--- Log opened Wed Jul 06 00:00:49 2011 | ||
-!- alesis-novik [~alesis@188.74.87.206] has quit [Quit: I'll be Bach] | 01:20 | |
-!- f-x [~user@117.192.193.217] has quit [Read error: Connection reset by peer] | 01:32 | |
-!- f-x [~user@117.192.193.217] has joined #shogun | 01:36 | |
-!- f-x [~user@117.192.193.217] has quit [Remote host closed the connection] | 02:32 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Remote host closed the connection] | 03:33 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 03:33 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 260 seconds] | 07:06 | |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 07:18 | |
CIA-32 | shogun: Shashwat Lal Das master * r8e55cd8 / (3 files): Removed unnecessary macro definitions of the get_vector functions in StreamingFileFrom*Features. - http://bit.ly/qvnOBh | 08:55 |
---|---|---|
CIA-32 | shogun: Soeren Sonnenburg master * re7e71c8 / (3 files): | 08:55 |
CIA-32 | shogun: Merge pull request #171 from frx/streamingfile | 08:55 |
CIA-32 | shogun: Removed unnecessary macro definitions of the get_vector functions - http://bit.ly/oYJXjL | 08:55 |
@sonney2k | f-x__, thanks | 08:57 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 09:19 | |
-!- sploving1 [~sploving@124.16.139.194] has joined #shogun | 09:21 | |
sploving1 | hi sonney2k | 09:22 |
@sonney2k | sploving1, IIRC jblas doesn't support types other than float/double? | 09:22 |
@sonney2k | thought only UJMP can handle that | 09:22 |
sploving1 | sonney2k, jblas could not support it. but we could use DoubleMatrix, and convert it into the types that shogun need | 09:23 |
sploving1 | array[i] = (SGTYPE)carr[i]; like this | 09:24 |
sploving1 | as doulbe could be converted any other type. | 09:24 |
@sonney2k | sploving1, I see - so you do that? | 09:24 |
@sonney2k | makes sense then | 09:24 |
@sonney2k | but needs to be documented... | 09:24 |
sploving1 | so the examples in java_modular like ushort could be added | 09:24 |
sploving1 | sonney2k, when documented, I will docted it. | 09:25 |
@sonney2k | I see | 09:25 |
sploving1 | now there is an error | 09:25 |
@sonney2k | merging then | 09:25 |
CIA-32 | shogun: Baozeng Ding master * r59fe842 / src/java_modular/swig_typemaps.i : support other types for jblas - http://bit.ly/rnAyiz | 09:26 |
CIA-32 | shogun: Baozeng Ding master * r6e66f91 / (3 files): add regression examples for java_modular - http://bit.ly/nAvDXF | 09:26 |
CIA-32 | shogun: Soeren Sonnenburg master * r2fc6eb1 / (4 files in 2 dirs): | 09:26 |
CIA-32 | shogun: Merge pull request #172 from sploving/master | 09:26 |
CIA-32 | shogun: support other types for jblas and add regression examples - http://bit.ly/n7ZpHQ | 09:26 |
@sonney2k | sploving1, error? | 09:26 |
sploving1 | compile: | 09:27 |
sploving1 | [javac] Compiling 623 source files to /home/sploving/shogun/src/java_modular/classes | 09:27 |
sploving1 | [javac] /home/sploving/shogun/src/java_modular/org/shogun/CustomKernel.java:68: in org.shogun.CustomKernel defineded alread set_full_kernel_matrix_from_full(org.jblas.DoubleMatrix) | 09:27 |
sploving1 | [javac] public boolean set_full_kernel_matrix_from_full(DoubleMatrix full_kernel_matrix) { | 09:27 |
sploving1 | 09:27 | |
sploving1 | when I make, it shows the above error | 09:27 |
sploving1 | sonney2k, you could make hava a try. git clean -dfx, ./configure --interfaces=libshougn,java_modular, make -j4 | 09:28 |
@sonney2k | sploving1, so it is because of your recent change or? | 09:29 |
@sonney2k | I mean I just compiled before your commit | 09:29 |
@sonney2k | sploving1, yeah look at this | 09:29 |
@sonney2k | public boolean set_full_kernel_matrix_from_full(DoubleMatrix full_kernel_matrix) { | 09:29 |
@sonney2k | return KernelJNI.CustomKernel_set_full_kernel_matrix_from_full__SWIG_0(swigCPtr, this, full_kernel_matrix); | 09:29 |
@sonney2k | } | 09:29 |
@sonney2k | public boolean set_full_kernel_matrix_from_full(DoubleMatrix full_kernel_matrix) { | 09:29 |
@sonney2k | return KernelJNI.CustomKernel_set_full_kernel_matrix_from_full__SWIG_1(swigCPtr, this, full_kernel_matrix); | 09:29 |
@sonney2k | } | 09:29 |
@sonney2k | Ifrom CustomKernel.java | 09:29 |
@sonney2k | sploving1, I guess that is due to the new typemaps | 09:30 |
sploving1 | sonney2k, how to fix it | 09:31 |
@sonney2k | sploving1, I don't know | 09:32 |
sploving1 | could we delete one method in shogun? | 09:32 |
@sonney2k | sploving1, can't you internally detect what target type is needed? | 09:32 |
@sonney2k | actually it is a bug in the typemaps | 09:33 |
@sonney2k | look at libshogun/kernel/CustomKernel.h | 09:33 |
sploving1 | sonney2k, I will take a look at it | 09:34 |
@sonney2k | there the function is defined for float32_t and for float64_t | 09:34 |
@sonney2k | but in the java code it is always DoubleMatrix | 09:34 |
@sonney2k | it should be FloatMatrix once and DoubleMatrix for the other | 09:34 |
sploving1 | it is easy. sonne2k, just add floatmatrix to typemap float_32 | 09:35 |
sploving1 | others using DoubleMatrix | 09:35 |
sploving1 | make sense? | 09:35 |
@sonney2k | where exactly? | 09:36 |
sploving1 | I mean splite float32_t in swig_type.i from other types | 09:37 |
sploving1 | sonne2k, ok? | 09:37 |
sploving1 | sonney2k | 09:37 |
@sonney2k | sploving1, I guess you can do that by enhancing the macro a bit | 09:37 |
@sonney2k | such that you always need to specify the *Matrix type | 09:38 |
@sonney2k | then you don't need two separate cases | 09:38 |
@sonney2k | sploving1, I would say give this a try | 09:38 |
sploving1 | sonney2k, I will have a try. another thing is Regression examples | 09:40 |
sploving1 | I could javac regression_libsvr_modular.java wokr | 09:40 |
sploving1 | but java regression_libsvr_modular does not work | 09:41 |
@sonney2k | sploving1, well I cannot test when java is not compiling | 09:41 |
@sonney2k | so please fix that first | 09:41 |
sploving1 | sonney2k, fixed it. now I am testing it | 09:51 |
@sonney2k | sploving1, and? does it work? | 10:02 |
sploving1 | sonney2k, not work well, still look where is the bug | 10:03 |
@sonney2k | ok, just let me know when you made progress - I will be around | 10:04 |
-!- sploving1 [~sploving@124.16.139.194] has quit [Ping timeout: 255 seconds] | 10:13 | |
-!- sploving1 [~sploving@124.16.139.134] has joined #shogun | 10:14 | |
sploving1 | my machine crashed justs now | 10:14 |
