--- Log opened Wed May 22 00:00:05 2013 | ||
--- Day changed Wed May 22 2013 | ||
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has left #shogun ["PING 1369173601"] | 00:00 | |
foulwall | Hi dvalcarce? | 01:25 |
---|---|---|
dvalcarce | foulwall, hi! | 01:37 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 01:41 | |
-!- iglesiasg [d58f329d@gateway/web/freenode/ip.213.143.50.157] has quit [Quit: Page closed] | 02:07 | |
-!- dvalcarce [5b757556@gateway/web/freenode/ip.91.117.117.86] has quit [Quit: Page closed] | 02:50 | |
-!- nube1 [~rho@49.244.29.35] has quit [Quit: Leaving.] | 04:51 | |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has left #shogun ["WeeChat 0.3.7"] | 05:15 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 06:08 | |
-!- foulwall [~foulwall@2001:da8:215:503:4172:782d:f853:e6d6] has quit [Remote host closed the connection] | 06:20 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Leaving] | 06:21 | |
-!- gsomix [~gsomix@37.61.180.77] has quit [Quit: Leaving] | 07:03 | |
-!- zxtx [~zv@ool-1892e23a.dyn.optonline.net] has joined #shogun | 07:14 | |
-!- nube [~rho@116.90.239.13] has joined #shogun | 07:30 | |
-!- nube [~rho@116.90.239.13] has quit [Client Quit] | 07:32 | |
-!- nube [~rho@116.90.239.13] has joined #shogun | 07:33 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:a4da:7547:abec:5503] has joined #shogun | 07:37 | |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has joined #shogun | 08:30 | |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has quit [Quit: WeeChat 0.3.7] | 08:39 | |
-!- lisitsyn [~blackburn@37.61.180.77] has quit [Quit: Leaving.] | 08:49 | |
@wiking | sonney2k: ok i'll check today before late night the thingy | 09:02 |
-!- iglesiasg [c1934402@gateway/web/freenode/ip.193.147.68.2] has joined #shogun | 09:15 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 09:15 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:16 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 09:16 | |
@lisitsyn | sonney2k: sonne|work: yes I will push it soon | 09:22 |
@lisitsyn | (about RNG in tapkee) | 09:22 |
sonne|work | lisitsyn: I assume you will then youse the CMath::random* functions? | 09:23 |
@lisitsyn | sonne|work: yes with some define magic | 09:23 |
sonne|work | ok | 09:23 |
sonne|work | lisitsyn: just make sure that shogun's random(min,max) functions are closed interval | 09:24 |
sonne|work | I mean that this fits what you need | 09:24 |
@wiking | lisitsyn: if not then you can use some other random_* functions in CRandom | 09:26 |
@wiking | ;P | 09:26 |
sonne|work | wiking: well in fact once my patch is merged - we should make introduce static random() functions in Random and drop the ones from CMath | 09:28 |
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.] | 09:56 | |
-!- nube [~rho@116.90.239.13] has joined #shogun | 09:57 | |
sonne|work | wiking: btw any feedback? | 09:57 |
@wiking | sonne|work: hard drives are dying around me | 10:28 |
@wiking | actually now it seems even mine | 10:28 |
@wiking | sorry but have to do seomthing about this first | 10:28 |
sonne|work | good luck | 10:28 |
-!- hushell [~hushell@c-67-189-100-116.hsd1.or.comcast.net] has joined #shogun | 10:46 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 11:03 | |
shogun-notifier- | shogun: lambday :develop * b005795 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/b0057954d3f6dda266b21c5e9c4e3062d9097849 | 11:03 |
shogun-notifier- | shogun: Jacobi Elliptic Functions added | 11:03 |
shogun-notifier- | shogun: Heiko Strathmann :develop * 8bb3fd5 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/8bb3fd559da99d5adc90dc8c2f97f26c4fb832bf | 11:03 |
shogun-notifier- | shogun: Merge pull request #1111 from lambday/develop | 11:03 |
shogun-notifier- | shogun: | 11:03 |
shogun-notifier- | shogun: JacobiEllipticFunctions added | 11:03 |
shogun-buildbot | build #859 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/859 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 11:08 |
shogun-buildbot | build #860 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/860 | 11:30 |
-!- iglesiasg [c1934402@gateway/web/freenode/ip.193.147.68.2] has quit [Quit: Page closed] | 12:16 | |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-kxrcuwzhyylfmgom] has joined #shogun | 12:22 | |
-!- gsomix [~gsomix@185.6.239.113] has joined #shogun | 12:22 | |
gsomix | hello! | 12:22 |
@HeikoS | hi! | 12:22 |
lambday | HeikoS: hi | 12:22 |
@HeikoS | hi lambday | 12:23 |
lambday | HeikoS: arprec seemed to have worked... | 12:24 |
lambday | HeikoS: we need to more methods that computes shifts and weights.. do you think Statistics is a good place for that? | 12:24 |
@HeikoS | lambday: good question | 12:26 |
@HeikoS | maybe a helper class for CG | 12:26 |
@HeikoS | ah no | 12:26 |
@HeikoS | its related to the rational approximations | 12:27 |
@HeikoS | so a helper class for this | 12:27 |
