojtech is? | 15:00 |
sploving1 | mikiobraun, not yet | 15:00 |
heiko | hi all | 15:00 |
@sonney2k | alesis-novik, ping? | 15:00 |
alesis-novik | sonney2k, pong | 15:00 |
@sonney2k | alesis-novik, hmmhh | 15:01 |
@sonney2k | alesis-novik, Any idea if Vojtech will join today? | 15:01 |
* blackburn closes WoW and Lineage 2 | 15:01 | |
alesis-novik | sonney2k, no idea really. | 15:01 |
@sonney2k | ok then lets start... | 15:02 |
@sonney2k | So now that everyone of you entered the midterm evals (thanks for that btw!) lets have a look where we want to be in 1 month in the future - which is the suggested coding deadline | 15:03 |
-!- VojtechFranc [~quassel@gw-101.scnet.cz] has joined #shogun | 15:03 | |
@sonney2k | alright | 15:03 |
@sonney2k | So now that everyone of you entered the midterm evals (thanks for that btw!) lets have a look where we want to be in 1 month in the future - which is the suggested coding deadline | 15:03 |
@sonney2k | ^Vojtech | 15:03 |
@sonney2k | (was the first sentence) | 15:03 |
@sonney2k | So ideally everyone would be finished with whatever was planned and we would all work on polishing shogun to get it into release ready conditions | 15:04 |
@sonney2k | such that by the end of August we can release shogun 1.0 that includes all your contributions | 15:04 |
@sonney2k | So it would be great if any student (or mentor - one of you is sufficient) just briefly says what the current state is and what his future plans are | 15:06 |
@sonney2k | let me start with heiko :) | 15:06 |
@sonney2k | heiko did the model selection stuff | 15:06 |
heiko | me or you? :) | 15:06 |
@sonney2k | and it is already working pretty well | 15:06 |
heiko | ok go ahead :) | 15:06 |
@sonney2k | heiko, please continue - you know what you want to do :) | 15:06 |
heiko | ok then, first model selection is working from python side | 15:07 |
@sonney2k | and C++ side of course :) | 15:07 |
heiko | yes, a grid search model selection is implemented | 15:08 |
heiko | evaluation is done via cross-validation (currently n-fold-stratified) | 15:08 |
heiko | and any parameter of an CSGObject may be selected for model selection | 15:08 |
heiko | so now simple parameters as SVM-C may be searched for, as well as CSGObjects, so different kernels may be chosen for example | 15:09 |
@sonney2k | ..and there is an example / test for that all already. | 15:09 |
@sonney2k | so now about the future :) | 15:09 |
heiko | yes, I plan to write some real-life examples of typical usage | 15:09 |
heiko | the current example does not use all possible features | 15:09 |
heiko | then, some work has to be done on the subset selection side (implement it for more classes) | 15:10 |
heiko | then, currently, only float64_t and CSGObject are possible for model selection, this should be generic | 15:10 |
heiko | and Soeren also suggested to have a more simple interface from the python side | 15:11 |
heiko | currently, its a bit complicated to specify a complex combination of parameters | 15:11 |
heiko | you may all have a look at the examples: | 15:11 |
heiko | undocumented/libshogun/modelselection_grid_search_simple.cpp | 15:11 |
heiko | undocumented/python_modular/modelselection_grid_search_simple_modulay.py | 15:12 |
heiko | also another big issue: | 15:12 |
@sonney2k | ok we don't need it that detailed now - just highlevel is sufficient | 15:12 |
heiko | cross-validation currently only works with LibLinear | 15:12 |
heiko | Kernel/Distance machines currently store their model by storing indices to features | 15:12 |
heiko | this has to be changed, they need to store feature copies to make cross-val work | 15:13 |
heiko | so much so far | 15:13 |
heiko | I think last thing has highest priority now | 15:13 |
