| --- Log opened Sun Mar 01 00:00:40 2015 | ||
| -!- HeikoS [~heiko@2.221.116.151] has joined #shogun | 00:09 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 00:09 | |
| @lisitsyn | HeikoS: yeah you get it right I think :) | 00:20 |
|---|---|---|
| @lisitsyn | HeikoS: the other thing - you know why the pointer is null | 00:20 |
| @HeikoS | because the thing was freed? | 00:20 |
| -!- vortex_ape [~vortex_ap@120.59.205.250] has joined #shogun | 00:20 | |
| @lisitsyn | no I mean | 00:20 |
| @lisitsyn | if you have some wrapper for the thing that is potentially absent | 00:21 |
| @lisitsyn | you can check whether it is absent | 00:21 |
| @lisitsyn | and then if it is absent you can log something | 00:21 |
| @HeikoS | I sewe | 00:21 |
| @HeikoS | see | 00:21 |
| @HeikoS | cool thing | 00:21 |
| @HeikoS | lisitsyn: what about this: https://github.com/shogun-toolbox/shogun/wiki/GSoC_2015_plugin | 00:22 |
| @lisitsyn | HeikoS: some very common use case is to find something in collection | 00:22 |
| @HeikoS | description is too short | 00:22 |
| @HeikoS | timeline, waypoints, entrance tasks, suggested reading, examples etc | 00:22 |
| @lisitsyn | HeikoS: oh yeah let me extend it | 00:23 |
| @lisitsyn | will do it tomorrow | 00:23 |
| @HeikoS | cool | 00:23 |
| @HeikoS | lisitsyn: and this one would also be good to have adescription for | 00:23 |
| @HeikoS | https://github.com/shogun-toolbox/shogun/wiki/GSoC_2015_project_parameters | 00:23 |
| @HeikoS | with your new ideas for that | 00:23 |
| @lisitsyn | ok | 00:23 |
| @lisitsyn | HeikoS: good | 00:24 |
| @HeikoS | btw did you look at this user guide thing? | 00:24 |
| @lisitsyn | sorry not yet | 00:24 |
| @HeikoS | https://github.com/karlnapf/shogun_manual | 00:24 |
| @lisitsyn | looks interesting | 00:25 |
| @HeikoS | we should get this before gsoc | 00:25 |
| @HeikoS | and the meta examples too | 00:25 |
| @lisitsyn | HeikoS: get like get merged? | 00:25 |
| @HeikoS | so that every student can write like 1 entry each week, and translate 3 minimal examples | 00:25 |
| @HeikoS | lisitsyn: yes, and easy to extend | 00:25 |
| @HeikoS | meta examples are in | 00:25 |
| @HeikoS | I am currently editing the travis file to include it in the build | 00:26 |
| @HeikoS | no idea how to change buildbot | 00:26 |
| @lisitsyn | hmmm I thought someone knows how to change buildbot | 00:28 |
| @lisitsyn | I guess only wiking | 00:28 |
| @HeikoS | lisitsyn: just emailed wiking :) | 00:43 |
| kingsley | Do you happen to know if anyone has tried optimizing the hyperparameters passed to machine learning algorithms (like the number of trees to be used by a random forest) with pre-existing non-linear optimization code like gradient descent or integer programming? | 00:44 |
| kingsley | I'd like to think that a guided search of hyperparaemters would be more efficient than guessing, trial and error. | 00:45 |
| @lisitsyn | HeikoS: ^ I think you're expert here ;) | 00:45 |
| @HeikoS | kingsley: for sure | 00:45 |
| @lisitsyn | kingsley: people like to model performance as gaussian processes | 00:46 |
| kingsley | HeikoS: Please elaborate. | 00:46 |
| @lisitsyn | that's how MOE and hyperopt work | 00:46 |
| @HeikoS | kingsley: what people often do is to use stochastic black box optimisers for example, see cma | 00:46 |
| @HeikoS | kingsley: or look up bayesian optimisation, which is also about this idea | 00:46 |
| @lisitsyn | https://github.com/Yelp/MOE | 00:46 |
| @lisitsyn | https://github.com/hyperopt/hyperopt | 00:47 |
| @lisitsyn | https://github.com/JasperSnoek/spearmint | 00:47 |
| @HeikoS | where the idea is to put a gaussian process on the surface of the function your are optimising, and then use information theoretic ideas to select the next point to look at using a exploration-exploitation tradeoff | 00:47 |
| kingsley | OK, you have given me much to think about. | 00:48 |
| @HeikoS | kingsley: would be a cool thing to have for Shogun | 00:48 |