@HeikoS | contour integral helper or so | 12:27 |
lambday | HeikoS: hmm.. okay | 12:28 |
lambday | so a separate class with two static methods, right? | 12:29 |
lambday | CContourIntegral or something | 12:29 |
@HeikoS | yes | 12:29 |
@HeikoS | with a reference etc | 12:29 |
lambday | umm.. reference? | 12:30 |
@HeikoS | just what the contour integral is etc | 12:31 |
@HeikoS | and paper reference | 12:31 |
lambday | ohh | 12:31 |
lambday | yes | 12:31 |
lambday | once this is done I'll chalk out a class diagram of the framework | 12:32 |
@HeikoS | okay | 12:32 |
@HeikoS | sounds good | 12:32 |
lambday | one question - while registering the systems (to be solved later) we need to store the shifts for those systems somewhere in the computation class, right? | 12:34 |
@HeikoS | yes exactly | 12:34 |
lambday | okay | 12:34 |
@HeikoS | maybe we can do this via indices or so | 12:34 |
@HeikoS | since redundant | 12:34 |
@HeikoS | so one passes a vector of shifts | 12:35 |
@HeikoS | and every problem has a list of indices for them | 12:35 |
@HeikoS | this minimises the communication overhead | 12:35 |
lambday | HeikoS: awesome idea | 12:35 |
lambday | okay... | 12:35 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun | 12:45 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 12:45 | |
sonne|work | hi gsomix! how is it going? | 13:00 |
sonne|work | and hey foulwall_! are you feeling better? | 13:00 |
gsomix | sonne|work, fine, have returned from the university. now testing and preparing PR. | 13:01 |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-kxrcuwzhyylfmgom] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 13:02 | |
@lisitsyn | sonne|work: http://jmlr.org/papers/volume14/curtin13a/curtin13a.pdf shameful tests for us | 13:10 |
sonne|work | lisitsyn: well KNN / k-means - not really shameful | 13:12 |
@wiking | lisitsyn: :) | 13:13 |
sonne|work | and you don't know which version of shogun this is | 13:13 |
@wiking | lisitsyn: we only need to work on knn and kmeans :P | 13:13 |
@lisitsyn | naywhayare: which version of shogun that is? ;) | 13:13 |
@lisitsyn | sonne|work: and they named us shogun toolkit | 13:14 |
@lisitsyn | :D | 13:14 |
sonne|work | lisitsyn: btw I wonder why shogun is like 4 times faster on the isolet dataset | 13:16 |
sonne|work | but consistently slower on everything else | 13:16 |
@wiking | sonne|work: btw: about bagging | 13:16 |
@wiking | u got a sec to talk about that? | 13:16 |
sonne|work | wiking: well talk | 13:16 |
@wiking | sonne|work: bagging class would simply just work only on one type of classifier at a time right? | 13:18 |
sonne|work | yes | 13:18 |
@wiking | sonne|work: i.e. Bagging(machine, features, labels) | 13:18 |
sonne|work | BaggingMachine yes | 13:18 |
@wiking | or not necessarily in constructor | 13:18 |
@wiking | or Bagging(features,labels) | 13:19 |
sonne|work | and some number of baggings to do | 13:19 |
@wiking | and then Bagging.run(machine, number of bags) | 13:19 |
sonne|work | well it needs a machine, features, labels, num_baggings | 13:19 |
sonne|work | wiking: wait no | 13:19 |
sonne|work | keep it to work with the machine interface | 13:19 |
sonne|work | so train(data) | 13:19 |
@wiking | aaah | 13:19 |
sonne|work | parameters have to be set before | 13:19 |
@wiking | but still then | 13:19 |
@wiking | Bagging(features,labels) | 13:20 |
@wiking | bagging.set_machine(machine) | 13:20 |
@wiking | bagging.train(data) | 13:20 |
@wiking | something like this | 13:20 |
@wiking | so that one could reuse Bagging object for the same feature,label pair | 13:20 |
@wiking | with different machine | 13:20 |
sonne|work | lisitsyn: yeah and it is all rather small data sets still | 13:21 |
@wiking | ah yeah there should be as well bagging.set_num_bags(bags) | 13:21 |
@wiking | ok | 13:21 |
@wiking | this should be pretty straightforward as far as i can see now | 13:21 |
sonne|work | wiking: yeah all you need is a good prng :D | 13:22 |
@wiking | sonne|work: lol | 13:22 |
@wiking | sonne|work: btw for the final Bagging.classify(data) we would need to be able to set different voting schemes | 13:22 |
@wiking | ok one remaining question | 13:23 |
@wiking | where should Bagging reside? | 13:23 |
@wiking | sonne|work: machine/Bagging.? | 13:23 |
@wiking | or modelselection? :) | 13:24 |
@wiking | it's pretty far fetched but still | 13:24 |
sonne|work | wiking: machine/BaggingMachine.h for the master class | 13:26 |
@wiking | k | 13:26 |
@wiking | and then we would still implement RandomForest : BaggingMachine? | 13:27 |
sonne|work | wiking: rather implement a tree | 13:28 |
@wiking | yep i thought so | 13:28 |
sonne|work | that is then directly used in baggingmachine to create the forest | 13:28 |
sonne|work | though it might be helpful for some to create a RandomForest class that just utilizes the baggingmachine with a tree | 13:29 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 13:32 | |
