--- Log opened Wed Jun 11 00:00:27 2014 | ||
-!- lambday [67157f4f@gateway/web/freenode/ip.103.21.127.79] has joined #shogun | 00:04 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 00:29 | |
shogun-notifier- | shogun: lambday :develop * d194676 / / (12 files): https://github.com/shogun-toolbox/shogun/commit/d194676f9fee8e3597a59ea790d1b9dc27a74d5e | 00:29 |
---|---|---|
shogun-notifier- | shogun: added proper lapack check to address issue #2291 | 00:29 |
shogun-notifier- | shogun: Soumyajit De :develop * 0e4bf9f / / (12 files): https://github.com/shogun-toolbox/shogun/commit/0e4bf9fe5a97e6318cf8f242e0c1553d51dcca1f | 00:29 |
shogun-notifier- | shogun: Merge pull request #2292 from lambday/develop | 00:29 |
shogun-notifier- | shogun: | 00:29 |
shogun-notifier- | shogun: added proper lapack check to address issue #2291 | 00:29 |
shogun-buildbot | build #2339 of bsd1 - libshogun is complete: Failure [failed compile test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2339 blamelist: lambday <heavensdevil6909@gmail.com> | 00:34 |
-!- thoralf [~thoralf@91-65-141-142-dynip.superkabel.de] has joined #shogun | 00:34 | |
thoralf | Heyhey. | 00:35 |
shogun-buildbot | build #2340 of bsd1 - libshogun is complete: Failure [failed compile test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2340 blamelist: Soumyajit De <heavensdevil6909@gmail.com> | 00:38 |
shogun-notifier- | shogun: Wu Lin :develop * 358fe87 / examples/undocumented/libshogun/variational_approx_example.cpp: https://github.com/shogun-toolbox/shogun/commit/358fe87eabdaefcaf0d8a82d53b4abd5481e4e56 | 00:38 |
shogun-notifier- | shogun: using small dataset | 00:38 |
shogun-notifier- | shogun: tklein23 :develop * 3a6592e / examples/undocumented/libshogun/variational_approx_example.cpp: https://github.com/shogun-toolbox/shogun/commit/3a6592e27fdb976e75a68bbbc7069b9f181a3ddf | 00:38 |
shogun-notifier- | shogun: Merge pull request #2286 from yorkerlin/variational-example | 00:38 |
shogun-notifier- | shogun: | 00:38 |
shogun-notifier- | shogun: variational_approx_example using small dataset (for memory check) | 00:38 |
shogun-notifier- | shogun: Wu Lin :develop * a64cbf0 / tests/unit/machine/gp/ProbitLikelihood_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/a64cbf07aa7fe5143089dbda3f4f2c6248ad20cc | 00:44 |
shogun-notifier- | shogun: fix the memory leak issue | 00:44 |
shogun-notifier- | shogun: Wu Lin :develop * 851a0be / src/shogun/machine/gp/ (2 files): https://github.com/shogun-toolbox/shogun/commit/851a0bea670fa93d7a8e174fd309ec9470805454 | 00:44 |
shogun-notifier- | shogun: add the base class for KL methods | 00:44 |
shogun-notifier- | shogun: tklein23 :develop * cac56dd / src/shogun/machine/gp/ (2 files): https://github.com/shogun-toolbox/shogun/commit/cac56dd961d573f6fde0dfded85b05e3fb848add | 00:44 |
shogun-notifier- | shogun: Merge pull request #2287 from yorkerlin/develop | 00:44 |
shogun-notifier- | shogun: | 00:44 |
shogun-notifier- | shogun: The base KL method class | 00:44 |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has quit [Ping timeout: 265 seconds] | 00:56 | |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has joined #shogun | 01:05 | |
-!- thoralf [~thoralf@91-65-141-142-dynip.superkabel.de] has quit [Ping timeout: 252 seconds] | 01:07 | |
-!- thoralf [~thoralf@91-65-141-142-dynip.superkabel.de] has joined #shogun | 01:24 | |
-!- thoralf [~thoralf@91-65-141-142-dynip.superkabel.de] has quit [Quit: Konversation terminated!] | 01:31 | |
shogun-buildbot | build #2341 of bsd1 - libshogun is complete: Failure [failed compile test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2341 blamelist: Wu Lin <yorker.lin@gmail.com>, tklein23 <tklein23@users.noreply.github.com> | 01:46 |
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has joined #shogun | 01:49 | |
