--- Log opened Wed Jun 20 00:00:41 2012 | ||
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] | 00:01 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has joined #shogun | 00:33 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has quit [Changing host] | 00:33 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 00:33 | |
-!- wiking_ [~wiking@108.170.25.86] has joined #shogun | 03:21 | |
-!- wiking_ [~wiking@108.170.25.86] has quit [Changing host] | 03:22 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 03:22 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 248 seconds] | 03:23 | |
-!- wiking_ is now known as wiking | 03:23 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 03:27 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 265 seconds] | 03:28 | |
-!- wiking_ is now known as wiking | 03:28 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 07:09 | |
n4nd0 | wiking: hey, what's up? how is it going with the PrimalMosekSOSVM example? I read in the logs you were running it | 07:12 |
---|---|---|
@sonney2k | n4nd0 long time no see | 07:24 |
@sonney2k | I've merged your stuff | 07:25 |
@sonney2k | Infrastructure wise I would need to know which functions should return specialized types | 07:26 |
n4nd0 | sonney2k: hi, apply should return an structure type probably | 07:30 |
n4nd0 | not a CStructuredLabels* but a subtype of it, e.g. CMulticlassSOLabels* | 07:31 |
n4nd0 | also the get_label(idx) of CStructuredLabels* | 07:31 |
@sonney2k | Will you be here in the evening ? | 07:31 |
@sonney2k | We can then easily fix those | 07:32 |
n4nd0 | I can do it if you want | 07:32 |
n4nd0 | ok, tell me when and I'll be here | 07:32 |
@sonney2k | After 19 hrs | 07:32 |
n4nd0 | all right, I will be around from 19h then | 07:33 |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 240 seconds] | 07:56 | |
-!- nji [31ca5015@gateway/web/freenode/ip.49.202.80.21] has joined #shogun | 08:43 | |
-!- zxtx [~zv@cpe-98-154-83-76.socal.res.rr.com] has joined #shogun | 08:47 | |
nji | hi!! does anyone know how to use img(erdas imagine raw format) images? | 08:47 |
-!- uricamic [~uricamic@2001:718:2:1634:5d58:8730:47cd:6cf8] has joined #shogun | 08:59 | |
-!- nji [31ca5015@gateway/web/freenode/ip.49.202.80.21] has quit [Ping timeout: 245 seconds] | 09:34 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 09:35 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has joined #shogun | 09:37 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has quit [Changing host] | 09:37 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 09:37 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 09:50 | |
-!- zxtx [~zv@cpe-98-154-83-76.socal.res.rr.com] has quit [Ping timeout: 246 seconds] | 10:40 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 265 seconds] | 10:53 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 11:34 | |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun | 12:16 | |
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun [] | 12:17 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has joined #shogun | 12:59 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has quit [Changing host] | 12:59 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 12:59 | |
-!- emrecelikten [~emre@176.40.254.205] has quit [Quit: Leaving.] | 13:39 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 14:25 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has joined #shogun | 14:48 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has quit [Changing host] | 14:48 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 14:48 | |
-!- alexlovesdata [82955843@gateway/web/freenode/ip.130.149.88.67] has joined #shogun | 15:39 | |
alexlovesdata | it is only a matter of time when I will fail with these too complex captchas | 15:40 |
-!- blackburn [d5578ec7@gateway/web/freenode/ip.213.87.142.199] has joined #shogun | 16:08 | |
-!- blackburn [d5578ec7@gateway/web/freenode/ip.213.87.142.199] has quit [Quit: Page closed] | 16:36 | |
-!- uricamic [~uricamic@2001:718:2:1634:5d58:8730:47cd:6cf8] has quit [Quit: Leaving.] | 16:43 | |
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun | 17:33 | |