sonne|work | lisitsyn: I think we should also create some kind of benchmark table - just for us to see if our algorithms became slower/faster | 13:34 |
@lisitsyn | sonne|work: makes sense | 13:36 |
gsomix | sonne|work, done | 13:40 |
@wiking | gsomix: benchmark table? woah that was fast! | 13:41 |
@lisitsyn | hahhahahah | 13:41 |
@lisitsyn | rofl | 13:41 |
gsomix | wiking, lol. >:3 | 13:41 |
sonne|work | gsomix: where? | 13:42 |
gsomix | sonne|work, about PR, I hope? https://github.com/shogun-toolbox/shogun/pull/1119 | 13:43 |
@lisitsyn | sonne|work: ha! why is it BSD? | 13:44 |
sonne|work | gsomix: am I missing something? | 13:44 |
sonne|work | lisitsyn: ? | 13:44 |
@lisitsyn | sonne|work: license of the file | 13:45 |
@lisitsyn | asciifile | 13:45 |
sonne|work | lisitsyn: ahh | 13:45 |
@lisitsyn | ahhh got it | 13:45 |
gsomix | sonney2k, mmm? O_O | 13:45 |
@lisitsyn | I saw just a few lines | 13:45 |
@lisitsyn | in the diff | 13:45 |
@lisitsyn | so GPL + BSD | 13:45 |
sonne|work | gsomix: where is the function? | 13:45 |
gsomix | sonne|work, line 1024. | 13:46 |
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.] | 13:46 | |
-!- nube [~rho@116.90.239.13] has joined #shogun | 13:47 | |
sonne|work | gsomix: you need to benchmark / test this thoroughly - the function is not easily readable :/ | 13:49 |
sonne|work | gsomix: take e.g. a file of a few GB in size (say 10GB) and one with a few bytes only | 13:50 |
sonne|work | and then test how long it takes to read it / count number of lines or sth as test | 13:50 |
gsomix | sonne|work, ok! | 13:50 |
-!- van51 [~van51@195.251.235.71] has joined #shogun | 13:53 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 14:04 | |
@wiking | sonne|work: actually nothing is against having the benchmark tests be part of the unit tests ;) | 14:11 |
@wiking | sonne|work: since the running time e.g. is measured by google unit test FW ;P | 14:11 |
sonne|work | wiking: haha we don't want to wait an hour :D | 14:12 |
@wiking | sonne|work: well then just create a new dir but still keep using gtest fw | 14:13 |
@wiking | :P | 14:13 |
-!- hushell [~hushell@c-67-189-100-116.hsd1.or.comcast.net] has quit [Ping timeout: 252 seconds] | 14:16 | |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Ping timeout: 250 seconds] | 14:16 | |
-!- van51 [~van51@195.251.235.71] has quit [Quit: Leaving.] | 14:19 | |
vgorbati | lisitsyn: here? | 14:26 |
@lisitsyn | vgorbati: yes | 14:27 |
vgorbati | lisitsyn: back to my question: is it ok to put the function 'get_feature_vectors_and_compute' into 'Kernel.h' file? | 14:28 |
@lisitsyn | vgorbati: yeah probably | 14:28 |
sonne|work | vgorbati, lisitsyn what is this for? | 14:30 |
sonne|work | wiking: haha | 14:30 |
vgorbati | sonne|work: a lot of kernels share the same pattern: they get feature vectors from lhs and rhs, and then do some computations | 14:30 |
vgorbati | sonne|work: I suggested to extract the part of getting the feature vectors into separate function and re-use it | 14:31 |
vgorbati | sonne|work: I have already tried this approach for one kernel, what if I send a PR, so that you can decide whether it makes sense, and if it does - I will do it for other kernels? | 14:32 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 14:45 | |
foulwall_ | hi sonne|work | 14:46 |
sonne|work | hi foulwall | 14:46 |
sonne|work | how are you? | 14:46 |
foulwall_ | sonne|work: all right. I'm doing the data importer | 14:47 |
sonne|work | foulwall_: so headache is gone very good | 14:47 |
sonne|work | foulwall_: data importer? | 14:47 |
sonne|work | you mean reading of .h5 files? | 14:48 |
foulwall_ | sonne|work: What I think is make a new view, when the client requires a data by filename, the view respond with a json | 14:49 |
-!- foulwall_ [~foulwall@2001:da8:215:503:a4da:7547:abec:5503] has quit [Remote host closed the connection] | 14:51 | |
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.] | 15:01 | |
-!- foulwall [~foulwall@2001:da8:215:6120:d86f:dfa7:6c8c:cc4] has joined #shogun | 15:08 | |
-!- zxtx [~zv@ool-1892e23a.dyn.optonline.net] has quit [Ping timeout: 256 seconds] | 15:12 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 15:29 | |
vgorbati | lisitsyn, sonne|work: sent a PR | 15:44 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun | 16:07 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 16:08 | |
@lisitsyn | sonne|work: first phase of dedup is here | 16:13 |
sonne|work | lisitsyn: cool | 16:15 |
sonne|work | lisitsyn: no conflicts? | 16:16 |
@lisitsyn | sonne|work: no | 16:16 |
@lisitsyn | sonne|work: https://github.com/shogun-toolbox/shogun/pull/1120/files#r4338851 please check when you get some time | 16:16 |
naywhayare | lisitsyn: sorry about the "shogun toolbox". I realized that after it was too late. I think I also botched "scikit.learn" which should be "scikit-learn" (how did I do that?) | 16:36 |
@lisitsyn | naywhayare: :) | 16:37 |