-!- khalednasr [~k.nasr92@41.69.158.33] has quit [Ping timeout: 240 seconds] | 02:26 | |
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has quit [Remote host closed the connection] | 02:27 | |
shogun-buildbot | build #24 of nightly_fedora is complete: Failure [failed git] Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_fedora/builds/24 blamelist: lambday <heavensdevil6909@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com>, Wu Lin <yorker.lin@gmail.com>, tklein23 <tklein23@users.noreply.github.com>, Soumyajit De <heavensdevil6909@gmail.com>, | 03:00 |
shogun-buildbot | Fernando Iglesias <fernando.iglesiasg@gmail.com> | 03:00 |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has quit [Quit: PirosB3] | 03:18 | |
shogun-buildbot | build #338 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/338 blamelist: lambday <heavensdevil6909@gmail.com>, Soumyajit De <heavensdevil6909@gmail.com> | 03:38 |
shogun-notifier- | shogun: lambday :develop * f3e1f59 / src/shogun/statistics/ (2 files): https://github.com/shogun-toolbox/shogun/commit/f3e1f598412d46a9978b9ff312c70387737adf85 | 03:41 |
shogun-notifier- | shogun: prevent failure in class_list.cpp when LAPACK is not present | 03:41 |
shogun-notifier- | shogun: Soumyajit De :develop * df92341 / src/shogun/statistics/ (2 files): https://github.com/shogun-toolbox/shogun/commit/df92341eff532339849dd5c7b405f1e90fee815a | 03:41 |
shogun-notifier- | shogun: Merge pull request #2294 from lambday/develop | 03:41 |
shogun-notifier- | shogun: | 03:41 |
shogun-notifier- | shogun: prevent failure in class_list.cpp when LAPACK is not present | 03:41 |
-!- travis-ci [~travis-ci@ec2-54-237-208-250.compute-1.amazonaws.com] has joined #shogun | 03:56 | |
travis-ci | [travis-ci] it's Soumyajit De'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/27272400 | 03:56 |
-!- travis-ci [~travis-ci@ec2-54-237-208-250.compute-1.amazonaws.com] has left #shogun [] | 03:56 | |
-!- soumyaC [uid15286@gateway/web/irccloud.com/x-zmdztwwnxsgnkaql] has quit [Quit: Connection closed for inactivity] | 04:26 | |
-!- lambday [67157f4f@gateway/web/freenode/ip.103.21.127.79] has quit [Ping timeout: 246 seconds] | 05:40 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 06:42 | |
-!- abinash_panda [~abinash@117.200.236.154] has joined #shogun | 07:58 | |
shogun-buildbot | build #823 of nightly_default is complete: Failure [failed notebooks] Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/823 blamelist: lambday <heavensdevil6909@gmail.com>, Parijat Mazumdar <mazumdarparijat@gmail.com>, Wu Lin <yorker.lin@gmail.com>, tklein23 <tklein23@users.noreply.github.com>, Soumyajit De | 08:12 |
shogun-buildbot | <heavensdevil6909@gmail.com>, Fernando Iglesias <fernando.iglesiasg@gmail.com> | 08:12 |
-!- lisitsyn [~qdrgsm@80.252.20.67] has joined #shogun | 08:18 | |
shogun-buildbot | build #2342 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2342 blamelist: lambday <heavensdevil6909@gmail.com>, Soumyajit De <heavensdevil6909@gmail.com> | 08:21 |
-!- abinash_panda [~abinash@117.200.236.154] has quit [Ping timeout: 245 seconds] | 09:14 | |
-!- lisitsyn [~qdrgsm@80.252.20.67] has quit [Quit: Leaving.] | 09:36 | |
-!- khalednasr [~k.nasr92@41.69.223.215] has joined #shogun | 09:54 | |
shogun-buildbot | build #339 of debian wheezy - memcheck is complete: Success [build successful] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/339 | 10:11 |
-!- khalednasr [~k.nasr92@41.69.223.215] has quit [Ping timeout: 260 seconds] | 10:13 | |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has joined #shogun | 10:39 | |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has quit [Quit: PirosB3] | 11:12 | |
-!- Saurabh7 [~Saurabh7@117.200.188.89] has joined #shogun | 11:15 | |