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds] | 17:58 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 18:21 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 18:24 | |
n4nd0 | alexlovesdata: hi, how is it going? do you have a moment? | 18:26 |
alexlovesdata | yes i have a moment | 18:28 |
alexlovesdata | did your MC-SVM got better results after your chat with Nico? | 18:29 |
alexlovesdata | the C hack still improves? | 18:32 |
n4nd0 | yeah, with the C hack I am getting results similar to one-vs-rest, what is nice | 18:33 |
n4nd0 | now I am trying to train the SOSVM with bias terms for the multiclass classification example | 18:34 |
n4nd0 | because the % of correct training without bias, about 40%, is too low, and I think it could be achieved even if the implementation is buggy | 18:34 |
@sonney2k | alexlovesdata, get a real IRC client | 18:35 |
n4nd0 | however, with bias one-vs-rest gets about 90% correct classification | 18:35 |
n4nd0 | :D | 18:35 |
n4nd0 | so if I manage to get the same result with SOSVM I will be more confident about the implementation | 18:35 |
@sonney2k | hi n4nd0 | 18:35 |
n4nd0 | sonney2k: hi | 18:35 |
@sonney2k | n4nd0, I am wondering whether you could compare to some reference | 18:36 |
@sonney2k | like some CS MC svm | 18:36 |
@sonney2k | or some matlab code from nico | 18:36 |
@sonney2k | cocococoo | 18:36 |
n4nd0 | I am comparing to liblinear MC, but it is not SO of course | 18:36 |
n4nd0 | Nico told me that he couldn't find his MC SO code :D | 18:37 |
n4nd0 | I can give it a shot with svm-struct, I have already tried their multiclass test locally | 18:37 |
@sonney2k | n4nd0, well then nico should just re-do it | 18:37 |
@sonney2k | it is trivial in matlab... | 18:37 |
@sonney2k | liblinear MC is not really the same thing | 18:37 |
@sonney2k | it uses regularized bias... | 18:38 |
@sonney2k | and it does not really converge too well | 18:38 |
@sonney2k | btw, what epsilon did you give liblinear? | 18:38 |
alexlovesdata | you are right Soeren | 18:38 |
alexlovesdata | I will try to get an irc client | 18:38 |
n4nd0 | sonney2k: epsilon = 1e-3 | 18:39 |
n4nd0 | 10e-3 sorry | 18:39 |
alexlovesdata | svm struct MC could be well for comparison, as I said yesterday :) | 18:40 |
n4nd0 | alexlovesdata, sonney2k if I want to train my sosvm for MC with bias, should I use regularized bias or not? | 18:40 |
alexlovesdata | good question | 18:42 |
@sonney2k | n4nd0, with 'true' bias | 18:42 |
n4nd0 | sonney2k: I thought that liblinear bias not regularized since the value of the bias I get is much larger than the rest of the components of w, a few orders of magnitude | 18:42 |
alexlovesdata | without regularization would be better | 18:42 |
@sonney2k | regularized bias is in liblinear because they cannot handly true bias | 18:43 |
@sonney2k | handle | 18:43 |
alexlovesdata | but it is easier to implement with | 18:43 |
@sonney2k | n4nd0, hmmhh I am not 100% sure with this but the rest of liblinear doesn't have a true bias so I would be surprised if the mc part had | 18:44 |
-!- alexlovesdata [82955843@gateway/web/freenode/ip.130.149.88.67] has quit [Quit: Page closed] | 18:44 | |
n4nd0 | ok | 18:44 |
n4nd0 | sonney2k: is it the epsilon value relevant? I tend to put it as small as possible (as long as the warnings don't show up) to improve classification results | 18:45 |
@sonney2k | n4nd0, I just checked the source | 18:45 |
@sonney2k | they use regularized bias in MC | 18:45 |
n4nd0 | ok | 18:45 |
@sonney2k | n4nd0, yes - the smaller the better | 18:46 |
n4nd0 | can you tell me the line where it appears? | 18:46 |
@sonney2k | if you compare to mosek - mosek will do 1e-12 or so optimization | 18:46 |
n4nd0 | I tried to check it on my own but lost trace of it | 18:46 |
@sonney2k | n4nd0, 108 in MulticlassLibLinear.cpp | 18:46 |
n4nd0 | sonney2k: ok, did you want to take a look to the modular issue related to my code? | 18:49 |
n4nd0 | I am surprised anyway sonney2k, look http://snipt.org/vahff0 | 18:50 |
n4nd0 | the first two columns are the weight vectors | 18:50 |
n4nd0 | the last column each of the bias | 18:50 |
n4nd0 | epsilon = 10e-5 in this case | 18:51 |