@sonney2k | certainly enough work for the next month I would say - but we will see | 15:13 |
heiko | yes, indeed :) | 15:13 |
@sonney2k | alright, thanks heiko | 15:13 |
cwidmer | just a side-note, I found the cross-validation stuff in scikits learn quite useful (http://scikit-learn.sourceforge.net/cross_validation.html). Might be worth a look for design considerations. | 15:14 |
@sonney2k | cwidmer, yes we looked at this | 15:14 |
cwidmer | ok, great | 15:14 |
* blackburn thinks that chris is scikit-learn spy | 15:14 | |
cwidmer | ;) | 15:14 |
@sonney2k | I think we will create typemaps inspired by that for convenience | 15:14 |
heiko | yes, we also lend some inspirations there ;) | 15:14 |
bettyboo | great | 15:14 |
@sonney2k | any further comments | 15:15 |
@sonney2k | if not then who wants to be next? f-x ? | 15:15 |
f-x | sonney2k: sure | 15:15 |
f-x | current status is: | 15:15 |
f-x | the online framework is fine and working, able to load arbitrarily large datasets | 15:15 |
f-x | though optimizations will surely have to be made for speed | 15:16 |
f-x | Also, versions of SGD, LibLinear, and most of SGD-QN have also been implemented using this framework | 15:16 |
f-x | The next step would be to complete SGD-QN and begin integration of Vowpal Wabbit | 15:17 |
f-x | While benchmarking using larger datasets at the same time | 15:17 |
-!- gsomix [~gsomix@85.26.233.175] has joined #shogun | 15:17 | |
f-x | Plus, examples using the newly implemented algorithms as and when they're ready | 15:18 |
f-x | VW should take a good deal of time, however | 15:18 |
f-x | sonney2k: Any specifics? | 15:18 |
@sonney2k | f-x, I agree - I guess you shouldn't forget that a lot of time will go into benchmarking / identifying and ironing out bottlenecks | 15:19 |
@sonney2k | and fixing bugs of course | 15:19 |
@sonney2k | but again certainly quite a lot of work again | 15:19 |
f-x | Definitely. | 15:19 |
@sonney2k | and it is important here that we stop adding features when we start realizing that it might be too much to fully test/support/document them | 15:20 |
@sonney2k | does anyone have questions to f-x? | 15:20 |
@sonney2k | if not thanks for your work! | 15:20 |
@sonney2k | and alesis-novik do you want to continue? | 15:20 |
alesis-novik | Sure | 15:21 |
alesis-novik | Essentially, the first part of my project is done and I have implemented, tested and created examples for the simple EM for GMM | 15:21 |
@sonney2k | alesis-novik, btw you wanted to send me some python examples - maybe I've missed them | 15:22 |
alesis-novik | sonney2k, I'll send a pull request later today | 15:22 |
@sonney2k | alesis-novik, thanks - please continue | 15:23 |
alesis-novik | it uses the 3 "improvements" over the vanilla one: doing computations in log domain, having a minimal covariance to avoid singular gaussians | 15:23 |
alesis-novik | and using svd to decompose covariances | 15:24 |
alesis-novik | For multivariate gaussians it also allows you to choose one of 3 types of covariances you want to use | 15:24 |
alesis-novik | full, diagonal or spherical | 15:24 |
alesis-novik | This implementation has been tested with both small examples (1d and 2d Gaussians, 200 points) | 15:25 |
alesis-novik | and a large dataset with 256d Gaussians and thousands of points if I remember correctly | 15:26 |
alesis-novik | Currently I'm working on implementing a simple Split-Merge EM | 15:26 |
alesis-novik | I should finish implementing it by the end of the week, then I'll test and work out any bugs | 15:27 |
alesis-novik | Hopefully it will be able to produce better fits and remove the problem of EM sometimes deciding to fit one Gaussian over all of the data (sometimes happens in the small examples( | 15:28 |
@sonney2k | yeah it would be great to see any illustrative examples for that | 15:28 |