-!- tp199911 [863a27df@gateway/web/freenode/ip.134.58.39.223] has joined #shogun | 11:28 | |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has joined #shogun | 11:50 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 12:03 | |
-!- khalednasr [~k.nasr92@41.69.235.207] has joined #shogun | 12:43 | |
-!- Saurabh7 [~Saurabh7@117.200.188.89] has quit [Quit: Leaving] | 13:03 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 13:13 | |
shogun-notifier- | shogun: Parijat Mazumdar :develop * 79ac6c8 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/79ac6c865d9d587720cd3e315d2522572d5444c2 | 13:13 |
shogun-notifier- | shogun: modular API, bug fix for C4.5 | 13:13 |
shogun-notifier- | shogun: Fernando Iglesias :develop * 4f85314 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/4f85314a51267c411318279e3d1dd46373abdad2 | 13:13 |
shogun-notifier- | shogun: Merge pull request #2293 from mazumdarparijat/rf | 13:13 |
shogun-notifier- | shogun: | 13:13 |
shogun-notifier- | shogun: modular API, bug fix for C4.5 | 13:13 |
shogun-buildbot | build #2343 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2343 blamelist: Parijat Mazumdar <mazumdarparijat@gmail.com> | 13:20 |
shogun-buildbot | build #2344 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2344 blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com> | 13:25 |
-!- khalednasr [~k.nasr92@41.69.235.207] has quit [Quit: Leaving] | 13:27 | |
-!- travis-ci [~travis-ci@ec2-54-82-55-141.compute-1.amazonaws.com] has joined #shogun | 14:17 | |
travis-ci | [travis-ci] it's Fernando Iglesias'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/27299542 | 14:17 |
-!- travis-ci [~travis-ci@ec2-54-82-55-141.compute-1.amazonaws.com] has left #shogun [] | 14:17 | |
-!- Saurabh7 [~Saurabh7@117.200.188.89] has joined #shogun | 14:20 | |
-!- kislay [~Abhijeet@103.240.206.231] has joined #shogun | 14:21 | |
-!- PirosB3 [~pirosb3@host80-223-dynamic.246-95-r.retail.telecomitalia.it] has quit [Quit: PirosB3] | 15:17 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 16:13 | |
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has joined #shogun | 16:14 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 16:19 | |
-!- pickle27 [~pickle27@192-0-136-118.cpe.teksavvy.com] has quit [Remote host closed the connection] | 16:20 | |
-!- tp199911 [863a27df@gateway/web/freenode/ip.134.58.39.223] has quit [Quit: Page closed] | 16:41 | |
-!- abinash_panda [~abinash@117.200.235.72] has joined #shogun | 19:14 | |
-!- Saurabh7 [~Saurabh7@117.200.188.89] has quit [Quit: Leaving] | 19:36 | |
-!- thoralf [~thoralf@91-65-141-142-dynip.superkabel.de] has joined #shogun | 20:06 | |
thoralf | Heyhey | 20:06 |
thoralf | abinash_panda: Good morning ,) | 20:07 |
abinash_panda | hey thoralf | 20:26 |
abinash_panda | its actually night over here ;) | 20:26 |
thoralf | abinash_panda: What are you currently working on? | 20:39 |
abinash_panda | Currently, I am working on the feature hashing. The gist you had sent me today was really helpful :) | 20:39 |
abinash_panda | thoralf: I have a few questions.. | 20:40 |
thoralf | abinash_panda: Shoot. | 20:40 |
abinash_panda | thoralf: First, what should be class name of hashed features, I propose HashedSparseMultilabelFeatures. Any suggestions? | 20:40 |
thoralf | I think "Sparse" is not neccessary. | 20:41 |
thoralf | necessary | 20:41 |
abinash_panda | thoralf: Okay, so we can go with HashedMultilabelFeatures | 20:41 |
abinash_panda | thoralf: But, I would be supporting mainly hashed features, so I was thinking of adding *Sparse*. Okay, no problem | 20:42 |
thoralf | If you find this important, let's call it HashedSparse... | 20:42 |
abinash_panda | Okay, fine. | 20:43 |
abinash_panda | thoralf: Second, the user should provide the seeds for hashing right? | 20:43 |
thoralf | abinash_panda: Well, for the beginning, you can just say seed for the i-th label is "i" ;) | 20:44 |