@sonney2k | n4nd0, what you could do is create some multiclass problem where you put 4 gaussians at the x and y axis at (-1,0), (+1,0), (0,-1), (0,+1) | 18:51 |
@sonney2k | n4nd0, wait doesn't help | 18:52 |
@sonney2k | n4nd0, how many classes do you use? | 18:52 |
n4nd0 | yeah, I have thought of that | 18:52 |
@sonney2k | I mean just use 2 classes | 18:52 |
@sonney2k | and then doing that would work | 18:52 |
n4nd0 | for more than three classes I cannot come up with any disposition in 2D that makes it linearly separable for hyperplanes without bias | 18:52 |
@sonney2k | n4nd0, btw why don't you compare to GMNPSVM? | 18:53 |
n4nd0 | I have no idea what is that, let me see | 18:53 |
@sonney2k | n4nd0, a true multiclass svm. I am not sure about the exact formulation though | 18:55 |
n4nd0 | sonney2k: why do you think it could be better with GMNPSVM? | 18:55 |
n4nd0 | I am taking a fast look and I think it is in the dual | 18:55 |
n4nd0 | I chose LibLinear because I can train in the primal as well, like it is done for the SO-SVM | 18:56 |
@sonney2k | n4nd0, I dont' know what the current state is | 18:58 |
@sonney2k | n4nd0, btw disabling bias is probably best then w/ liblinear and in your code | 18:58 |
@sonney2k | reducing epsilon from 1e-2 ... 1e-6 you should get more and more close results | 18:59 |
n4nd0 | n4nd0: without bias I obtain similar % of correct classification with liblinear and my code, what is good | 18:59 |
n4nd0 | but the thing is that for the example I am trying, 10 Gaussians with 100 samples each quite separated | 18:59 |
n4nd0 | this % is quite low, about 40%-50% | 18:59 |
n4nd0 | I believe there could be an error in my code and still get this rate | 19:00 |
n4nd0 | that's why I wan to try it with bias, LibLinear goes up to 90% with bias for the same example | 19:00 |
n4nd0 | if I get something similar with so-svm then I'll be more confident that the code may be ok | 19:00 |
@sonney2k | n4nd0, don't compare accuracy | 19:01 |
@sonney2k | n4nd0, compute the output of each f(x) | 19:01 |
n4nd0 | what should I compare then? | 19:01 |
n4nd0 | scores? | 19:01 |
@sonney2k | and compare the real valued numbers | 19:01 |
n4nd0 | weight vectors? | 19:01 |
@sonney2k | n4nd0, normally when you compare convex optimizers you compute the objective value | 19:02 |
@sonney2k | and if it deviates by not too much all is good | 19:02 |
n4nd0 | yeah I understand that | 19:03 |
n4nd0 | but Nico told me that since they are different solvers for the optimization problem | 19:03 |
n4nd0 | they could be not so equal | 19:03 |
@sonney2k | n4nd0, so what? | 19:04 |
@sonney2k | if they are close then it is ok | 19:04 |
n4nd0 | I am testing right now and giving you the results | 19:04 |
@sonney2k | and results shouldb e more close when epsilon -> 0 | 19:04 |
@sonney2k | n4nd0, how can you compute objectives for liblinear? | 19:04 |
@sonney2k | I guess in your code you have it as output of mosek | 19:04 |
n4nd0 | for liblinear I just do get_w() for each machine | 19:06 |
n4nd0 | aaah! I forgot to say, I am using a MulticlassMachine formed by binary LibLinear | 19:06 |
n4nd0 | not MulticlassLibLinear | 19:06 |
n4nd0 | I should have said that before, sorry | 19:06 |
@sonney2k | n4nd0, ?? | 19:07 |
@sonney2k | how is this comparable then? | 19:07 |
@sonney2k | n4nd0, don't you do true multiclass? | 19:07 |
n4nd0 | http://snipt.org/vahfh6 | 19:07 |
n4nd0 | that's how it is exactly done | 19:07 |
@sonney2k | n4nd0, well not you should compare it to true multiclass liblinear! | 19:08 |
n4nd0 | aham, why? shouldn't they be more or less the same? | 19:08 |
@sonney2k | n4nd0, it is like comparing apples and bananas... | 19:09 |
@sonney2k | n4nd0, really the best thing you can do is compare it to nicos' true MC code | 19:10 |
n4nd0 | :O, really, why? I thought it had actually sense | 19:10 |
@sonney2k | how? | 19:10 |
@sonney2k | I mean you need to compare if objectives are the same | 19:10 |
@sonney2k | classification accuracy is only a vague indication that things are ok | 19:11 |
n4nd0 | aham, exactly the same optimization problem | 19:11 |
@sonney2k | as alex said compare it to the true CS thing http://svmlight.joachims.org/svm_multiclass.html | 19:12 |
n4nd0 | I am not sure why they shouldn't be equivalent though | 19:12 |