@lisitsyn | naywhayare: I am finishing my paper for jmlr mloss as well | 16:38 |
@lisitsyn | I've been trying to finish it last two months haha :D | 16:38 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Ping timeout: 250 seconds] | 16:38 | |
naywhayare | I was using shogun out of git probably about eight months ago | 16:38 |
naywhayare | I think with kNN, shogun was using the brute-force N^2 search | 16:39 |
@lisitsyn | I see | 16:39 |
naywhayare | which explains why it seems to do better with ISOLET -- ISOLET has rather high dimensionality, and trees don't do as well in those settings | 16:39 |
naywhayare | the commands I was using was the best I could find in the documentation without digging too deep into the code | 16:39 |
naywhayare | using the text interface (I think that's what it's called) it looked like this: | 16:40 |
naywhayare | new_classifier KNN | 16:40 |
naywhayare | train_classifier 4 | 16:40 |
naywhayare | set_features TEST /some/file | 16:40 |
naywhayare | out.txt = classify | 16:40 |
naywhayare | I think in the paper we used k = 3 (I can't remember right now), but that doesn't make a huge difference | 16:41 |
sonne|work | naywhayare: knn / kmeans are not exactly fast in shogun anyway... | 16:42 |
naywhayare | yeah; shogun focuses on SVM (to my understanding) so it's not like shogun is claiming to be the best in those things | 16:43 |
-!- foulwall [~foulwall@2001:da8:215:6120:d86f:dfa7:6c8c:cc4] has quit [Remote host closed the connection] | 16:43 | |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun | 16:44 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 16:44 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 17:03 | |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has joined #shogun | 17:03 | |
sonne|work | naywhayare: well we have lots of other stuff too but we only claim to be best in SVM related business :D | 17:04 |
@wiking | sonne|work: we should diversify ;P | 17:04 |
sonne|work | wiking: we do! | 17:05 |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun | 17:10 | |
-!- foulwall [~foulwall@2001:da8:215:c252:bcbe:6ebf:7c7f:6188] has joined #shogun | 17:14 | |
-!- romi__ [~mizobe@187.101.139.4] has joined #shogun | 17:38 | |
foulwall | sonney2k: sonney, there's only a australian.libsvm.h5 in shogun-data, respect to the other data files in the repo, I must do standardize job. | 17:46 |
foulwall | sonne|work: and when I standardize, I'll convert all of them to .h5? | 17:47 |
@sonney2k | foulwall, no just use this single one | 17:48 |
@sonney2k | foulwall, we can add more data from mldata.org later | 17:48 |
foulwall | sonne|work: sorry I went wrong again... | 17:49 |
@sonney2k | foulwall, better always ask - I totally don't mind to the contrary | 17:50 |
@sonney2k | it is frustrating for you / me otherwise | 17:50 |
foulwall | sonne|work: ok~ | 17:51 |
@sonney2k | gtg | 17:51 |
foulwall | sonney2k: bye | 17:52 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 17:53 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 17:59 | |
-!- sumit [75e1c0a1@gateway/web/freenode/ip.117.225.192.161] has joined #shogun | 18:14 | |
-!- sumit [75e1c0a1@gateway/web/freenode/ip.117.225.192.161] has quit [Quit: Page closed] | 18:49 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 18:58 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 19:09 | |
shogun-notifier- | shogun-demo: foulwall :data * 073026a / / (8 files): https://github.com/shogun-toolbox/shogun-demo/commit/073026a3893fd4028d8a651ee59d8397f118801d | 19:09 |
shogun-notifier- | shogun-demo: 1. import the shogun-data as a submodule. 2. add serveral lines for | 19:09 |
shogun-notifier- | shogun-demo: gp-demo, for fetch the data from australian.libsvm.h5. 3. add a feature | 19:09 |
shogun-notifier- | shogun-demo: on all available demos, when user press the button, when the browser do | 19:09 |
shogun-notifier- | shogun-demo: the ajax activity, the button are disabled to get rid of the duplicate | 19:09 |
shogun-notifier- | shogun-demo: calculation. | 19:09 |
@sonney2k | Re | 19:09 |
foulwall | sonney2k: ah, notifier! | 19:09 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Quit: Page closed] | 19:11 | |
@sonney2k | foulwall, something is wrong with data though | 19:23 |
@sonney2k | foulwall, it points to your data repository | 19:23 |
foulwall | sonney2k: oh I fix | 19:24 |
foulwall | sonney2k: The 'australian.libsvm.h5' file has 15 features, Is that means what I can do is extract any pair of features and treat them as 2 dims data, then send them into svr/gp demo? | 19:25 |
@sonney2k | foulwall, yes - btw please still work in your foulwall fork and interact with the official shogun-demo via PR's | 19:25 |
foulwall | sonney2k: ok | 19:26 |
@sonney2k | wiking, help help! | 19:28 |
foulwall | sonney2k: sent the fix pr. I'll try to import the data to the demos as much as I can tomorrow. goodnight~ | 19:36 |