abinash_panda | So, I am thinking the constructor would be something like CHashedSparseMultilabelFeatures<ST>(CSparseFeatures* feats, int32_t d, SGVector<uint32_t> seeds,...) | 20:44 |
thoralf | What's d? | 20:44 |
abinash_panda | d is the dimension of the new feature | 20:45 |
thoralf | Ah, of course. | 20:45 |
thoralf | Yes. | 20:45 |
abinash_panda | So, if the user doesn't provide the seeds we can go for seed of i-th label as "i" | 20:45 |
thoralf | Yes, that would be helpful. | 20:46 |
abinash_panda | so another constructor would be something like CHashedSparseMultilabelFeatures(feats, dim, num_classes, ...) | 20:46 |
thoralf | num_labels? | 20:46 |
thoralf | Let me check. ;) | 20:47 |
abinash_panda | okay | 20:47 |
thoralf | num_classes, you're right. | 20:48 |
thoralf | This is a bit confusing. | 20:48 |
abinash_panda | okay...num_classes means the total number of classes we are having...right? | 20:49 |
thoralf | abinash_panda: Yes. | 20:49 |
thoralf | And num_labels should be called num_examples to avoid confusion. | 20:50 |
thoralf | But it's too late now. | 20:50 |
thoralf | (It's never too late... but. ;) | 20:50 |
abinash_panda | yeah...I would try to modify it wherever I can | 20:50 |
thoralf | abinash_panda: I guess the implementation is straight forward. Do you think you have to implement more that get_joint_feature_vector and get_dim? | 20:51 |
abinash_panda | get_joint_feature and get_dim are the methods of a StructuredModel, but I am implementing a DotFeature class. | 20:52 |
abinash_panda | I am thinking of having a class for hashed features and not modifying the MultilabelModel | 20:53 |
abinash_panda | Any suggestions? | 20:53 |
thoralf | I think it would be easier to don't implement own features, because the joint features is not a tensor any more. | 20:56 |
abinash_panda | Now, I am getting confused. | 20:57 |
abinash_panda | If you could explain me the hashing once again. | 20:58 |
abinash_panda | What I thought that we would be hashing the feature vector, so it would be better to implement our own features. | 20:59 |
thoralf | normal so learning optimizes < w ; PSI(x,y) > with PSI(x,y) = x (x) y where (x) denotes tensor product | 21:04 |
thoralf | hashing optimizes < w ; PSI(x,y) > with PSI(x,y) = sum_{i \in y} phi(x,i) | 21:06 |
thoralf | where phi is a hashing function parametrized by i | 21:06 |
thoralf | or sum_i [[y_i == 1]] * phi_i(x) | 21:07 |
thoralf | Same thing, different notation | 21:07 |
abinash_panda | Okay, now I get it. | 21:08 |
abinash_panda | Previously, I was thinking of something else. | 21:08 |
abinash_panda | so, now just PSI(x, y) changes meaning I have to change get_joint_feature_vector and get_dim | 21:08 |
abinash_panda | So, now the MultilabelModel should have a constructor something like this MultilabelModel(CFeatures* features, CStructuredLabels* labels, bool hashing = false, int32_t dim = 0) | 21:10 |
thoralf | I still would create a new Hashing(Sparse;))MultilabelModel | 21:18 |
thoralf | There is some duplication, but the functionality would be too different. | 21:18 |
abinash_panda | Yeah that would be better. | 21:18 |
thoralf | If we see too much duplicated behaviour, we can talk about it. | 21:19 |
thoralf | Ah, well. You also have to implement a slightly different version of argmax. | 21:19 |
thoralf | Because the current one assumes that each class has it's own features. | 21:19 |
thoralf | The new one can just do < w, phi_(x) > to find out the score for label i | 21:20 |
abinash_panda | you mean <w, phi_i(x)> where phi_i(x) would be hashing function parameterized by i..right? | 21:22 |
thoralf | Yeah. | 21:22 |
abinash_panda | Okay. Got it. | 21:23 |
thoralf | This can be used to get the score of label i. | 21:23 |
thoralf | It still decomposes in a linear way, so we can put the sum into the scalar product or outside. | 21:23 |