@sonney2k | VojtechFranc, do you want to add anything? | 15:28 |
VojtechFranc | just a small comment | 15:29 |
VojtechFranc | 1. I would like to add some other examples on real data | 15:29 |
VojtechFranc | 2. if there is time it would be cool to implement support for other cov type- Factor analysis | 15:30 |
VojtechFranc | otheriwise, we need testing/debuggin/tuning and writting documentation | 15:30 |
VojtechFranc | that's it | 15:30 |
@sonney2k | OK, does anyone else have questiosn? | 15:31 |
alesis-novik | I've been meaning to ask, what should I do about the large example? The small ones are self sufficient but the large one uses the dataset VojtechFranc provided | 15:31 |
VojtechFranc | it is USPS database of handwritten characters, it is publicly available | 15:32 |
@sonney2k | I guess we provide a download link then | 15:32 |
VojtechFranc | I'll collect other data which can be made public | 15:32 |
blackburn | +1 for USPS nums - I'll make an example for dimreduction | 15:32 |
blackburn | anyway we could share some data of that kind | 15:32 |
@sonney2k | if we are allowed to distribute it we can add it to shogun-data | 15:32 |
@sonney2k | I fully agree | 15:32 |
blackburn | just licence issues | 15:33 |
@sonney2k | we need some real-world but intuitive data sets | 15:33 |
@sonney2k | could very well be.... | 15:33 |
@sonney2k | alright | 15:33 |
@sonney2k | sploving1 - do you want to be next? | 15:33 |
sploving1 | okay | 15:33 |
sploving1 | I mainly finised java_modular: Now it can support jblas library well, includes vector, matrix, and string features. In addition, add java_modular 131 examples, and docs. | 15:34 |
sploving1 | also, it can support ujmp matrix well | 15:34 |
sploving1 | that is the current state | 15:35 |
@sonney2k | sploving1, as far as I can see some of the examples don't compile yet though | 15:35 |
sploving1 | it maybe some reason by shogun | 15:36 |
sploving1 | some bug not fixed yet | 15:36 |
@sonney2k | I guess you will need to do a bit of testing for them too (compare if python modular returns the same result) | 15:36 |
@sonney2k | yes I am working on some of them | 15:36 |
@sonney2k | sploving1, please continue | 15:37 |
sploving1 | about the testsuit, it cannot work now. | 15:38 |
sploving1 | I think the object need implement seriazation | 15:38 |
@sonney2k | sploving1, yes lets ignore the test suite for now | 15:38 |
sploving1 | javac can work for examples, but java *.java not work for all. some bugs still need fix | 15:40 |
mikiobraun | I'm just answering your last email. Sorry for my replying so late. | 15:40 |
sploving1 | it doesnot matter | 15:40 |
@sonney2k | sploving1, mikiobraun so what about future? | 15:41 |
blackburn | .. it will be utterly fantastic :D | 15:41 |
@sonney2k | and shogun world domination of course ;-) | 15:42 |
mikiobraun | Existing tests should compile and run, of course. Otherwise we should probably use the rest of the time for the Lua thing. (assuming that sploving has more fun on Lua) ;) | 15:42 |
@sonney2k | IIRC sploving1 did many typemaps for lua and ruby too already - right? | 15:43 |
@sonney2k | so I think they mostly need examples again and some missing typemaps | 15:43 |
@sonney2k | though this is a lot easier to do this time with a scripting language than java | 15:43 |
sploving1 | yeap. I would like to work on lua. The main plan is to support lua matrix | 15:44 |
-!- f-x [~user@117.192.197.185] has quit [Ping timeout: 260 seconds] | 15:44 | |
sploving1 | and add lua examples, docs and clean up java/lua binding | 15:44 |
blackburn | man down | 15:44 |
@sonney2k | my suggestion would be to this time start earlier with adding examples | 15:44 |
@sonney2k | this way we can test for breakage more early | 15:44 |