sploving1 | okay. when it works, I will push it | 10:16 |
@sonney2k | ok | 10:16 |
-!- blackburn [~blackburn@188.122.238.99] has joined #shogun | 10:21 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Quit: Ooops..!!] | 10:26 | |
sploving1 | sonney2k, pushed it | 10:28 |
@sonney2k | sploving1, looking | 10:28 |
@sonney2k | sploving1, where? | 10:29 |
@sonney2k | did you issue a pull request | 10:29 |
sploving1 | https://github.com/sploving/shogun/commit/d05e6bbf6e7a106d0c6025152760557ab4fcca5e | 10:29 |
sploving1 | no | 10:29 |
sploving1 | I will pull request | 10:29 |
sploving1 | okay | 10:30 |
CIA-32 | shogun: Baozeng Ding master * rd05e6bb / src/java_modular/swig_typemaps.i : improve jblas typemap for various types - http://bit.ly/nJe9oD | 10:32 |
CIA-32 | shogun: Soeren Sonnenburg master * rf23f64a / src/java_modular/swig_typemaps.i : | 10:32 |
CIA-32 | shogun: Merge pull request #173 from sploving/master | 10:32 |
CIA-32 | shogun: improve typemap for jblas - http://bit.ly/p3zJZC | 10:32 |
-!- blackburn1 [~blackburn@188.122.238.99] has joined #shogun | 10:37 | |
sploving1 | sonney2k, java regression_krr_modular still have problem | 10:39 |
sploving1 | Exception in thread "main" java.lang.UnsatisfiedLinkError: org.shogun.ClassifierJNI.KernelMachine_SWIGUpcast(J)J | 10:40 |
sploving1 | at org.shogun.ClassifierJNI.KernelMachine_SWIGUpcast(Native Method) | 10:40 |
sploving1 | 10:40 | |
@sonney2k | yes I see - can you make any sense of that | 10:40 |
@sonney2k | ? | 10:40 |
-!- blackburn [~blackburn@188.122.238.99] has quit [Ping timeout: 255 seconds] | 10:40 | |
sploving1 | sonney2k, I have no idea | 10:41 |
blackburn1 | UnsatisfiedLinkError... I love it | 10:41 |
@sonney2k | running regression_libsvr_modular .. | 10:42 |
@sonney2k | Exception in thread "main" java.lang.UnsatisfiedLinkError: org.shogun.ClassifierJNI.SVM_SWIGUpcast(J)J | 10:42 |
@sonney2k | at org.shogun.ClassifierJNI.SVM_SWIGUpcast(Native Method) | 10:42 |
@sonney2k | at org.shogun.SVM.<init>(Unknown Source) | 10:42 |
@sonney2k | at org.shogun.LibSVR.<init>(Unknown Source) | 10:42 |
@sonney2k | at org.shogun.LibSVR.<init>(Unknown Source) | 10:42 |
@sonney2k | at regression_libsvr_modular.main(regression_libsvr_modular.java:28) | 10:42 |
blackburn1 | ain't we fixed it with some import? | 10:42 |
@sonney2k | it seems to only happen with Regression | 10:42 |
@sonney2k | so I suspect there is a classification / machine / regression clash somehow | 10:42 |
@sonney2k | blackburn1, a new day a new problem :) | 10:42 |
@sonney2k | blackburn1, for you :D | 10:43 |
-!- blackburn1 is now known as blackburn | 10:43 | |
blackburn | why for me, my headache is only arpack now :D | 10:43 |
@sonney2k | blackburn, yeah - I forgot that :) | 10:43 |
blackburn | shift-invert mode is still f-ked | 10:44 |
blackburn | and is not used anywhere | 10:44 |
@sonney2k | arpack is suffitient punishment :D | 10:46 |
@sonney2k | c | 10:46 |
@sonney2k | http://i.imgur.com/iYnGb.png | 10:49 |
blackburn | ahahha | 10:49 |
@sonney2k | hmmh could be a problem with KernelMachine being included in classifiers and Regression | 10:51 |
blackburn | the first time in my life the solution could be not to include something? :D | 10:53 |
@sonney2k | blackburn, heh | 10:54 |
@sonney2k | I think we need to have a separate Machine.i | 10:54 |
sploving1 | for i in xrange(10),(python) does java have similar function? anyone knows? | 11:10 |
blackburn | ehh | 11:11 |
blackburn | for(int i=0; i<10; i++)? :D | 11:12 |
sploving1 | blackburn, that is not native I think | 11:12 |
blackburn | native? | 11:12 |
sploving1 | but is useful | 11:12 |
blackburn | what could be more 'native'? | 11:12 |
sploving1 | if java hava a build-in function "arange" then it is more native | 11:13 |
sploving1 | I do not know whether it has or not | 11:14 |
blackburn | of course not | 11:14 |
blackburn | python has no cycles - that is why it has ranges | 11:14 |
blackburn | you can create Collection containing 1,2,3,4,5,6, ... and use for(Integer i: collection) | 11:14 |
blackburn | but it is the very bad way | 11:14 |
sploving1 | oh | 11:15 |
blackburn | java way to iterate over numbers is definitely loops | 11:16 |
sploving1 | you are a java/python expert? hero! | 11:16 |
blackburn | why hero? :) | 11:17 |
sploving1 | I am not good at them. just know basic knowlege | 11:18 |
blackburn | well C++/java/python is kinda standard :) | 11:19 |
bettyboo | haha | 11:19 |
sploving1 | in python, we could use 3*'a', in java, could we use 3*"a"? | 11:21 |
blackburn | no, no way | 11:22 |
sploving1 | they how to get that? blackburn | 11:23 |
blackburn | with loops.. :) | 11:23 |
blackburn | and StringBuilder | 11:23 |
blackburn | you can '+' strings | 11:23 |
sploving1 | blackburn, did you see test_check__comman...modular.py in python exmaple | 11:24 |
sploving1 | there are a lot of num* 'ACGT' | 11:24 |
blackburn | aha I see | 11:25 |
sploving1 | a bit trival for java, do you agree with me? | 11:26 |
blackburn | strange test anyways | 11:27 |
blackburn | I'll make a function repeating string num times | 11:27 |
blackburn | in a minute | 11:27 |
sploving1 | in java?or python | 11:28 |
blackburn | java | 11:28 |
sploving1 | thx in advance, blackburn | 11:31 |
blackburn | sploving1: http://pastebin.com/AzSFDx1v | 11:32 |
* blackburn wasn't code in java for a two months: what a wonderful feeling | 11:33 | |
sploving1 | blackburn, lookes good~ | 11:35 |
blackburn | will work fast ;) | 11:36 |
blackburn | the bad way to do that - just + strings | 11:36 |
blackburn | it will cause high memory usage | 11:36 |
blackburn | i.e. then you do a = a+b | 11:36 |
blackburn | it makes string a+b | 11:36 |
blackburn | as well as a | 11:37 |
blackburn | so O(n^2) memory usage :D | 11:37 |
sploving1 | blackburn, I think in the test, there is no need a+b | 11:39 |
blackburn | I mean if you use loop where String + String | 11:39 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 11:39 | |
blackburn | it will produce as much strings as 'num' | 11:39 |
blackburn | if not Stringbuilder or Stringbuffer is used | 11:40 |
sploving1 | if String POS = repeat("ACGT", 100); POS+NEG = repeat("ACGT", 200)? | 11:40 |
blackburn | eh? | 11:41 |
sploving1 | blackburn, in the test, POS is the same as NEG | 11:43 |
sploving1 | why POS+NEG, not construct just a POS | 11:43 |