@sonney2k | so download TJ's svm multiclass | 19:12 |
@sonney2k | make sure to disable bias in your code | 19:12 |
@sonney2k | and to scale C by number of examples in your code (seems like TJ is doing it...) | 19:13 |
n4nd0 | my bias doesn't change anything so ... | 19:13 |
n4nd0 | ok | 19:13 |
@sonney2k | n4nd0, it changes objective | 19:13 |
@sonney2k | and that is all what counts | 19:13 |
@sonney2k | then compare objectives | 19:14 |
@sonney2k | difference should be *very small* | 19:14 |
@sonney2k | in the order of epsilon | 19:14 |
n4nd0 | there's no epsilon in my code | 19:15 |
@sonney2k | n4nd0, in TJs there is | 19:16 |
n4nd0 | yes | 19:16 |
@sonney2k | and yours is using mosek's default so should be <1e-12 | 19:16 |
n4nd0 | I am not sure this epsilon is mosek's default | 19:17 |
n4nd0 | I have seen that in the paper the bias they use is for the condition to add a new constraint | 19:17 |
n4nd0 | sorry | 19:17 |
n4nd0 | don't read bias ^, read epsilon | 19:17 |
n4nd0 | http://www.cs.cornell.edu/People/tj/publications/tsochantaridis_etal_04a.pdf | 19:18 |
n4nd0 | page 4, algorithm 1, line 8 | 19:18 |
@sonney2k | n4nd0, it basically means they tolerate some constraint violation | 19:21 |
@sonney2k | up to epsilon | 19:21 |
@sonney2k | so if you set epsilon smaller it will take longer to converge (or will never :D | 19:21 |
n4nd0 | yes | 19:22 |
n4nd0 | what I meant is that it is not the mosek's default, but another thing | 19:22 |
@sonney2k | n4nd0, yes of course | 19:23 |
n4nd0 | in my code it will be simply H(y_pred) > xi_i, in line 8 of the algo | 19:23 |
n4nd0 | that's how it is in the code | 19:23 |
@sonney2k | just run svm multiclass once with say 0.1 as epsilon | 19:23 |
@sonney2k | then compare objectives | 19:23 |
@sonney2k | then run it with 1e-4 or so | 19:23 |
@sonney2k | and compare again | 19:23 |
@sonney2k | results shoudl become more similar | 19:23 |
@sonney2k | when decreasing epsilon in svm multiclass | 19:24 |
n4nd0 | my liblinear multiclass right? | 19:24 |
n4nd0 | or should I repeat for MulticlassLiblinear? | 19:24 |
@sonney2k | dont' care about liblinear | 19:25 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 19:25 | |
@sonney2k | n4nd0, compare you MCSO svm w/ svm multiclass | 19:26 |
@sonney2k | puffin444, hey there .... about your patch - shall we merge it or what is the state | 19:26 |
@sonney2k | puffin444, other known issues except for this live user? | 19:26 |
puffin444 | I was hoping that it would be merged, as the additional model selection code would make it a pretty huge patch | 19:27 |
@sonney2k | puffin444, well you never said so | 19:27 |
@sonney2k | so we didn't know | 19:27 |
puffin444 | I have tested it as much as possible for accuracy against GPML and for memory leaks via Valgrind | 19:27 |
@sonney2k | ok | 19:27 |
@sonney2k | so only the live user issue is in there right? | 19:27 |
puffin444 | Yes. | 19:28 |
@sonney2k | puffin444, so you now use enums to compare classes? | 19:29 |
@sonney2k | no longer get_name? | 19:29 |
puffin444 | Yes. | 19:29 |
@sonney2k | ok | 19:29 |
puffin444 | I'm using enums to determine the likelihood models and will do so for other classes for now on. | 19:29 |
CIA-18 | shogun: Soeren Sonnenburg master * r5609bab / (18 files in 3 dirs): Merge pull request #585 from puffin444/master (+11 more commits...) - http://git.io/Aj_NBg | 19:29 |
@sonney2k | puffin444, btw can we have some example for that? | 19:31 |
puffin444 | Absolutely | 19:31 |
@sonney2k | puffin444, ahh and one more comment - what do you need iostream for? | 19:31 |
puffin444 | Nothing. I had it in there for debugging and looks like I forgot to take it out. sorry :( | 19:31 |
@sonney2k | puffin444, did you already include it in the modular interfaces or are you using it from C? | 19:31 |
@sonney2k | puffin444, you can always use SG_PRINT | 19:32 |
puffin444 | Will do so in the future. | 19:32 |
@sonney2k | and nowadays matrix.display() / vector.display() etc | 19:32 |
@sonney2k | puffin444, thanks for the patch | 19:32 |
puffin444 | Right now it's all in C/C++, I can have a python example if you wish. | 19:33 |