-!- romi__ [~mizobe@187.101.139.4] has quit [Quit: Leaving] | 19:37 | |
@sonney2k | foulwall, good night! | 19:38 |
@sonney2k | HeikoS, I am getting this | 19:45 |
@sonney2k | ./statistics_mmd_kernel_selection_combined.py:73: RuntimeWarning: [WARN] ������MMDKernelSelectionComb::solve_optimization(): all mmd estimates are negative. This is techically possible, although extremely rare. Consider using different kernels. This combination will lead to a bad two-sample test. Since anycombination is bad, will now just return equally distributed kernel weights | 19:45 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 19:46 | |
@HeikoS | really? | 19:46 |
@HeikoS | sonney2k, consistently? | 19:46 |
@sonney2k | HeikoS, well I changed the rng so maybe this is causign this | 19:49 |
-!- foulwall [~foulwall@2001:da8:215:c252:bcbe:6ebf:7c7f:6188] has quit [Remote host closed the connection] | 19:53 | |
shogun-notifier- | shogun-data: Soeren Sonnenburg :master * 03cbc66 / testsuite/tests/ (12 files): https://github.com/shogun-toolbox/shogun-data/commit/03cbc66469bde59145b03048144a6b003c991274 | 19:58 |
shogun-notifier- | shogun-data: update integration tests to new random functions | 19:58 |
shogun-notifier- | shogun-data: Soeren Sonnenburg :master * 4831243 / testsuite/clustering/KMeans_EuclideanDistance_3.m: https://github.com/shogun-toolbox/shogun-data/commit/483124318ef0cbd64b7d844520322898cc2b5582 | 19:58 |
shogun-notifier- | shogun-data: fix old integration tests | 19:58 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * a2519cf / / (6 files): https://github.com/shogun-toolbox/shogun/commit/a2519cf7694872290fb1ffc050aee9f10e084867 | 19:58 |
shogun-notifier- | shogun: use lower order bits in prng | 19:58 |
shogun-notifier- | shogun: | 19:58 |
shogun-notifier- | shogun: - This is about 2 times faster compared to the previous rng. | 19:58 |
shogun-notifier- | shogun: - Add unit tests to ensure things are correct | 19:58 |
shogun-notifier- | shogun: - Update integration test data | 19:58 |
shogun-notifier- | shogun-demo: foulwall :master * d45d160 / / (8 files): https://github.com/shogun-toolbox/shogun-demo/commit/d45d160283b7e6095435cd23c1e4466150aa2887 | 20:00 |
shogun-notifier- | shogun-demo: 1. import the shogun-data as a submodule. 2. add serveral lines for | 20:00 |
shogun-notifier- | shogun-demo: gp-demo, for fetch the data from australian.libsvm.h5. 3. add a feature | 20:00 |
shogun-notifier- | shogun-demo: on all available demos, when user press the button, when the browser do | 20:01 |
shogun-notifier- | shogun-demo: the ajax activity, the button are disabled to get rid of the duplicate | 20:01 |
shogun-notifier- | shogun-demo: calculation. | 20:01 |
shogun-notifier- | shogun-demo: Soeren Sonnenburg :master * 083c75b / / (8 files): https://github.com/shogun-toolbox/shogun-demo/commit/083c75bd3caca92adc98a99f5609844548b37137 | 20:01 |
shogun-notifier- | shogun-demo: Merge pull request #3 from foulwall/data | 20:01 |
shogun-notifier- | shogun-demo: | 20:01 |
shogun-notifier- | shogun-demo: correct the shogun-data submodule repo url | 20:01 |
-!- HeikoS [~heiko@nat-163-207.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 20:01 | |
shogun-buildbot | build #861 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/861 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:11 |
-!- iglesiasg [d58f32a1@gateway/web/freenode/ip.213.143.50.161] has joined #shogun | 20:22 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 20:22 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 20:25 | |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has quit [Read error: Connection reset by peer] | 20:27 | |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun | 20:27 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Client Quit] | 20:28 | |
@sonney2k | van51, do you still have time for some stuff? | 20:29 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 20:29 | |
@sonney2k | I need to reduce this blacklist https://github.com/shogun-toolbox/shogun/blob/develop/tests/integration/python_modular/generator.py | 20:29 |
van51 | sonney2k: yea sure | 20:29 |
@sonney2k | to close to 0 | 20:29 |
@sonney2k | these are all tests that fail for one reason or another | 20:29 |
-!- lisitsyn [~blackburn@185.6.239.113] has joined #shogun | 20:29 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 20:29 | |
@sonney2k | some trivial some not fixable by a human being | 20:30 |
van51 | sonney2k: haha | 20:30 |
@sonney2k | (without super powers) | 20:30 |
van51 | you want me to go over them? | 20:30 |
@sonney2k | I am going over them too | 20:30 |
@sonney2k | basically what one needs to do is remove .py example from the blacklist | 20:31 |
@sonney2k | run python generator.py example.py | 20:31 |
@sonney2k | and then tester.py example.py | 20:31 |
@sonney2k | and if it works - all fine | 20:31 |
@sonney2k | if not make it work | 20:31 |
van51 | ok | 20:32 |