blackburn | I don't know, if it is the same just construct it this way | 11:44 |
sploving1 | blackburn, POS is not only 'ACGT' there is 'TTGT' | 11:45 |
sploving1 | also | 11:45 |
blackburn | well so just do POS = repeat("ACGT",some X) + repeat("TTGT",some Y) | 11:46 |
blackburn | or so | 11:46 |
sploving1 | blackburn, in python, POS+NEG will result a string or string[]? | 11:50 |
blackburn | eh? | 11:50 |
blackburn | what is pos and what is neg? | 11:51 |
sploving1 | in test_check_comm...modular.py | 11:51 |
blackburn | ah | 11:53 |
blackburn | non-trivial shit | 11:53 |
blackburn | you have to do it this way | 11:53 |
sploving1 | tring POS = repeat("ACGT", 600) + repeat("TTGT", 210) + repeat("ACGT", 600);String NEG = repeat("ACGT", 600) + repeat("TTGT", 210) + repeat("ACGT", 600); | 11:54 |
sploving1 | then use POS+NEG? | 11:54 |
blackburn | String[] POS = new String[how many]; | 11:54 |
blackburn | for (int i=0; i<how many; i++) | 11:54 |
blackburn | POS[i] = repeat("ACGT", 100); | 11:54 |
blackburn | something like that | 11:54 |
blackburn | then construct NEG | 11:55 |
blackburn | and then construct POSNEG | 11:55 |
blackburn | you can't do that POS+NEG in java | 11:55 |
sploving1 | you mean POS is 1 * x array | 11:55 |
sploving1 | NEG is too | 11:55 |
blackburn | yes it is :) | 11:56 |
sploving1 | POS + NEG is 1* 2x array | 11:56 |
blackburn | yes | 11:56 |
blackburn | when constructing posneg you could just set first x elements with according pos' ones | 11:57 |
blackburn | and last x with according neg's | 11:57 |
sploving1 | String POSNEG[] = new String[2]; POSNEG[0] = POS; POSNEG[1] = NEG | 11:59 |
sploving1 | ? | 11:59 |
sploving1 | blackburn, am I right? | 11:59 |
blackburn | no | 12:00 |
blackburn | String[] pos = new String[x]; | 12:01 |
blackburn | String[] neg = new String[x]; | 12:01 |
blackburn | String[] posneg = new String[2*x]; | 12:01 |
sploving1 | then a for look will be ok? | 12:02 |
sploving1 | thx blackburn | 12:02 |
sploving1 | I gtg bye | 12:02 |
-!- sploving1 [~sploving@124.16.139.134] has left #shogun [] | 12:02 | |
bettyboo | kiss sploving1 | 12:02 |
blackburn | Shogun® java™ consulting program | 12:03 |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Quit: Ooops..!!] | 12:17 | |
-!- heiko1 [~heiko@main.uni-duisburg.de] has joined #shogun | 12:56 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Read error: Operation timed out] | 13:06 | |
@sonney2k | blackburn, hmmhh I have a fix for that java thing - guess what the solution is? | 13:13 |
blackburn | sonney2k: import something? | 13:13 |
@sonney2k | include / import yes :) | 13:13 |
bettyboo | interesting | 13:13 |
blackburn | what a surprise :D | 13:13 |
@sonney2k | blackburn, the reason is simply that LibSVM is derived from KernelMachine | 13:14 |
@sonney2k | and that is all defined in classifier | 13:14 |
@sonney2k | now LibSVR (regression!!!) is derived from LiBSVM | 13:14 |
@sonney2k | so java needs to import classifier stuff to do the upcast from LibSVR -> SVM | 13:14 |
blackburn | kinda intricated affairs | 13:15 |
@sonney2k | don't like that ... I would prefer that one does not need to import classifiers to use regression methofs | 13:15 |
@sonney2k | methods | 13:15 |
@sonney2k | like it is in python | 13:15 |
@sonney2k | but sigh - I don't see any workaround other than having SVM not in classifiers but something else | 13:16 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 13:19 | |
CIA-32 | shogun: Soeren Sonnenburg master * rcfc559d / (4 files in 2 dirs): | 13:22 |
CIA-32 | shogun: (temporary?) fix for java & regression: since LibSVR etc are derived | 13:22 |
CIA-32 | shogun: from SVM (which is in module classifiers) we need to import Classifiers | 13:22 |
CIA-32 | shogun: too (in Regression.i and also in the regression*.java examples) to fix | 13:22 |
CIA-32 | shogun: compilation issues. - http://bit.ly/nNBw0Y | 13:22 |
@sonney2k | heiko1, all good with you swig fiddling? | 13:22 |
heiko1 | sonney2k, hi, yes makeing progress but it is fiddeling indeed :) | 13:23 |
blackburn | is there any LU expert? :D | 13:23 |
@sonney2k | ask early and often is the only thing that can rescue you from becoming mad - and I know what I am talking about | 13:23 |
heiko1 | yes, I will | 13:24 |
heiko1 | problem is currently that I dont have a large overview how all this works | 13:24 |
heiko1 | but its getting better | 13:24 |
heiko1 | nearly have the example running | 13:24 |
heiko1 | blackburn, LU factorization? | 13:24 |
blackburn | heiko1: yeap | 13:24 |
heiko1 | blackburn, I did a course in numerical mathematics in spring, we did LU, I know a little bit about it, what is your problem? | 13:25 |
blackburn | I got an ill-posed matrix with det~e-12 | 13:25 |
blackburn | I have to do invers | 13:25 |
blackburn | e | 13:25 |
blackburn | as I can see lapack do it with LU | 13:25 |
blackburn | then L^-1 U^-1 | 13:25 |
blackburn | the result is completely wrong | 13:25 |
blackburn | the result is non-sym matrix while given is sym | 13:26 |
heiko1 | mmh | 13:26 |
blackburn | octave gave me matrix with elements ~ e+05 | 13:27 |
heiko1 | perhaps try it with another method? or implementation? | 13:27 |
blackburn | hmm I will try dpotri now | 13:27 |
blackburn | cholesky inverse | 13:27 |
blackburn | hmm | 13:28 |
blackburn | no, I can't do it in wrapper | 13:28 |
blackburn | it makes me mad | 13:28 |
heiko1 | i understand :( | 13:28 |
blackburn | heiko1: I don't know if LU inverse is stable for this problem | 13:28 |
heiko1 | yes, that might be the problem | 13:29 |
blackburn | I guess it should be, right? | 13:29 |
blackburn | well det(L) is 1 | 13:29 |
blackburn | det(U) is product of diag | 13:29 |
blackburn | so there shouldn't be det of e-10 | 13:29 |
heiko1 | if the matrix is pos definite you might use cholseky, wikipedia says it is more stable | 13:31 |
heiko1 | and with the product of diagonal entries I think you have to be carefull when numerical errors are in the game. But I dont remember perfectly | 13:31 |
blackburn | well I do wrapper for symmetrical eigenproblems | 13:32 |
blackburn | I can't be sure pos-def will be given | 13:32 |
heiko1 | oh ok :( | 13:32 |
blackburn | arpack uses inverse communication | 13:33 |
blackburn | and on each iteration I should solve system Aw=v | 13:33 |
blackburn | where w, v will be given to me | 13:33 |