@sonney2k | puffin444, well some C/C++ one for the moment is OK - later we need some eye candy in python / matplotlib :) | 19:34 |
puffin444 | Absolutely. Would you like me to submit a small patch including an example right away or would if be fine if I just included it when I submit the model selection code? | 19:34 |
n4nd0 | sonney2k: is there something to talk about the modular interface? | 19:36 |
@sonney2k | n4nd0, yes - you said you wanted to return some specialize code | 19:37 |
@sonney2k | puffin444, the earlier we have an example the better - these things get checked on the buildbot so they are kind of a minimal test | 19:38 |
puffin444 | Okay. | 19:38 |
@sonney2k | n4nd0, err derived class | 19:38 |
@sonney2k | so I need to know the exact names to get things to work | 19:38 |
n4nd0 | sonney2k: first, since I am using specialized StructuredLabels, e.g. MulticlassSOLabels; I wonder whether apply should return StructuredLabels and later cast or return MulticlassSOLabels directly | 19:39 |
n4nd0 | and then when I do CStructuredLabels::get_label() | 19:39 |
n4nd0 | that returns a CStructuredData* | 19:39 |
n4nd0 | but maybe it would be better to return the derived class, e.g. CRealNumber for the multiclass example | 19:40 |
@sonney2k | n4nd0, yeah we can do that | 19:47 |
@sonney2k | but I need some list for that and an example (in python to see if things work now) | 19:48 |
n4nd0 | sonney2k: things won't work in python now since the swig interface is not fully completed | 19:52 |
n4nd0 | sonney2k: but I can try to do it, make the example work and let you know | 19:52 |
n4nd0 | is that ok? | 19:53 |
@sonney2k | n4nd0, sure | 19:54 |
n4nd0 | but just as an idea | 19:55 |
n4nd0 | what is the preference, to do swig magic to return correct types or to use things like RealNumber.obtain_from_generic(multiclassSOLabels) | 19:56 |
n4nd0 | however for PrimalMosekSOSVM I doubt if swig magic will work since one may want to return MulticlassSOLabels or other types of labels depending on the application at hand | 19:56 |
@sonney2k | n4nd0, I would always provide the obtain_from_generic helpers | 20:04 |
@sonney2k | n4nd0, it is only 'syntactic sugar' to not having to use them from the type unaware interfaces in cases where these are not needed of course | 20:04 |
CIA-18 | shogun: Soeren Sonnenburg master * r140515f / src/interfaces/modular/Machine.i : revert patch deactivating apply magic - http://git.io/woCRXw | 20:08 |
@sonney2k | n4nd0, if you want to see how it is done look at interfaces/modular/Machine.i | 20:11 |
@sonney2k | basically all I do is some %extend classname { CRegressionLabels* apply(CFeatures* data=NULL) { return $self->apply_regression(data); } } | 20:11 |
@sonney2k | and then a %rename of the original apply function | 20:12 |
@sonney2k | that's it | 20:12 |
CIA-18 | shogun: Soeren Sonnenburg master * rf4d348f / (43 files in 7 dirs): move kernel normalizers to shogun/kernel/normalizer - http://git.io/40CGTQ | 20:15 |
n4nd0 | sonney2k: ok, thanks | 20:16 |
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun | 20:18 | |
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host] | 20:18 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 20:18 | |
shogun-buildbot | build #972 of cmdline_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/972 | 20:18 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 20:19 | |
@sonney2k | blackburn, hey there | 20:20 |
blackburn | hey | 20:20 |
@sonney2k | blackburn, any reason you disabled the apply magic in interfaces/modular/Machine.i | 20:20 |
blackburn | hmm did I? | 20:20 |
@sonney2k | yeah | 20:21 |
@sonney2k | all commented | 20:21 |
@sonney2k | I fixed that in 140515f139e1dd9605d96ea28a80d1bfd8e92a2d | 20:21 |
blackburn | I do not remember I did that | 20:22 |
@sonney2k | blackburn, git blamed you | 20:22 |
@sonney2k | and git never lies | 20:22 |
@sonney2k | :D | 20:22 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 20:26 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 20:30 | |
blackburn | sonney2k: I do not really understand why it became commented O_o | 20:30 |
blackburn | it was rather unintentional | 20:31 |