van51 | can do | 20:32 |
@sonney2k | vgorbati, help much appriciated | 20:32 |
vgorbati | sonney2k: in fixing integration tests? | 20:32 |
@sonney2k | vgorbati, yeah ... | 20:34 |
van51 | sonney2k:should we report them back here? | 20:34 |
vgorbati | sonney2k: what do you want more: converting to SGVector<..> in kernels or fixing tests? | 20:34 |
@sonney2k | van51 yeah | 20:37 |
@sonney2k | I will also keep you updated about my findings | 20:37 |
@sonney2k | vgorbati, tests first - maybe you choose the *kernel* ones | 20:39 |
@sonney2k | van51, doing library_time.py | 20:39 |
van51 | sonney2k: had you started them from the top? | 20:39 |
van51 | sonney2k: I took it from the bottom, I'm at distance_mahalanobis | 20:40 |
@sonney2k | van51, no random picks | 20:40 |
-!- hushell [~hushell@c-67-189-100-116.hsd1.or.comcast.net] has joined #shogun | 20:43 | |
van51 | sonney2k: I went over them | 20:54 |
van51 | I have six not working | 20:54 |
van51 | sonney2k: and four say that I should recompile with --enable-swig-directors | 20:55 |
van51 | but they report ok, should I recompile? | 20:55 |
@sonney2k | van51, ohh the swig director ones are tough | 20:55 |
@sonney2k | van51, that needs superhuman powers ... | 20:56 |
van51 | sonney2k: the other six it is then | 20:56 |
@sonney2k | van51, IDK if you know but this is C++ class -> python class -> python class extended | 20:57 |
@sonney2k | and the extensions in python are used by the C++ class! | 20:57 |
@sonney2k | heureka! | 20:57 |
van51 | sonney2k: you kind of lost me :p | 20:58 |
@sonney2k | van51, think of a shogun C++ class | 20:58 |
@sonney2k | say CKernel | 20:58 |
@sonney2k | swig gives you CKernel inside python | 20:58 |
@sonney2k | now think of a virtual function in CKernel | 20:58 |
@sonney2k | say compute(int idxa, int idxb) | 20:59 |
@sonney2k | you overload that class in python | 20:59 |
@sonney2k | and whoever in shogun uses CKernel::compute | 20:59 |
@sonney2k | will call the python function! | 20:59 |
van51 | aah sweet | 20:59 |
van51 | and what was the heureka for? | 21:00 |
@sonney2k | van51, gsomix wrote that last gsoc - swig supports that but it is crazy | 21:00 |
@sonney2k | crazy feature IMHO | 21:00 |
@sonney2k | highly unexpected to even work | 21:00 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 21:01 | |
@sonney2k | van51, do all the transfer_* things work? | 21:01 |
van51 | you think the errors in those examples come from this? | 21:01 |
@sonney2k | van51, ? | 21:01 |
van51 | sonney2k : yea | 21:01 |
@sonney2k | van51, ok then enable what works and send a PR | 21:02 |
van51 | ok | 21:02 |
@lisitsyn | sonney2k: why, people used it | 21:02 |
@sonney2k | ohh I need another round of time zone collection | 21:02 |
@sonney2k | van51, lisitsyn, votjakovr, pickle27, hushell, gsomix, iglesiasg - you time zones please | 21:03 |
@sonney2k | lisitsyn, Heiko is in UTC+1 ? | 21:03 |
@iglesiasg | sonney2k: your time zone ;) | 21:03 |
@sonney2k | iglesiasg, UTC+2 | 21:03 |
@lisitsyn | sonney2k: heiko is in london I guess? | 21:03 |
@sonney2k | iglesiasg, so 21hrs now? | 21:04 |
@lisitsyn | sonney2k: mine and gsomix is +4 | 21:04 |
@iglesiasg | sonney2k: yeah | 21:04 |
van51 | sonney2k: UTC+3 now | 21:05 |
votjakovr | sonney2k: UTC+7 | 21:05 |
@sonney2k | van51, you are travelling? | 21:05 |
@sonney2k | or why now? | 21:05 |
gsomix | lisitsyn, tnx | 21:06 |
van51 | sonney2k: heh no | 21:06 |
van51 | sonney2k: i was sure it was utc+2 | 21:06 |
van51 | and you said 21hrs | 21:06 |
van51 | so I looked it up | 21:06 |
van51 | because it's 10 o'clock here | 21:07 |
van51 | and it says UTC+2/3 | 21:07 |
van51 | so I'm guessing it has to do with daylight saving time | 21:07 |
gsomix | sonney2k, van51 not crazy, just bad documented | 21:08 |
@sonney2k | van51, where is that? | 21:08 |
van51 | sonney2k: athens. greece | 21:08 |
van51 | gsomix: I never said anything :p | 21:08 |
@sonney2k | van51, uh and you still talk to me - I am honored | 21:08 |
van51 | sonney2k: just because I have to :p | 21:09 |
@sonney2k | hehe | 21:10 |
@sonney2k | lisitsyn, do you recall which timezone flouwall is in? | 21:11 |
@lisitsyn | sonney2k: whole china is in the one timezone iirc | 21:12 |
@lisitsyn | sonney2k: +8 or so | 21:12 |
@sonney2k | lisitsyn, pickle27 was from ? | 21:13 |
@sonney2k | europe still? | 21:13 |
@lisitsyn | sonney2k: canada | 21:13 |
@lisitsyn | or so | 21:13 |
@sonney2k | I suckk | 21:13 |
@sonney2k | alright I recall | 21:13 |
@lisitsyn | ahh | 21:14 |
@lisitsyn | sonney2k: yes | 21:14 |
@lisitsyn | sonney2k: ontario | 21:14 |
@sonney2k | he is the one who doesn't want to get up early | 21:14 |
@sonney2k | but will have to if accepted | 21:14 |
@lisitsyn | sonney2k: haha | 21:14 |
@sonney2k | maybe he will reconsider :D | 21:14 |
pickle27 | sonney2k: lisitsyn yes Canada | 21:15 |
pickle27 | pacific time | 21:15 |
@sonney2k | pickle27, so UTC-4 ? | 21:15 |
pickle27 | wait woops | 21:15 |