blackburn | and A is given matrix | 13:33 |
blackburn | so the good way to do that is to factorize A | 13:33 |
blackburn | or compute A^-1 | 13:34 |
heiko1 | yes, indeed | 13:34 |
blackburn | I guess inverse is bad way | 13:34 |
blackburn | LLE gives me matrices with ~0 dets | 13:34 |
* heiko1 will get a coffe | 13:36 | |
blackburn | heiko1: hey is it a standard practice that students does a courses? | 13:36 |
heiko1 | what do you mean? | 13:37 |
blackburn | you take it or teach it? | 13:37 |
blackburn | you give a lessons, right? | 13:37 |
heiko1 | I took numerical maths, but I also give lessons in discrete mathematics, not the lecture but the exercises | 13:37 |
heiko1 | its usual here that student tutors do this | 13:38 |
blackburn | I see | 13:38 |
heiko1 | so, I am off for coffee :) see you later | 13:38 |
blackburn | see you | 13:38 |
blackburn | oh haha | 13:40 |
blackburn | http://f.asset.soup.io/asset/2119/0975_2d73.jpeg | 13:40 |
-!- f-x [~user@117.192.192.180] has joined #shogun | 13:41 | |
@sonney2k | blackburn, :D | 13:43 |
blackburn | sonney2k: do you know how to tell arpack that matrix is already factorized? ;( | 13:45 |
blackburn | ah | 13:45 |
blackburn | lapack | 13:45 |
blackburn | not arpack | 13:45 |
* blackburn is crying | 13:45 | |
@sonney2k | blackburn, sorry but no... I guess you should ask on some lapack/blas/atlas mailinglist and do sth else to recover | 13:45 |
blackburn | okaaaay | 13:46 |
blackburn | hm | 13:48 |
blackburn | I think it is not possible! | 13:48 |
@sonney2k | blackburn, and I don't mean vodka this time | 13:48 |
blackburn | sonney2k: yes but I already took two glasses of vodka this morning | 13:49 |
@sonney2k | blackburn, of course - you are Russian :D | 13:51 |
bettyboo | he | 13:51 |
@sonney2k | it is widely known that you brush your teeth with vodka! | 13:51 |
blackburn | my day begins from power on my nuclear missile site, then I brush my teeth with vodka, play some balalayka and clean my house roof out of snow | 13:52 |
blackburn | after I have to walk out with bear and make some human rights violation :D | 13:53 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 13:53 | |
@sonney2k | blackburn, like burning US flags I suppose | 13:55 |
@sonney2k | exactly what I expected | 13:55 |
blackburn | yes, I burn US flags every day | 13:56 |
blackburn | as well as haunting chodorkovsky :D | 13:56 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 244 seconds] | 13:57 | |
heiko1 | lol whats going on here? :) | 14:01 |
heiko1 | sonney2k, i got a question on this python/swig stuff | 14:01 |
heiko1 | I want to do a parameter tree example | 14:01 |
@sonney2k | heiko1, normal day-to-day madness | 14:02 |
@sonney2k | shoot | 14:02 |
heiko1 | in c++ i create all node with "new" and then when the root node is deleted, the tree is deleted automatically | 14:02 |
heiko1 | in python, i create all nodes via the normal constructor | 14:02 |
heiko1 | but then they are deleted automatically | 14:02 |
heiko1 | and then, the root node tries to delete them again | 14:02 |
heiko1 | segfault | 14:02 |
heiko1 | what to do against this? | 14:02 |
heiko1 | cann I tell python not to delete variables? | 14:03 |
heiko1 | (I want to translate the modelselection_parameter_tree example to python() | 14:03 |
@sonney2k | heiko1, you need to use the SG_REF mechanism in C++ | 14:04 |
@sonney2k | that is within your tree SG_REF the nodes that are in there | 14:04 |
@sonney2k | and SG_UNREF when you are done | 14:05 |
heiko1 | ah ok | 14:05 |
heiko1 | when appending child nodes | 14:05 |
heiko1 | alright, that saved some time :) | 14:05 |
@sonney2k | heiko1, basically everything that you keep a reference on (and is derived from SGObject) your should SG_REF | 14:06 |
@sonney2k | and never use delete but SG_UNREF | 14:06 |
heiko1 | yes, | 14:06 |
heiko1 | i usually do this, you already told me, but i forgot about the tree structure | 14:06 |
@sonney2k | swig then automagically will increase the refcount for the objects it still holds | 14:06 |
@sonney2k | and whoever calls unref last will cause the object to die | 14:07 |
heiko1 | ok, will try now | 14:07 |
heiko1 | ehm. | 14:08 |
heiko1 | I use delete this at some point | 14:08 |
heiko1 | maybe have to think about this some more | 14:08 |
@sonney2k | delete this? | 14:09 |
@sonney2k | heiko1, won't that kill the object under your feet? | 14:09 |
heiko1 | yes | 14:10 |
heiko1 | its for tree destruction | 14:10 |
blackburn | delete this?! | 14:10 |
blackburn | :D | 14:10 |
heiko1 | yes, I have a destroy method | 14:10 |
heiko1 | that does something other than destructor | 14:10 |
blackburn | uff | 14:10 |
heiko1 | destructor only deletes node, destroy deleted tree of which node is the root | 14:10 |
blackburn | why not make destructor capable of this? | 14:11 |
heiko1 | I want to be able to choose whether tree gets destroyed or not | 14:12 |
blackburn | hmm | 14:12 |
heiko1 | well, I will think of it a bit now, perhaps I will change | 14:12 |
heiko1 | I changed some stuff afterwards, perhaps its possible to do it with only the destructor nwo | 14:13 |
heiko1 | now | 14:13 |
blackburn | delete this looks strange to me | 14:13 |
heiko1 | yes its a bit esoterical :) | 14:13 |
heiko1 | but its possible | 14:13 |
blackburn | sonney2k: I have one suggestion: lets make shogun version in assembly language? | 14:14 |
heiko1 | omg :D | 14:14 |
heiko1 | please not | 14:14 |
blackburn | :D | 14:14 |
heiko1 | building products of sets of trees in assembler ;) | 14:14 |
@sonney2k | blackburn, I prefer one that is totally written in java script | 14:14 |
blackburn | oh yes | 14:15 |
blackburn | javascript works too | 14:15 |
heiko1 | what about google go if we already are paid by google :) | 14:15 |
blackburn | hm | 14:17 |
blackburn | I have an idea | 14:17 |
@sonney2k | too easy | 14:17 |
blackburn | how to involve more developers | 14:17 |
blackburn | PHP! | 14:17 |
blackburn | :D | 14:17 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 14:20 | |
blackburn | heiko1: I guess I found a solution to my trouble | 14:24 |
heiko1 | blackburn, ok, what did you do? | 14:25 |
blackburn | that is the most applied mathematician did if they can't solve something | 14:25 |
blackburn | :D | 14:25 |
blackburn | I just will modify diag of matrix | 14:25 |
blackburn | regularize it | 14:25 |