shogun-buildbot | build #973 of cmdline_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/973 blamelist: walke434@msu.edu, sonne@debian.org | 20:36 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Quit: Page closed] | 20:42 | |
@sonney2k | puffin444, can you please have a look at examples/undocumented/libshogun/regression_gaussian_process.cpp ? | 20:55 |
@sonney2k | it fails currently... | 20:55 |
@sonney2k | fails to compile that is | 20:55 |
puffin444 | Oh No. It looks like I forgot to update the regression_gaussian_process.cpp file for the new interface :-/ | 20:56 |
puffin444 | I am uploading a new example ASAP. | 20:58 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 21:03 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 21:17 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 21:18 | |
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun | 21:39 | |
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host] | 21:39 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 21:39 | |
@sonney2k | blackburn, chris' patch breaks stuff - examples in libshogun won't work | 21:51 |
CIA-18 | shogun: puffin444 master * r0051549 / (4 files in 3 dirs): Updated Gaussian Process Regression example. - http://git.io/5RGkUw | 21:56 |
CIA-18 | shogun: Soeren Sonnenburg master * rcf381d6 / (4 files in 3 dirs): Merge pull request #591 from puffin444/master - http://git.io/fv0d5w | 21:56 |
@sonney2k | puffin444, btw you can use init_shogun_with_defaults() | 21:56 |
@sonney2k | then you don't need to set a print etc function | 21:56 |
puffin444 | ok. | 21:56 |
puffin444 | I'll quickly change the example if you want with that. | 21:57 |
puffin444 | Sorry about breaking the build. | 21:57 |
@sonney2k | puffin444, blackburn/chris broke it before you so relax :) | 22:00 |
@sonney2k | and they are both hiding | 22:00 |
@sonney2k | wiking, any news on your PR? | 22:00 |
wiking | jes | 22:06 |
@sonney2k | wiking, a real one tomorrow? | 22:12 |
wiking | i have a merging error here | 22:13 |
wiking | trying to resolve | 22:13 |
@sonney2k | ? | 22:13 |
wiking | something happened with the Makefile in the example | 22:13 |
wiking | that i cannot merge in | 22:13 |
wiking | debugging | 22:14 |
puffin444 | sonney2k, are we still using NLOPT in shogun? | 22:17 |
@sonney2k | puffin444, if you need it - you can. pluskid found another lib that seems to be better suited for his optimization problem | 22:18 |
@sonney2k | (see his pull request - maybe it fits to what you want to do) | 22:18 |
puffin444 | sonney2k, I saw pluskid's patch and I wondered if it would be easier just to use his lib | 22:19 |
@sonney2k | puffin444, look at it! | 22:19 |
@sonney2k | it seemed very promising to me | 22:19 |
puffin444 | It does to me too. My only concern is that it doesn't support constraints | 22:20 |
puffin444 | But I think the only constraints I have are that the hyperparameters must be positive. | 22:20 |
@sonney2k | puffin444, try it out on some toy example - I would bet it supports box constraints... | 22:22 |
puffin444 | Would optimizing over log space work? | 22:22 |
@sonney2k | puffin444, no idea - have a look at this lib (I don't know it either) or ask oliver | 22:23 |
puffin444 | I think it would - Oliver mentioned something like this. | 22:24 |
@sonney2k | IIRC will a log transform keep things convex | 22:28 |
@sonney2k | actually log(x) is a counter example - haha | 22:30 |
n4nd0 | puffin444: about optimizing after applying log, it depends on what are you applying log to I think | 22:35 |
puffin444 | ok | 22:35 |
n4nd0 | depending on its convexity/concavity and if it's decreasing/increasing it can turn the problem into convex and so | 22:36 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 264 seconds] | 22:44 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 22:46 | |
-!- emrecelikten [~emre@176.41.10.108] has joined #shogun | 22:53 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed] | 23:22 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has joined #shogun | 23:47 | |
-!- wiking [~wiking@ip68-14-87-205.ri.ri.cox.net] has quit [Changing host] | 23:47 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 23:47 | |
--- Log closed Thu Jun 21 00:00:41 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!