@sonney2k | java caused a lot of unforeseen problems | 15:45 |
mikiobraun | I agree | 15:45 |
@sonney2k | sploving1, like when typemaps for vectors are there - illustrate how Labels() work | 15:45 |
blackburn | we will have a Kernel problem in ruby ;) | 15:45 |
blackburn | I even would say the Kernel Problem | 15:45 |
@sonney2k | when the ones for matirx are there, e.g. use gaussian kernel | 15:46 |
@sonney2k | blackburn, but that is ruby :) lua first | 15:46 |
blackburn | yeah I just recalled The Kernel Problem serialhex reported month ago or so | 15:46 |
@sonney2k | sploving1, again a lot of work but still - it is do-able. | 15:47 |
sploving1 | I have no idea how to make Labels work now. I write them nearly a month ago | 15:47 |
sploving1 | and forgot it | 15:47 |
@sonney2k | sploving1, copy and paste the example from python :) | 15:47 |
@sonney2k | alright, does anyone else have questions? | 15:47 |
sploving1 | I suggest not write so many examples | 15:48 |
sploving1 | it is borning | 15:48 |
sploving1 | My main advantage is to write typemap | 15:48 |
sploving1 | there will be no much time to debug | 15:49 |
blackburn | I can help writing examples if it is needed ;) | 15:49 |
blackburn | or gsomix can too | 15:49 |
@sonney2k | sploving1, well the examples will tell us if your typemaps work | 15:49 |
@sonney2k | so we definitely will need some for each feature type | 15:49 |
@sonney2k | of course lua is much more close to python | 15:49 |
@sonney2k | so it will be a lot easier now | 15:50 |
sploving1 | thx. blackburn. | 15:50 |
@sonney2k | but sploving1, yes if you want to e.g. write more typemaps for other languages like c# instead we can arrange sth | 15:50 |
-!- in3xes_ is now known as in3xes | 15:51 | |
blackburn | I have an suggestion - I could finish up work with java examples | 15:51 |
@sonney2k | blackburn, that would be great... | 15:51 |
blackburn | okay so let sploving1 just work on typemaps | 15:52 |
@sonney2k | this way sploving1 can have some more fun with typemaps | 15:52 |
-!- f-x [~user@27.57.28.101] has joined #shogun | 15:52 | |
@sonney2k | so sploving1 what do you think - if you now just do lua / ruby /c# typemaps - no examples? | 15:52 |
@sonney2k | better | 15:52 |
@sonney2k | ? | 15:52 |
sploving1 | no examples? that is impossible, how to test? | 15:53 |
blackburn | we could help with examples, you could work only on typemaps | 15:53 |
sploving1 | I suggest I write some of them, not all | 15:53 |
@sonney2k | sploving1, well then port 1 example for each datatype - the others we will do | 15:53 |
sploving1 | that souds good | 15:53 |
sploving1 | okay. | 15:54 |
sploving1 | I will try my best. lua/ruby/c# | 15:54 |
@sonney2k | ok - is everyone (more) happy now? | 15:54 |
cwidmer | one happy family | 15:55 |
blackburn | more communism for shogun | 15:55 |
@sonney2k | that's how it should be | 15:55 |
@sonney2k | which brings us to sweet soviet russia | 15:55 |
@sonney2k | blackburn, your turn | 15:55 |
blackburn | okay I have finished working on three base algorithms with some modifications | 15:55 |
blackburn | in particular it is | 15:56 |
blackburn | LocallyLinearEmbedding | 15:56 |
blackburn | ClassicMDS | 15:56 |
blackburn | stands for Classic Multidimensional Scaling | 15:56 |
blackburn | and LandmarkMDS | 15:56 |
blackburn | in addition to both of MDSs I have implemented Isomaps | 15:56 |
blackburn | so, ClassicIsomap and LandmarkIsomap | 15:56 |
blackburn | during work on these algos I have notified we should use some other solver for eigenproblems | 15:57 |
blackburn | in all of them eigenproblem is stated | 15:57 |
blackburn | so, I have integrated ARPACK (which is considered to be fastest for that kind of eigenproblems) | 15:58 |
blackburn | even faster than lobpcg | 15:58 |