heiko1 | hehe | 14:25 |
heiko1 | lol that reminds me of a funny hint, a guy once gave me | 14:25 |
heiko1 | for my bachelor thesis I firstly used kernlab, an R implementation of SVM | 14:26 |
heiko1 | and it froze regulary | 14:26 |
heiko1 | I asked the developer | 14:26 |
heiko1 | He tolds me that due to numerical errors, the kernel matrix had some smal smal negative values | 14:26 |
blackburn | haha | 14:26 |
heiko1 | his solution was to add 0.1 to every diag entry :) | 14:27 |
heiko1 | that solved the problem for some cases, but not for every one, | 14:27 |
heiko1 | thats actually why i switched to shogun :) | 14:27 |
blackburn | I have used similar technique in LLE already: when k>dim covariance matrix is singular | 14:27 |
heiko1 | wel"technique is a nice word for that :) | 14:28 |
-!- in3xes1 [~in3xes@180.149.49.227] has joined #shogun | 14:36 | |
@sonney2k | heiko1, actually we do the same thing... | 14:37 |
@sonney2k | all this numerical business is so damn wobbly / unstable or whatever you want to call it | 14:37 |
@sonney2k | and every solver has its own oddities / problems | 14:39 |
@sonney2k | In your case I suspect your data was not in a 0-1 ball / not nicely normalized... | 14:39 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 14:39 | |
-!- heiko1 [~heiko@main.uni-duisburg.de] has quit [Ping timeout: 258 seconds] | 14:39 | |
@sonney2k | I mean try a polynomial kernel on unnormalized data | 14:39 |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 258 seconds] | 14:40 | |
@sonney2k | kaboom 0.0001**12 = 0 | 14:40 |
@sonney2k | or 1000**12 == inf - well almost :) | 14:40 |
bettyboo | 8) | 14:40 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 14:52 | |
f-x | sonney2k: copied my online sgd on top of shogun's and made a new pull request (not for merging, only for comparing).. the backend stuff is always open for changes, but i guess the code within the online version should remain more or less similar to the version i've committed | 14:59 |
f-x | Although, again - I'm in favour of separate online/batch versions | 15:00 |
@sonney2k | f-x, I wonder if we should create a new directory for online methods or just keep them in classifier etc | 15:00 |
@sonney2k | I am not | 15:00 |
@sonney2k | :) | 15:00 |
@sonney2k | I mean code duplication is a mess | 15:01 |
f-x | sonney2k: but as you pointed out, the batch version can take advantage of stuff like lambda=1/num_vec | 15:01 |
f-x | we'd be losing all that | 15:01 |
@sonney2k | f-x, not really... features defines get_num_vectors() | 15:02 |
@sonney2k | if it is != -1 you can use it | 15:02 |
f-x | sonney2k: having those checks every time - i guess the code would look messy | 15:03 |
@sonney2k | and surely we only do this for online methods like sgd / liblinear | 15:03 |
f-x | but as always, there might be a way to do it cleanly | 15:03 |
f-x | sonney2k: and i expect that the batch techniques are already faster when we give input through say a SimpleFeatures feature matrix | 15:04 |
f-x | than the whole conversion to streaming features | 15:04 |
@sonney2k | f-x, I don't think so | 15:04 |
@sonney2k | not at all ... | 15:04 |
@sonney2k | I mean you just pass pointers / everything is implemented in *.h files - the compiler will do the rest | 15:05 |
f-x | i hope so - anyway i think there's a good enough scope for optimization for that which i haven't exploited yet.. that should speed up things when done | 15:08 |
@sonney2k | f-x, your patch looks good to me btw. | 15:08 |
@sonney2k | I would suggest to first have an OnlineMachine/StreamingMachine etc | 15:09 |
@sonney2k | and then derive the things from there | 15:09 |
@sonney2k | but I guess you did already | 15:09 |
f-x | sonney2k: i did Machine->OnlineLinearMachine directly | 15:09 |
@sonney2k | for now I am also fine with having OnlineSGD | 15:10 |
@sonney2k | and when that works to try it out on some real examples and then do real benchmarks comparing it to the current SGD | 15:10 |
f-x | sonney2k: right. this test version i have right now works, but i need to benchmark and make changes in the parser and stuff depending on the results | 15:12 |
-!- heiko1 [~heiko@main.uni-duisburg.de] has joined #shogun | 15:12 | |
f-x | so for the time being I will throw in my OnlineLinearMachine and OnlineSVMSGD separately and make a pull request soon | 15:13 |
@sonney2k | f-x, full ack | 15:13 |
f-x | sonney2k: for a non-native, you know plenty of english colloquial stuff :) | 15:14 |
f-x | ooh but full ack has a german origin, says google | 15:16 |
blackburn | volle zustimmung :D | 15:17 |
@sonney2k | f-x, hmmh question now is - is this good or bad | 15:18 |
f-x | sonney2k: separate classes? | 15:19 |
@sonney2k | f-x, get thsi data set here http://largescale.ml.tu-berlin.de/submission/evaluation/1/5/ | 15:19 |
@sonney2k | and try to do the sgd experiment | 15:19 |
@sonney2k | if you get around 0.14 aoPRC we are fine | 15:19 |
@sonney2k | f-x, slang | 15:20 |
@sonney2k | f-x, btw instructions are here http://largescale.ml.tu-berlin.de/instructions/ | 15:23 |
-!- f-x` [~user@117.192.200.70] has joined #shogun | 15:23 | |
f-x` | sonney2k: okay - got the dataset link | 15:24 |
@sonney2k | f-x`, and data here ftp://largescale.ml.tu-berlin.de/largescale | 15:24 |
@sonney2k | alpha is the smalles data set ... | 15:24 |
@sonney2k | 500,000 examples | 15:25 |
@sonney2k | so you could train on 400k examples | 15:25 |
@sonney2k | and measure performance on the remaining 100k | 15:25 |
-!- f-x [~user@117.192.192.180] has quit [Ping timeout: 260 seconds] | 15:25 | |
f-x` | sonney2k: do you have the results for the batch version? | 15:26 |
@sonney2k | f-x__, f-x` - look on the website or what I wrote above: 0.14 aoPRC should be OK | 15:26 |
heiko1 | sonney2k, could you help me with this: | 15:26 |
heiko1 | i got a c++ method get_combinations(DynArray<ParameterCombination*> c) | 15:26 |
heiko1 | i got a swig type | 15:27 |
heiko1 | %template(DynamicParameterCombinationPointerArray) DynArray<ParameterCombination*>; | 15:27 |
f-x` | sonney2k: okay, will do | 15:27 |
heiko1 | and i want to do this in python | 15:27 |
heiko1 | combinations=DynamicParameterCombinationPointerArray() | 15:27 |
heiko1 | root.get_combinations(combinations) | 15:27 |
heiko1 | but i get a type error | 15:27 |
@sonney2k | heiko1, you should shorten the name - say ParameterCombinations | 15:27 |