thoralf | < w ; sum_{i \in y} phi(x,i) > = sum_{i \in y} < w ; phi(x,i) > | 21:24 |
abinash_panda | and if the score of label i > 0, then we should predict i | 21:25 |
thoralf | Exact. | 21:25 |
thoralf | As usual :) | 21:25 |
abinash_panda | Okay. So, I going to implement it now :) | 21:26 |
thoralf | Yeah, good luck. | 21:27 |
thoralf | You should start thinking about some small plot/example/ipython notebook. | 21:27 |
thoralf | Only gathering ideas what to evaluate/write/plot on which (toy?) data... | 21:28 |
abinash_panda | Yeah...I would do it. I was writing it earlier, and suddenly my computer got crashed :( | 21:29 |
abinash_panda | All of it got deleted, as I had to format it. | 21:30 |
abinash_panda | So, I would do it. | 21:30 |
abinash_panda | BTW, have you used the PrimalMosekSOSVM solver? | 21:31 |
-!- lisitsyn [~qdrgsm@85.114.187.192] has joined #shogun | 21:46 | |
kislay | hey lisitsyn :) | 21:46 |
lisitsyn | kislay: hey | 21:46 |
kislay | congo for the MSC man :D | 21:46 |
lisitsyn | kislay: thanks! | 21:47 |
lisitsyn | kislay: how is it going? | 21:47 |
kislay | Fine. I have tried to implement the macro thing you commented there. | 21:47 |
lisitsyn | kislay: cool, is it better now? | 21:48 |
kislay | yeah | 21:49 |
kislay | was able to cut lots of code. | 21:49 |
kislay | See the PR when you get free. I have to keep improving it. | 21:50 |
thoralf | abinash_panda: No, didn't use it so far. | 21:50 |
thoralf | Hey Master L | 21:51 |
kislay | Hey T!! how's you? :) | 21:51 |
thoralf | kislay: I'm fine. :) | 21:52 |
abinash_panda | thoralf: Actually, I have used it on my machine and found that it took a lot of time to train. | 21:52 |
thoralf | Very hot here and glad it's evening and got a bit colder :) | 21:52 |
thoralf | abinash_panda: Yeah, they all take some time. I thing SGD was the fastest one. | 21:52 |
thoralf | think | 21:53 |
kislay | hah! we have 46 degrees here! in India!! | 21:53 |
abinash_panda | thoralf: The toy example provided in so_multiclass.cpp worked fine, but when I had increased the features and examples in hundreds of thousands, I had to reboot my computer. | 21:53 |
thoralf | kislay: degree celsius? | 21:53 |
kislay | yup yup | 21:53 |
thoralf | abinash_panda: Before starting the script, call "ulimit -v <kbytes>" to save your machine ;) | 21:54 |
kislay | thoralf, Thats actually normal summer temp. here. | 21:54 |
thoralf | Set it to something like 80% of your RAM. | 21:54 |
thoralf | kislay: That's crazy. Who's responsible for this decision? ;) | 21:54 |
abinash_panda | thoralf: Okay. I didn't know about. Thanks :) | 21:55 |
thoralf | abinash_panda: Must be called in the same shell as used for running the command. | 21:55 |
thoralf | Has no affect on already running processes or other shells. | 21:55 |
abinash_panda | thoralf: yeah..it would use it | 21:56 |
kislay | thoralf, the equator crosses India... probably thats why! | 21:57 |
lisitsyn | thoralf: hey | 21:59 |
-!- kislay [~Abhijeet@103.240.206.231] has quit [Ping timeout: 245 seconds] | 22:12 | |
-!- abinash_panda [~abinash@117.200.235.72] has quit [Quit: leaving] | 22:13 | |
-!- zxtx [~zv@ool-45763ca9.dyn.optonline.net] has quit [Ping timeout: 260 seconds] | 22:22 | |
-!- lisitsyn [~qdrgsm@85.114.187.192] has quit [Read error: Connection reset by peer] | 22:32 | |
-!- lisitsyn [~qdrgsm@85.114.187.192] has joined #shogun | 22:32 | |
-!- Netsplit *.net <-> *.split quits: sonney2k, naywhayare | 22:40 | |
-!- Netsplit over, joins: sonney2k, naywhayare | 22:40 | |
-!- lisitsyn [~qdrgsm@85.114.187.192] has quit [Read error: Connection reset by peer] | 23:45 | |
-!- khalednasr [~k.nasr92@41.69.234.202] has joined #shogun | 23:59 | |
--- Log closed Thu Jun 12 00:00:29 2014 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!