blackburn | my plan for next month is to implement three variations of LLE: Hessian LLE, Modified LLE and LTSA (local tangent space) | 15:59 |
blackburn | and I would like to integrate SuperLU for sparse LU factorization | 15:59 |
blackburn | it is needed in shift-invert mode of ARPACK | 15:59 |
* sonney2k googles superlu | 16:00 | |
blackburn | moreover, some real data examples should be good - I have been thinking about it for a while | 16:00 |
blackburn | have some ideas, will add some of them during next month | 16:01 |
cwidmer | blackburn, on the bioinf side, I can provide you with some examples from biological sequences | 16:01 |
blackburn | yes it would be wonderful | 16:01 |
blackburn | I'm still interested if manifold learning is applicable to biological data | 16:02 |
cwidmer | same here ^^ | 16:02 |
@sonney2k | blackburn, ok thanks... | 16:03 |
@sonney2k | does anyone else want to comment? | 16:03 |
blackburn | about superlu | 16:03 |
blackburn | it is funny why I want to do it: scikit-learn use it | 16:04 |
blackburn | it seems that I'm a scikit-learn spy too | 16:04 |
cwidmer | :) | 16:04 |
bettyboo | :) | 16:04 |
blackburn | ah not scikit-learn, scipy uses it | 16:04 |
cwidmer | spying seems to pay off | 16:04 |
@sonney2k | :D | 16:05 |
cwidmer | are there any license issues with that? | 16:05 |
blackburn | with what? | 16:05 |
cwidmer | superlu | 16:05 |
cwidmer | or is it not a lib | 16:05 |
blackburn | it is a lib | 16:05 |
blackburn | e.g. it is available even in ubuntu | 16:06 |
cwidmer | sounds good | 16:06 |
blackburn | okay I have no idea what licence do they use | 16:06 |
@sonney2k | it seems to be 3-clause BSD | 16:07 |
@sonney2k | so all good | 16:07 |
blackburn | ah ok | 16:07 |
heiko | when its in ubuntu they probably use some gpl stuff :) | 16:07 |
blackburn | not really | 16:07 |
* sonney2k just checked from the source | 16:07 | |
@sonney2k | so good to go with superlu | 16:07 |
blackburn | http://crd.lbl.gov/~xiaoye/SuperLU/License.txt | 16:07 |
blackburn | I'm not sure it will speed it up really | 16:07 |
blackburn | :D | 16:07 |
blackburn | i.e. it will but may be only for LARGE scale | 16:08 |
@sonney2k | blackburn, I mean the alternative to doing optimizations is to do other algorithms... | 16:08 |
@sonney2k | but up to you and cwidmer :) | 16:08 |
blackburn | I have chosen the most essential ones ;) if we will notify any very good algo - of course it should be implemented | 16:09 |
blackburn | notice* | 16:09 |
@sonney2k | OK, so we will keep you busy if we do :) | 16:09 |
bettyboo | he | 16:09 |
cwidmer | we have quite a few algorithms already, but blackburn, if you still have a few on the stack, it might be worth discussing doing these before the optimizations | 16:09 |
VojtechFranc | sonney2k, is there any other global issue to discuss? I'll need to go. | 16:09 |
blackburn | hehe sorry my report took a while | 16:10 |
@sonney2k | VojtechFranc, yes next meeting | 16:10 |
@sonney2k | I would say we do it in 2 weeks from today | 16:11 |
cwidmer | fine with me | 16:11 |
@sonney2k | how about Wednesday Jul 27 13:00 UTC? | 16:11 |
blackburn | okay | 16:11 |
heiko | ok | 16:11 |
cwidmer | k | 16:11 |
VojtechFranc | ok | 16:11 |
@sonney2k | mikiobraun, alesis-novik sploving1 ? | 16:11 |
f-x | okay | 16:12 |
sploving1 | no problem | 16:12 |
@sonney2k | and then we see how far everyone is | 16:12 |
@sonney2k | and readjust | 16:12 |
@sonney2k | re-adjust | 16:12 |
alesis-novik | Sounds fine | 16:12 |
@sonney2k | if possible I'd like to see all the big changes in the next 2 weeks | 16:12 |
@sonney2k | afterwards only incremental work | 16:13 |
@sonney2k | such that we can stabalize everything in August | 16:13 |
@sonney2k | ...to make a release | 16:13 |
blackburn | release race ;) | 16:13 |