heiko1 | ok will do.. | 15:27 |
heiko1 | but still: | 15:27 |
heiko1 | TypeError: in method 'ModelSelectionParameters_get_combinations', argument 2 of type 'DynArray< shogun::ParameterCombination * > &' | 15:27 |
@sonney2k | but this works: combinations=DynamicParameterCombinationPointerArray() ? | 15:28 |
heiko1 | yes | 15:28 |
@sonney2k | yeah | 15:28 |
@sonney2k | there is a ptr in this | 15:28 |
heiko1 | ? | 15:29 |
@sonney2k | let me check | 15:29 |
heiko1 | defi nition is : | 15:29 |
heiko1 | void CModelSelectionParameters::get_combinations( | 15:30 |
heiko1 | DynArray<ParameterCombination*>& result) | 15:30 |
@sonney2k | heiko1, yeah that won't work | 15:30 |
@sonney2k | you should do DynArray<ParameterCombination*>* get_whatever() | 15:30 |
@sonney2k | there is not typemap for the & stuff | 15:31 |
heiko1 | why is the reference stuff not possible? | 15:31 |
@sonney2k | ^ | 15:31 |
@sonney2k | you need to return things... | 15:31 |
heiko1 | is it possible to return the original DynArray which was given in the parameter? | 15:32 |
@sonney2k | heiko1, I don't understand | 15:36 |
@sonney2k | you have DynArray<ParameterCombination*>* get_whatever( DynArray<ParameterCombination*>* p) | 15:36 |
@sonney2k | and want to return p ? | 15:36 |
heiko1 | yes, is that possible? | 15:36 |
@sonney2k | yes why not | 15:37 |
heiko1 | ok | 15:37 |
heiko1 | thanks | 15:37 |
@sonney2k | but I see one problem this array won't do SG_REF/UNREF | 15:37 |
heiko1 | mmh ok | 15:38 |
heiko1 | ParameterCombination is not an SGObject | 15:38 |
@sonney2k | why not? | 15:39 |
@sonney2k | I mean you can do it w/o for testing now but you will run into memory related trouble | 15:39 |
@sonney2k | memory leak / crash | 15:40 |
heiko1 | i did this because it is not serializable | 15:47 |
heiko1 | but ok you are right, I can add this anyways | 15:47 |
@sonney2k | heiko1, well you don't have to register any parameters for serializaton | 15:55 |
heiko1 | I will change this later then. | 15:57 |
@sonney2k | so adding a C prefix won't add marginal overhead (reference count and say 70 bytes) | 15:57 |
heiko1 | First want to get it working :) | 15:57 |
heiko1 | ok alright then | 15:57 |
heiko1 | mmh | 15:57 |
heiko1 | Then i might also use DynamicArray instead of DynArray | 15:57 |
@sonney2k | yes | 15:57 |
heiko1 | *sigh* | 15:57 |
@sonney2k | but that is compatible or? | 15:57 |
heiko1 | then I have to rethink of all the memory stuff | 15:57 |
@sonney2k | well except for SG_UNREF stuff | 15:57 |
heiko1 | yes, | 15:57 |
heiko1 | ok will do this at some point | 15:58 |
heiko1 | but for now I want to get this python example working | 15:58 |
@sonney2k | that is admittedly painful :( | 15:58 |
@sonney2k | you need to review the whole code then | 15:58 |
@sonney2k | heiko1, if you try it now - you will see random crashes I suspect | 15:58 |
heiko1 | You know I used this get_combinations(DynArray...) stuff because then I could use stack variables and did not have to worry about all this memory | 15:58 |
@sonney2k | heiko1, well it is not a good idea to use stack variables (at least not too many of them) ... we had at some point weird crashes because we were running out of stack | 16:00 |
heiko1 | oh really? | 16:00 |
heiko1 | ok then its better anyways :) | 16:00 |
@sonney2k | but I agree that it is easyer | 16:00 |
@sonney2k | heiko1, well we had arrays on stack - that is sth one just shouldn't do | 16:01 |
heiko1 | ok, did not know this | 16:01 |
@sonney2k | ulimit -s | 16:01 |
@sonney2k | 8192 | 16:01 |
@sonney2k | that is 8M stack on my machine | 16:01 |
@sonney2k | and we used things like double[1000] | 16:02 |
heiko1 | ok same here | 16:02 |
heiko1 | ok, I just learned something new :) | 16:02 |
@sonney2k | or what is even worse we used double x[num] | 16:03 |
@sonney2k | with num being a parameter | 16:03 |
@sonney2k | that is something I would consider a bug | 16:03 |
@sonney2k | and it is not compiling even under non-gcc compilers | 16:03 |
blackburn | time to rewrite shogun | 16:04 |
blackburn | time to bomb saddam | 16:04 |
heiko1 | ok, I will then now change all this stuff | 16:04 |
heiko1 | also will use DynamicArray | 16:04 |
blackburn | I've got a permanent pain in ass with arpack | 16:05 |
heiko1 | pain in the ass everywhere today ;) | 16:05 |
blackburn | the spirit of pain in the ass | 16:05 |
@sonney2k | I think only f-x__ / f-x` is feeling fine today :) | 16:05 |
heiko1 | uhuh thunderstorm coming in here .... damn I wanted to go to a barbecue later :( | 16:06 |
blackburn | we have to invent some pain in ass for him too | 16:06 |
@sonney2k | blackburn, he is just through this - give him a rest :) | 16:06 |
f-x` | sonney2k, blackburn : not so much! i have to test on a huge dataset | 16:06 |
@sonney2k | heiko1, great weather today finally in Berlin | 16:07 |
blackburn | f-x' you have segfault | 16:07 |
blackburn | *joke* | 16:07 |
blackburn | :D | 16:07 |
heiko1 | same here until two hours ago .... | 16:07 |
f-x` | blackburn: i;m sure i will :) | 16:07 |
* sonney2k takes out his vodoo computer | 16:07 | |
* sonney2k takes a needle | 16:07 | |
* sonney2k inserts needle into WHAT'S THAT SMELL | 16:07 | |
blackburn | sonney2k: is that you called synergy effects? | 16:07 |
@sonney2k | and I am serious this time | 16:07 |
heiko1 | uhoh | 16:07 |
blackburn | synergy pain in the ass | 16:08 |
@sonney2k | I got this as a present | 16:08 |
@sonney2k | for heiko1 lightning storm is more appropriate though | 16:08 |
blackburn | we've got x10 synergy pain in the ass | 16:08 |
heiko1 | lol ;) | 16:08 |
bettyboo | smart | 16:08 |
@sonney2k | bettyboo, nope not smart | 16:10 |
bettyboo | sonney2k: nope...irritated :| | 16:10 |
@sonney2k | exactly | 16:10 |
* blackburn is praying on DGETRI | 16:13 | |
blackburn | DGETRI cult | 16:14 |
blackburn | fail :D | 16:20 |
bettyboo | :^) blackburn | 16:20 |
blackburn | gtg | 16:26 |
-!- blackburn [~blackburn@188.122.238.99] has quit [Quit: Leaving.] | 16:26 | |
-!- f-x` [~user@117.192.200.70] has quit [Remote host closed the connection] | 16:32 | |
-!- sploving1 [~sploving@2001:cc0:2020:2022:5eff:35ff:fe04:f091] has joined #shogun | 16:41 | |
@sonney2k | sploving1, seen the patch / fix/ workaround? | 16:45 |
@sonney2k | at least your code compiles now for regression | 16:45 |