pickle27 | eastern time | 21:15 |
pickle27 | TUC - 5 | 21:16 |
pickle27 | UTC* | 21:16 |
@sonney2k | pickle27, whats your local time now? | 21:16 |
pickle27 | 3:15 pm | 21:16 |
pickle27 | so 15:15 | 21:16 |
@sonney2k | UTC is 7pm | 21:16 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 21:16 | |
@sonney2k | err 9pm | 21:16 |
@sonney2k | err 7 pm | 21:17 |
* sonney2k dies | 21:17 | |
@sonney2k | so UTC-4 | 21:17 |
@lisitsyn | sonney2k: just use 24h format | 21:17 |
pickle27 | sonney2k: time zones are confusing | 21:17 |
pickle27 | sorry yeah its UTC - 4 for summer | 21:17 |
@sonney2k | lisitsyn, pickle27 started with am/pm to confuse me | 21:17 |
@lisitsyn | I don't know any reason to use it | 21:18 |
@sonney2k | lisitsyn, do you recall hushell's UTC? | 21:19 |
@sonney2k | timezone I mean | 21:19 |
@lisitsyn | sonney2k: oregon | 21:19 |
@lisitsyn | sonney2k: may be utc -5 | 21:19 |
@sonney2k | so alsoe UTC-4 | 21:19 |
@lisitsyn | sonney2k: west coast | 21:19 |
@sonney2k | vgorbati, your time zone? | 21:19 |
pickle27 | sonney2k I can get up early for meetings lol | 21:20 |
vgorbati | sonney2k: UTC+3:00 | 21:20 |
@sonney2k | so let see | 21:21 |
@sonney2k | cheng is UTC+10 | 21:21 |
@sonney2k | pickle27/hushell UTC-4 | 21:21 |
@sonney2k | so if we had a meeting at 11 UTC it would be 7-8 for pickle27/hushell and 21-22 for cheng | 21:23 |
@sonney2k | not too bad | 21:23 |
pickle27 | sonney2k: yeah I could do that | 21:24 |
vgorbati | sonney2k: just want to be sure: if I run 'python generator.py test.py', followed by 'python tester.py test.py' and they both give 'OK' - then the test can be un-blacklisted? | 21:29 |
van51 | vgorbati: yeah, but I just made a PR about those | 21:37 |
van51 | if you have gone over some, maybe we can double-check | 21:38 |
van51 | although, probably travis will let me know, right? | 21:38 |
vgorbati | van51: have you gone through all black-listed tests? | 21:42 |
-!- Netsplit *.net <-> *.split quits: sanyam | 21:42 | |
van51 | vgorbati: yeah | 21:43 |
vgorbati | van51: so, the green ones here https://github.com/shogun-toolbox/shogun/pull/1121/files are the tests that still fail? | 21:44 |
vgorbati | van51: not green ones, but white ones* | 21:44 |
vgorbati | van51: actually, both of them:) | 21:45 |
van51 | vgorbati: actually, both the green and white ones | 21:45 |
van51 | yea :) | 21:45 |
van51 | vgorbati: but that is on my machine, for some reason I see them failing on travis | 21:45 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Leaving] | 21:46 | |
@sonney2k | van51, well you didn't submit the new data | 21:46 |
@sonney2k | van51, let me try with your blacklist | 21:46 |
@sonney2k | lisitsyn, converter_tdistributedstochasticneighborembedding_modular.py setting 1/1 ERROR | 21:51 |
@lisitsyn | sonney2k: again | 21:51 |
@lisitsyn | :D | 21:51 |
vgorbati | lisitsyn: this one should be refactored | 21:52 |
@lisitsyn | oh yes sure | 21:52 |
vgorbati | sonney2k: it can't be OK without any changes | 21:52 |
@sonney2k | vgorbati, ok blacklisting it again | 21:52 |
vgorbati | sonney2k: I guess I will take this one to fix, since I am the one who wrote it:) | 21:53 |
van51 | does that one require a specific library? | 21:54 |
van51 | because it passed as ok to me | 21:54 |
vgorbati | van51: no, this one basically can return new values at every run, since it is calling randomized method | 21:56 |
vgorbati | van51: and since tester.py just compares the results with serialized file for strict equality, it will fail | 21:56 |
shogun-notifier- | shogun-data: Soeren Sonnenburg :master * afe8ada / testsuite/tests/ (28 files): https://github.com/shogun-toolbox/shogun-data/commit/afe8adad2b7dd633f94e2e20a9b0269db03aa16c | 21:57 |
shogun-notifier- | shogun-data: re-enable integration tests | 21:57 |
vgorbati | van51: or, at least, it CAN fail even if the method is OK | 21:57 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 683cab4 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/683cab4f9ace33f363ba2b0b6bb80456c50cf6e7 | 21:58 |
shogun-notifier- | shogun: re-enable integration tests (thanks to van51!) | 21:58 |
@sonney2k | vgorbati, maybe you could use the shogun random framework - then it should give some reproducible result | 21:58 |
@lisitsyn | sonney2k: I am merging tapkee with shogun random | 21:59 |
vgorbati | lisitsyn: so, if tapkee will use shogun random, I can set the seed at the beginning of the test, and then it should produce similar results? | 22:02 |
@lisitsyn | vgorbati: no idea to be honest | 22:02 |
@lisitsyn | vgorbati: as I am merging your code to the repo you will be mentioned in the acknowledgments of my paper :) | 22:06 |
vgorbati | lisitsyn: Does it mean I will become world-famous?:) | 22:06 |
@lisitsyn | vgorbati: not really but a bit :D | 22:07 |
@sonney2k | iglesiasg, btw what is up with https://github.com/shogun-toolbox/shogun/pull/1089 ? | 22:08 |