@sonney2k | (e.g. documenting, examples, minor bugfixes etc all in august) | 16:13 |
@sonney2k | so we will probably have to have some bug squashing party at the end of august with everyone of us in the chat for some time | 16:14 |
@sonney2k | anyway that's it form my side | 16:14 |
@sonney2k | any questions? | 16:14 |
@sonney2k | concerns? | 16:14 |
@sonney2k | vodka related issues? | 16:15 |
blackburn | I have enough vodka | 16:15 |
@sonney2k | if not then thanks again for attending | 16:15 |
blackburn | :D | 16:15 |
@sonney2k | and keep up the good work!!! | 16:15 |
alesis-novik | blackburn, if you ever have vodka quantity related issues, we are all doomed | 16:15 |
blackburn | yeeeeah | 16:16 |
blackburn | let the vodka begins okay | 16:16 |
blackburn | :D | 16:16 |
* gsomix away | 16:17 | |
@sonney2k | blackburn, regarding the java examples | 16:18 |
-!- [1]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has joined #shogun | 16:18 | |
alesis-novik | it would probably mean that the Russian economy has collapsed. I remember seeing an ad where they use vodka to pay taxes in czar Russia | 16:18 |
blackburn | hehe | 16:18 |
VojtechFranc | so, I need to go. bye | 16:18 |
bettyboo | zzz | 16:18 |
-!- VojtechFranc [~quassel@gw-101.scnet.cz] has quit [Remote host closed the connection] | 16:18 | |
@sonney2k | VojtechFranc, cu! | 16:18 |
blackburn | in fact russian economy is more related to oil and gas | 16:19 |
CIA-32 | shogun: Soeren Sonnenburg master * r377c55b / (8 files): fix a few more java examples ... even more includes - http://bit.ly/qR5IQE | 16:19 |
cwidmer | ok, I gotta go, happy hacking! | 16:19 |
CIA-32 | shogun: Soeren Sonnenburg master * rd531d4c / (3 files): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/omeALF | 16:19 |
blackburn | cwidmer: see ya | 16:19 |
@sonney2k | CIA-32, cu | 16:19 |
@sonney2k | cwidmer, cu | 16:19 |
blackburn | sonney2k: CIA-32 is not leaving | 16:19 |
blackburn | sonney2k: what about java examples? | 16:19 |
-!- CIA-32 was kicked from #shogun by sonney2k [CIA-32] | 16:19 | |
@sonney2k | he is :D | 16:19 |
blackburn | oh | 16:19 |
-!- CIA-20 [~CIA@cia.atheme.org] has joined #shogun | 16:19 | |
blackburn | :D | 16:19 |
blackburn | killer | 16:20 |
@sonney2k | I knew that this would happen | 16:20 |
@sonney2k | some of the java examples currently issue a couple of warnings or don't compile | 16:20 |
@sonney2k | so just fix those | 16:20 |
blackburn | okay, I'll take a look | 16:21 |
@sonney2k | then we still need to run every example one-by-one (that works using sploving1 check.sh script) | 16:21 |
@sonney2k | and at some point create functions in similar spirit to java_modular | 16:21 |
@sonney2k | though I am not so sure about serialization at the moment... | 16:21 |
@sonney2k | because I had the following idea | 16:22 |
@sonney2k | one could use python to write down the serialized shogun objects | 16:22 |
@sonney2k | and nothing more | 16:22 |
-!- cwidmer [~quassel@connect.tuebingen.mpg.de] has quit [Remote host closed the connection] | 16:22 | |
@sonney2k | if we do that, we could serialize the shogun objects from java too | 16:22 |
@sonney2k | and load the ones from python and compare... | 16:22 |
blackburn | I think there is no problem with serialization | 16:23 |
@sonney2k | for now I would just check if e.g. kernel matrices are the same by just printing them | 16:23 |
@sonney2k | blackburn, in which respect? | 16:23 |
blackburn | I mean it is possible to compare python and java outputs | 16:24 |
-!- f-x` [~user@117.192.203.179] has joined #shogun | 16:24 | |
blackburn | now compiling java_modular :) | 16:25 |
alesis-novik | blackburn, I don't know why I decided to look this up, but "The taxes on vodka became a key element of government finances in Tsarist Russia, providing at times up to 40% of state revenue." | 16:26 |