sploving1 | oh. thx. | 16:46 |
sploving1 | I have not git fetch now. I am test other example: test_check_commonwordkernel_memleak_modular | 16:47 |
sploving1 | sonne2k, have you seen my pushed example. | 16:48 |
CIA-32 | shogun: Baozeng Ding master * r7850285 / examples/undocumented/java_modular/tests_check_commwordkernel_memleak_modular.java : add tests_check_commonwordkernel_memleak_modular example - http://bit.ly/oDwxob | 16:48 |
CIA-32 | shogun: Soeren Sonnenburg master * rdb4dbc6 / examples/undocumented/java_modular/tests_check_commwordkernel_memleak_modular.java : | 16:48 |
CIA-32 | shogun: Merge pull request #175 from sploving/master | 16:48 |
CIA-32 | shogun: add memleak example - http://bit.ly/qcy2Th | 16:48 |
@sonney2k | yes | 16:48 |
sploving1 | but there maybe some mem leak | 16:48 |
sploving1 | javac could work | 16:49 |
sploving1 | but java get error | 16:49 |
@sonney2k | ?? | 16:49 |
@sonney2k | sorry I don't understand what you are saying | 16:50 |
sploving1 | sonney2k, since you called it memleak_modular, you could know where it may leak memory | 16:50 |
@sonney2k | sploving1, I used this to check if it leaks memory | 16:50 |
@sonney2k | and fixed the leak back then | 16:50 |
@sonney2k | so now it shouldn't | 16:50 |
sploving1 | I mean javac tests_.. .java could compile well. but when I run java ... commond | 16:51 |
sploving1 | A fatal error has been detected by the Java Runtime Environment: | 16:51 |
sploving1 | # | 16:51 |
sploving1 | # SIGSEGV (0xb) at pc=0xb6e9755a, pid=17037, tid=3064814448 | 16:51 |
@sonney2k | sploving1, hmmhh - please run it through valgrind | 16:51 |
@sonney2k | and see where it dies | 16:52 |
sploving1 | valgrind? | 16:52 |
sploving1 | a tool? | 16:52 |
@sonney2k | sploving1, yes | 16:52 |
@sonney2k | call it like | 16:53 |
@sonney2k | valgrind PROGRAM | 16:53 |
@sonney2k | e.g. /usr/bin/ls | 16:53 |
@sonney2k | it will trace all memory allocs and detect leaks etc | 16:53 |
sploving1 | how to install the tool | 16:53 |
@sonney2k | uninited memory leaks | 16:53 |
sploving1 | how to run it? | 16:53 |
@sonney2k | http://www.google.com/search?q=valgrind | 16:53 |
@sonney2k | http://en.wikipedia.org/wiki/Valgrind | 16:54 |
sploving1 | I am fraind I have no time to learn it now. there are still 60 exampels remaining. I need add ten everyday in avange | 16:57 |
@sonney2k | sploving1, ok then just go ahead | 16:57 |
-!- in3xes [~in3xes@180.149.49.227] has quit [Quit: Ooops..!!] | 16:57 | |
sploving1 | I will debug it when all are in good shape | 16:58 |
@sonney2k | sploving1, OK | 16:59 |
@sonney2k | sploving1, it crashes in this line traindat.set_features(POSNEG); | 17:31 |
@sonney2k | So I suspect the typemaps for strings are not yet working? | 17:31 |
sploving1 | sonne2y, oh, I will take a look at it | 17:31 |
-!- sploving1 [~sploving@2001:cc0:2020:2022:5eff:35ff:fe04:f091] has left #shogun [] | 17:37 | |
CIA-32 | shogun: Baozeng Ding master * re04e91c / (6 files): add some preprocessor examples - http://bit.ly/qr6FNJ | 17:52 |
CIA-32 | shogun: Soeren Sonnenburg master * r172024e / (6 files): | 17:52 |
CIA-32 | shogun: Merge pull request #176 from sploving/master | 17:52 |
CIA-32 | shogun: add some preprocessor examples - http://bit.ly/qPM8xh | 17:52 |
-!- in3xes1 is now known as in3xes | 17:55 | |
-!- [1]warpy [~warpy@bzq-79-179-0-217.red.bezeqint.net] has joined #shogun | 18:33 | |
* sonney2k is applying hefty force to some pine apple | 19:58 | |
-!- blackburn [~blackburn@188.122.238.99] has joined #shogun | 20:39 | |
-!- [1]warpy [~warpy@bzq-79-179-0-217.red.bezeqint.net] has quit [Quit: HydraIRC -> http://www.hydrairc.com <- The alternative IRC client] | 20:51 | |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 21:26 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Read error: Operation timed out] | 21:28 | |
blackburn | HOOOOOORRRRRRRRRRRRRRAAAAAAAAAAAAAYY | 21:31 |
@sonney2k | blackburn, ass problem solved :D | 21:38 |
blackburn | sonney2k: half-ass | 21:38 |
blackburn | it is slow and wrong anyway :D | 21:38 |
blackburn | but now I know | 21:38 |
blackburn | always call dgetrf before dgetri | 21:39 |
blackburn | 1500 iterations hehe | 21:41 |
blackburn | who stole convergence | 21:41 |
-!- f-x [~user@117.192.204.35] has joined #shogun | 21:55 | |
@sonney2k | f-x, how is sgd doing? | 22:10 |
f-x | sonney2k: haven't tested it yet - finished downloading the dataset.. uncompressing it right now.. | 22:12 |
f-x | some additions have to be made | 22:12 |
f-x | to take labels and features from separate files | 22:12 |
f-x | or maybe I could join the labels and features files for now, but is that allowed? | 22:13 |
@sonney2k | f-x, or you concatenate the files via the cmdline command 'paste' | 22:13 |
f-x | sonney2k: yes, exactly.. i'll follow that till the new interface is ready | 22:13 |
blackburn | 10Gb database? :D | 22:14 |
f-x | blackburn: not 10, but 2.3 GB training data | 22:15 |
blackburn | oh sh | 22:15 |
blackburn | really? | 22:15 |
blackburn | :D | 22:15 |
bettyboo | :) | 22:15 |
blackburn | sonney2k: another one celebration - java 7 - I simply can't live without vodka | 22:15 |
blackburn | http://damnyouautocorrect.com/images/fuhrer1.jpg | 22:17 |
@sonney2k | blackburn, tiny data sets for online methods... | 22:20 |
blackburn | sonney2k: you know my knowledge about online learning tends to zero :) | 22:20 |
@sonney2k | I did batch learning with datasets of 20G in size... | 22:20 |
blackburn | what was it?? | 22:20 |
@sonney2k | just DNA | 22:21 |
blackburn | ehh | 22:21 |
blackburn | sonney2k: I thought human dna is ~850 mb | 22:21 |
@sonney2k | 3 G yes | 22:21 |
blackburn | was it an alien? :D | 22:22 |
@sonney2k | but when you extract windows around certain positions you can explode that | 22:22 |
blackburn | ahh | 22:22 |
@sonney2k | (unfortunately) | 22:22 |
@sonney2k | but yes of course one could have loaded the whole dna and then just specify the centers of the window | 22:22 |
@sonney2k | ...which is now supported in shogun (obtain_by_positionlist :) | 22:22 |
blackburn | sonney2k: is it true say tomato have more nucleotides in dna than us? | 22:23 |
@sonney2k | many organisms have more | 22:23 |
blackburn | why? ain't we the most cool ones? :D | 22:23 |
blackburn | we have complex splicing and etc, right? | 22:23 |
@sonney2k | but amoeba is the winner :) | 22:23 |
bettyboo | ^_^ | 22:24 |