@sonney2k | iglesiasg, the code looked pretty ok, add some unit tests and get this merged | 22:09 |
@sonney2k | otherwise it will rot | 22:09 |
@sonney2k | (tm) | 22:10 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Quit: Page closed] | 22:11 | |
gsomix | sonney2k, hope, it's readable now. https://github.com/shogun-toolbox/shogun/pull/1119 | 22:20 |
@sonney2k | van51, haha look at https://travis-ci.org/shogun-toolbox/shogun/jobs/7404385 | 22:20 |
@sonney2k | basically all the things we enabled fail now | 22:20 |
@sonney2k | search for 'The following tests failed" | 22:21 |
shogun-buildbot | build #862 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/862 | 22:22 |
van51 | haha | 22:22 |
van51 | is it every single one that was enabled? | 22:22 |
@sonney2k | gsomix, no still not readable - the function is 100 lines | 22:23 |
@sonney2k | can you break it down to say 20 line functions? | 22:23 |
gsomix | sonney2k, of course. I'll do this. | 22:24 |
van51 | sonney2k: i'm telling you they are reporting OK here! | 22:24 |
@sonney2k | van51, yeah same here! | 22:25 |
van51 | sonney2k: good, I thought I was going crazy :p | 22:25 |
@sonney2k | van51, but not travis... I have no idea what travis does / runs on if you can figure out how we can simulate running on travis we could debug things | 22:25 |
@sonney2k | if not then we have to somehow ``fix'' things from where we are | 22:25 |
@iglesiasg | sonney2k: I will try to get some time for it ASAP | 22:27 |
-!- travis-ci [~travis-ci@ec2-184-73-46-130.compute-1.amazonaws.com] has joined #shogun | 22:28 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7404381 | 22:28 |
-!- travis-ci [~travis-ci@ec2-184-73-46-130.compute-1.amazonaws.com] has left #shogun [] | 22:28 | |
van51 | sonney2k: apart from "classifier_conjugateindex_modular.py" all other error are very small, right? | 22:29 |
@sonney2k | van51, yeah seems like | 22:30 |
@sonney2k | -bias float64 1.511179333277558 | 22:30 |
@sonney2k | +bias float64 1.511179333277562 | 22:30 |
@sonney2k | gsomix, btw please don't do 2*buffer size - use some configurable upper limit for a line (say 10MB) | 22:32 |
@sonney2k | van51, hmm the diff of the transfer_multitask_l12_logistic_regression* is weird | 22:33 |
@sonney2k | seems like it is empty or bug? | 22:34 |
van51 | sonney2k: it seems empty | 22:34 |
van51 | and in other files too | 22:34 |
shogun-buildbot | build #1155 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1155 blamelist: Soeren Sonnenburg <sonne@debian.org> | 22:46 |
hushell | sonney2k: I am in UTC -7, now is 13:52 | 22:53 |
hushell | If the meeting is UTC 11 am, seems would be 4 am for me :D | 22:53 |
@sonney2k | hushell, argh | 23:01 |
hushell | sonney2k: could you make it a bit eariler or late few hours? But I can also stay until 4am | 23:01 |
hushell | I am available from UTC 14:00 to UTC 09:00 | 23:02 |
@sonney2k | difficult! | 23:02 |
hushell | haha it's hard to figure out the right time | 23:03 |
@sonney2k | van51, I suggest to blacklist again for now | 23:03 |
van51 | sonney2k: probably.. I've been looking at it | 23:03 |
van51 | but nothing.. | 23:03 |
van51 | sonney2k: could it be a compiler thing? | 23:03 |
@sonney2k | hushell, think of UTC+10 in australia | 23:03 |
@sonney2k | van51, it could already be a different version of eigen/lapack/blas etc | 23:04 |
@sonney2k | and compiler too yes | 23:04 |
@sonney2k | van51, maybe try with clang... | 23:04 |
van51 | sonney2k: i just compiled wih that | 23:04 |
van51 | and ran just tester.py with some failing examples | 23:04 |
van51 | and they passed.. | 23:05 |
hushell | Well, I can make the meeting, but not everytime, that could be difficult :) | 23:05 |
van51 | sonney2k:w8 | 23:05 |
hushell | Maybe I should change a bit my schedule in summer | 23:05 |
@lisitsyn | hushell: we are going to have just a few meetings I think | 23:06 |
@lisitsyn | that's not day to day I mean :) | 23:06 |
hushell | :D once a week is fine for me | 23:07 |
hushell | and you know in summer always hot in daytime | 23:07 |
@sonney2k | hushell, about once a month | 23:07 |
hushell | sounds great! no problem! | 23:08 |
hushell | sonney2k: what the day would be the first meeting? | 23:10 |
-!- dvalcarce [5b757556@gateway/web/freenode/ip.91.117.117.86] has joined #shogun | 23:11 | |
dvalcarce | hi all | 23:11 |
van51 | sonney2k: classifier_conjugateindex_modular.py is failing for me too now | 23:16 |
van51 | and it's a big diff | 23:17 |
van51 | the others worked with either compiler | 23:17 |
@sonney2k | hushell, no idea yet | 23:20 |
@sonney2k | but post official google announcement :D | 23:20 |
gsomix | sonney2k, how about 30+ line functions? >:3 | 23:27 |
hushell | I'll have a nap that after dinner :) | 23:31 |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-oldqzgtrfytyuyka] has joined #shogun | 23:44 | |
--- Log closed Thu May 23 00:00:13 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!