blackburn | hehe | 16:26 |
blackburn | na samom dele eto strashno :) | 16:26 |
-!- f-x [~user@27.57.28.101] has quit [Ping timeout: 260 seconds] | 16:26 | |
blackburn | sonney2k: will you have some fancy lecture about shogun 1.0? ;) just like for older ones | 16:28 |
f-x` | sonney2k: any ideas about the SGD-QN? | 16:28 |
@sonney2k | f-x`, lets ask the mighty masters, leaon and antoine | 16:29 |
@sonney2k | leon | 16:29 |
@sonney2k | f-x`, I will email them | 16:29 |
f-x` | sonney2k: cool :) thanks! | 16:29 |
@sonney2k | f-x`, I know them pretty well so they might even respond :D | 16:30 |
heiko | sonney2k, does shogun actually use boost? | 16:33 |
@sonney2k | heiko, no | 16:33 |
f-x` | great :) they should surely know what modifications will be needed for online behaviour. i'm done implementing most of the functions except that determine_t0() | 16:33 |
heiko | and you probably do not want to add it ? :) | 16:33 |
@sonney2k | heiko, boost is a huge dependency so one has to be very careful about that | 16:33 |
@sonney2k | heiko, what do you need? | 16:33 |
heiko | ok then I'll compute confidence intervals myself | 16:33 |
@sonney2k | f-x`, sent the email | 16:33 |
heiko | but now, off for some coffee :) | 16:34 |
heiko | see you later | 16:34 |
@sonney2k | coffee?! vodka! | 16:34 |
blackburn | coffee is not that shoguneers drink | 16:35 |
-!- mikiobraun [~mikio@squid.ml.tu-berlin.de] has left #shogun ["Leaving."] | 16:39 | |
-!- sploving1 [~sploving@124.16.139.134] has left #shogun [] | 16:41 | |
@sonney2k | blackburn, do you see a way to modify the SGObject.java to have an implements serializable? | 16:47 |
blackburn | I have to take a look on it first ;) | 16:47 |
blackburn | sonney2k: where it is? | 16:49 |
@sonney2k | blackburn, recall that we can pickle objects from python? | 16:50 |
blackburn | yes | 16:50 |
@sonney2k | wouldn't it be cool to have serializable shogun objects in java too? | 16:50 |
blackburn | I mean where to find SGObject.java?? | 16:50 |
@sonney2k | so all that is needed for that is that the generated SGObject.java derives from serializable | 16:50 |
blackburn | фр | 16:51 |
blackburn | ah | 16:51 |
@sonney2k | blackburn, in java_modular/org/shogun | 16:51 |
@sonney2k | (org/shogun is being generated) | 16:51 |
blackburn | ah okay, compiling | 16:52 |
f-x` | sonney2k: do you know of any scripts to readily convert between common input formats? like convert.py from the lsl site did from dense->SVMLight | 16:54 |
f-x` | would be great for dense<->libsvm<->svmlight etc | 16:54 |
@sonney2k | f-x`, you could use shogun for that, i.e. load in svmlight format, convert to dense features and write out | 16:56 |
@sonney2k | and vice versa | 16:56 |
@sonney2k | blackburn there seems to be a "%pragma(java) jniclassinterfaces=x" | 16:56 |
blackburn | no idea wtf is it :D | 16:57 |
@sonney2k | I will have a look at this - would be too cute if that worked :) | 16:59 |
@sonney2k | %pragma(java) moduleimports=%{ | 17:05 |
@sonney2k | import java.io.*; // For Serializable | 17:05 |
@sonney2k | %} | 17:05 |
@sonney2k | %pragma(java) jniclassinterfaces="Serializable" | 17:05 |
@sonney2k | %pragma(java) moduleinterfaces="Serializable" | 17:05 |
@sonney2k | lets see what happens | 17:05 |
@sonney2k | kaboom of course | 17:09 |
blackburn | I guess readObject and writeObject should be implemented somehow.. | 17:12 |
blackburn | how it works with python? | 17:12 |
@sonney2k | blackburn, yes sure | 17:19 |
@sonney2k | that would be easy | 17:19 |
@sonney2k | the only problem is that we need SGObject to implement Serializable | 17:19 |
blackburn | okay compiles now | 17:19 |
@sonney2k | and currently I don't see how to do this (if not using sed on the swig'd file) | 17:19 |