blackburn | what's up with amoeba? | 22:24 |
@sonney2k | 670 billion base pairs! | 22:24 |
@sonney2k | we have just 3 billion | 22:24 |
blackburn | why it have so many? | 22:24 |
@sonney2k | because it only has one cell? | 22:24 |
@sonney2k | blackburn, I don't know the masterplan behind this | 22:25 |
@sonney2k | maybe spaghetti coding | 22:25 |
blackburn | sonney2k: why are you asking, I was drinking vodka instead of biology :D | 22:25 |
blackburn | how many genes it have? | 22:25 |
@sonney2k | blackburn, needs some blackburn kalaschnikov refactoring | 22:25 |
@sonney2k | no idea | 22:26 |
blackburn | калашников refactoring? :) | 22:26 |
blackburn | I mean how it can have much lesser genes but much more np? | 22:26 |
blackburn | the most code is junk? | 22:26 |
@sonney2k | blackburn, well I don't know... I would say nicely written code uses lots of subroutines .... if you don't know how to do that you end up with many more LoC | 22:29 |
blackburn | sonney2k: so the code was reducing during this evolving? | 22:31 |
@sonney2k | I would think so... I mean it is a lot easier to write spaghetti code | 22:31 |
@sonney2k | so I would expect evolution to compact things | 22:32 |
blackburn | it seems like entropy is lesser for us, fantastic | 22:32 |
blackburn | sonney2k: is amoeba=its cell more complex than our ones? | 22:35 |
blackburn | we have fantastic generic code :D | 22:35 |
blackburn | with templates I guess :D | 22:35 |
@sonney2k | blackburn, well it is a shapeshifter - I mean it can change its shape | 22:36 |
@sonney2k | that is impressive enoguh | 22:36 |
blackburn | sonney2k: our cells lose this ability? | 22:39 |
@sonney2k | blackburn, I am not a biologist ... | 22:40 |
blackburn | sonney2k: well you know something about it :) | 22:40 |
@sonney2k | still not a lot | 22:41 |
@sonney2k | f-x, but don't hesitate to send a OnlineSGD patch before that | 22:57 |
f-x | sonney2k: thanks - i was wondering about that | 22:58 |
f-x | shall send it in a few hours after some initial testing | 22:59 |
@sonney2k | f-x, yeah just send it if you think you are ready | 22:59 |
f-x | sonney2k: sure | 23:00 |
blackburn | sonney2k: gunnar pointed some bug with classic mds, working on it | 23:14 |
@sonney2k | blackburn, I see his email | 23:14 |
@sonney2k | but it is compiling here | 23:14 |
blackburn | sonney2k: without arpack? | 23:15 |
@sonney2k | ahh didn't try that | 23:15 |
@sonney2k | but the other thing is not an issue imho | 23:15 |
@sonney2k | the include path is correct | 23:15 |
@sonney2k | blackburn, and I don't think sploving has arpack installed ... | 23:16 |
blackburn | sonney2k: hmm yes | 23:16 |
@sonney2k | actually I don't think anyone except you and me | 23:16 |
blackburn | can my test give 'false positive'? :) | 23:17 |
@sonney2k | blackburn, hardly... | 23:18 |
blackburn | asked gunnar if he have arpack or not | 23:19 |
blackburn | and was it detected or not | 23:19 |
@sonney2k | great | 23:20 |
@sonney2k | me too | 23:20 |
blackburn | :D | 23:22 |
@sonney2k | could be some OSX thing though | 23:26 |
@sonney2k | maybe we should buy 2 mac-minis one running osx and one windows as build servers | 23:27 |
blackburn | sonney2k: he said he used ubuntu | 23:27 |
@sonney2k | true | 23:27 |
@sonney2k | ubuntu is crap anyways :D | 23:27 |
blackburn | why? | 23:27 |
blackburn | I hadn't experienced any differences in debian :) | 23:28 |
@sonney2k | even though I am half-way kidding - it is really not particularly stable | 23:28 |
blackburn | returned to 10.04 - at least gnome looks better by default | 23:28 |
@sonney2k | it is a fork of debian unstable... | 23:28 |
blackburn | 10.04 is pretty stable now ;) | 23:28 |
@sonney2k | like debian :) | 23:29 |
blackburn | yes but there are some more developed ehh | 23:29 |
blackburn | ecosphere? :D | 23:29 |
blackburn | I mean software (gui things and etc) more tightly connected here in ubuntu | 23:30 |
@sonney2k | sure ;) | 23:31 |
@sonney2k | blackburn, look you have no chance - you are talking to a debian developer | 23:31 |
blackburn | sonney2k: yeah I understand :) | 23:32 |
blackburn | well I don't like the way of 'modern' ubuntu | 23:32 |
blackburn | unity, software centers, blabla | 23:32 |
blackburn | so I think the next my dist is debian | 23:32 |
@sonney2k | some people here at work use it - some turned away from ubuntu... | 23:33 |
@sonney2k | and it is the first time that I stick to debian stable | 23:33 |
@sonney2k | shogun is unstable enough for me and currently all my attention | 23:34 |
blackburn | sonney2k: will be any gnome 3 debian some day? :) | 23:34 |
@sonney2k | I guess so... | 23:34 |
@sonney2k | I hope by then it is feature complete and usable even by people like us | 23:34 |
@sonney2k | but hey - the gnome2 switch was also a big bummer | 23:35 |
bettyboo | roger | 23:35 |
@sonney2k | bettyboo, that | 23:35 |
bettyboo | sonney2k: I guess there are scripts doing that though | 23:35 |
blackburn | ahaha | 23:36 |
blackburn | sonney2k: btw my laptop brightness doesn't work neither in debian or ubuntu | 23:36 |
@sonney2k | blackburn, samsung? | 23:37 |
blackburn | yes | 23:37 |
@sonney2k | this got fixed in kernel 2.6.39 | 23:37 |
blackburn | really? | 23:37 |
blackburn | sonney2k: do you have samsung? :) | 23:38 |
@sonney2k | a netbook | 23:38 |
@sonney2k | but working on a macbook pro running debian usually | 23:39 |
blackburn | you have installed debian on macbook?! | 23:39 |
@sonney2k | the superior OS - yes of course | 23:41 |
blackburn | why not osx?? | 23:42 |
@sonney2k | osx is too difficult to use | 23:42 |
@sonney2k | I mean not when just clicking around - but when coding | 23:42 |
blackburn | hmm I se | 23:42 |
blackburn | e | 23:42 |
@sonney2k | just try once to get all the dependencies installed | 23:42 |
@sonney2k | takes years | 23:43 |
@sonney2k | and then I cannot move windows with alt+mouse etc | 23:43 |
blackburn | sonney2k: no news from gunnar.. :) | 23:54 |
@sonney2k | heiko1, what is going on? you are still online? | 23:54 |
@sonney2k | blackburn, I guess we muted him | 23:54 |
blackburn | i'm wondering how can heiko be online if he uses internet at university and he have no internet access at home | 23:56 |
@sonney2k | blackburn, exactly... | 23:56 |
--- Log closed Thu Jul 07 00:00:52 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!