@sonney2k | the jniclassinterface indeed did add a implements | 17:20 |
@sonney2k | but not sure what the *.java class has to say about this | 17:20 |
@sonney2k | anyway | 17:20 |
@sonney2k | gtg now | 17:20 |
@sonney2k | l8r | 17:20 |
blackburn | okay | 17:20 |
blackburn | see you | 17:20 |
-!- in3xes is now known as gpk | 17:33 | |
-!- gpk is now known as in3xes | 17:33 | |
-!- heiko [~heiko@main.uni-duisburg.de] has quit [Ping timeout: 258 seconds] | 18:31 | |
-!- gsomix [~gsomix@85.26.233.175] has quit [Ping timeout: 255 seconds] | 18:33 | |
blackburn | sonney2k: have problem with java modular installation :) | 18:43 |
bettyboo | not bad | 18:43 |
blackburn | bettyboo: it is bad | 18:43 |
bettyboo | blackburn: ah, you mean that some kernels are 'bad'? sure it is | 18:43 |
-!- gsomix [~gsomix@85.26.232.71] has joined #shogun | 18:45 | |
blackburn | install -m755 java_modular/shogun.jar "//share/java" | 19:03 |
blackburn | install: cannot stat `java_modular/shogun.jar': No such file or directory | 19:03 |
blackburn | sonney2k: ^^ am I doing something wrong? | 19:03 |
-!- gsomix [~gsomix@85.26.232.71] has quit [Ping timeout: 255 seconds] | 19:07 | |
-!- gsomix [~gsomix@85.26.232.71] has joined #shogun | 19:09 | |
-!- blackburn1 [~blackburn@188.122.238.99] has joined #shogun | 19:36 | |
-!- blackburn [~blackburn@188.122.238.99] has quit [Ping timeout: 255 seconds] | 19:38 | |
-!- blackburn1 [~blackburn@188.122.238.99] has quit [Client Quit] | 19:38 | |
-!- blackburn [~blackburn@188.122.238.99] has joined #shogun | 19:38 | |
-!- [1]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has quit [Ping timeout: 246 seconds] | 20:16 | |
-!- gsomix [~gsomix@85.26.232.71] has quit [Ping timeout: 250 seconds] | 21:17 | |
-!- in3xes_ [~in3xes@180.149.49.230] has joined #shogun | 21:41 | |
-!- warpy2 [~androirc@bzq-219-227-204.pop.bezeqint.net] has joined #shogun | 21:52 | |
CIA-20 | shogun: Sergey Lisitsyn master * r32a8bd0 / (12 files in 4 dirs): Renamed and cleaned PCA - http://bit.ly/pdXwNP | 22:04 |
-!- in3xes_ [~in3xes@180.149.49.230] has quit [Quit: Leaving] | 22:14 | |
-!- warpy2 [~androirc@bzq-219-227-204.pop.bezeqint.net] has quit [Read error: Connection reset by peer] | 22:36 | |
-!- in3xes_ [~in3xes@180.149.49.230] has joined #shogun | 22:50 | |
-!- CIA-87 [~CIA@cia.atheme.org] has joined #shogun | 23:01 | |
-!- Netsplit *.net <-> *.split quits: CIA-20 | 23:07 | |
-!- [1]warpy [~warpy@bzq-79-181-19-104.red.bezeqint.net] has joined #shogun | 23:09 | |
-!- f-x` [~user@117.192.203.179] has quit [Read error: Connection reset by peer] | 23:33 | |
-!- f-x` [~user@117.192.203.179] has joined #shogun | 23:42 | |
-!- in3xes_ [~in3xes@180.149.49.230] has quit [Quit: Leaving] | 23:43 | |
-!- Todd_ [4526dd82@gateway/web/freenode/ip.69.38.221.130] has joined #shogun | 23:48 | |
Todd_ | Hi-I am organizer for the San Francisco Data Mining meetup...was wondering to any of you (contributors to shogun) are located near San Francisco, and would be interested in giving a talk about shogun to our group on Aug 16? | 23:50 |
blackburn | Hi, Todd_ | 23:51 |
blackburn | I don't really know but I guess no, most of us are located in europe | 23:52 |
Todd_ | thats too bad...I'm a fan of the project...thanks anyhow! | 23:53 |
blackburn | Todd_: you could leave some email or so, I'll ask 'a boss', sonney2k | 23:53 |
Todd_ | that would be great...my email is todd.holloway@gmail.com | 23:53 |
blackburn | okay, thanks for interest anyway :) | 23:54 |
-!- Todd_ [4526dd82@gateway/web/freenode/ip.69.38.221.130] has quit [Quit: Page closed] | 23:54 | |
blackburn | serialhex: give a talk at san francisco :D | 23:55 |
--- Log closed Sat Jul 16 00:00:25 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!