| kingsley | I'll check them out. | 00:48 |
| @lisitsyn | HeikoS: btw | 00:49 |
| @HeikoS | discrete case is another thing | 00:49 |
| @HeikoS | much much harder to do | 00:49 |
| @lisitsyn | I was curious about MOE | 00:49 |
| @lisitsyn | have you seen how it works? | 00:49 |
| @lisitsyn | wanted to ask you whether we can do the same in shogun | 00:49 |
| @HeikoS | MOE uses bayesian optimisation | 00:49 |
| @HeikoS | I invited Matthew Hoffman to give a talk here at UCL on Friday | 00:50 |
| @lisitsyn | yes with GPs | 00:50 |
| @HeikoS | http://mlg.eng.cam.ac.uk/hoffmanm/ | 00:50 |
| @lisitsyn | so can we do this with shogun and if not what is missed? | 00:50 |
| @HeikoS | lisitsyn, kingsley https://github.com/mwhoffman/pybo | 00:50 |
| @HeikoS | thats the way to do this | 00:50 |
| kingsley | HeikoS and lisitsyn: Thank you very much for saving me from re-inventing the wheel. | 00:51 |
| @HeikoS | lisitsyn: would be cool to use MOE in shogun, what do you think? | 00:51 |
| @lisitsyn | HeikoS: yes or do the same thing with our own GPs | 00:51 |
| @HeikoS | lisitsyn: though its not parallelisable | 00:51 |
| @lisitsyn | that's why I am asking ;) | 00:51 |
| @HeikoS | lisitsyn: cma is better in that respect | 00:51 |
| @HeikoS | we could use our own gps sure | 00:52 |
| @HeikoS | but why do that? | 00:52 |
| @HeikoS | people have done it, its just a matter of plugging things together | 00:52 |
| @HeikoS | the bayesian thing is hard to parallelise | 00:52 |
| @HeikoS | cma is easier to | 00:52 |
| @HeikoS | based on particles | 00:52 |
| @lisitsyn | what's cma? | 00:53 |
| @lisitsyn | covariance matrix adaptation? | 00:53 |
| @HeikoS | yep | 00:53 |
| @HeikoS | that works very nicely | 00:53 |
| @HeikoS | I use it for some things sometimes | 00:53 |
| @HeikoS | lisitsyn: though the thing is | 00:54 |
| @HeikoS | cma is a one-guy library and also method | 00:54 |
| @HeikoS | whereas bayesina optimisation is a large field with lots of excellent people doing research on | 00:54 |
| @HeikoS | and yelp is a kind of plus too | 00:54 |
| @lisitsyn | we should have both | 00:54 |
| @lisitsyn | HeikoS: interfacing with MOE is easy | 00:55 |
| @HeikoS | lisitsyn: thats cool | 00:55 |
| @HeikoS | we should do that | 00:55 |
| @HeikoS | its important | 00:55 |
| @HeikoS | but next year | 00:55 |
| @HeikoS | thats part of making model-selection easier and offer full pipelines | 00:55 |
| @HeikoS | but we are not there yet | 00:55 |
| @HeikoS | need other things first | 00:55 |
| @lisitsyn | I'd do that in a few hours I guess | 00:55 |
| @HeikoS | maybe | 00:56 |
| @HeikoS | yeah cool thing in fact | 00:56 |
| @HeikoS | though I wonder... maybe you should spend these hours on something else currently?? | 00:56 |
| @HeikoS | we have so many things that are not working properly | 00:56 |
| @lisitsyn | yes :) | 00:56 |
| @lisitsyn | I just mean it is not a big deal to interface with that | 00:57 |
| @lisitsyn | they have a few examples | 00:57 |
| @HeikoS | hyperparameters are not really an issue currently | 00:57 |
| @HeikoS | yeah I have seen | 00:57 |
| @HeikoS | but then we need | 00:57 |
| @HeikoS | cmake options | 00:57 |
| @HeikoS | a build | 00:57 |
| @HeikoS | tests | 00:57 |
| @HeikoS | etc | 00:57 |
| @HeikoS | so not sure | 00:57 |
| @HeikoS | whether its a good time now | 00:57 |
| @HeikoS | hackathon thing | 00:57 |
| @HeikoS | actually | 00:57 |
| @lisitsyn | no it is not :) | 00:57 |
| @HeikoS | that'd be cool | 00:57 |
| @lisitsyn | do you think hackathon thing is possible still? | 00:58 |
| @HeikoS | yes | 00:59 |
| @lisitsyn | HeikoS: UK won't let me in :D | 00:59 |
| @HeikoS | but I cannot do this alone | 00:59 |
| @HeikoS | we need: | 00:59 |
| @HeikoS | 1.) BSD | 00:59 |
| @HeikoS | 2.) foundation | 00:59 |
| @HeikoS | 3.) user guide | 00:59 |
| @HeikoS | without that it will not work | 00:59 |
| @HeikoS | oh and | 00:59 |
| @HeikoS | 0.) installation | 00:59 |
| @HeikoS | lisitsyn: why wont they let you in? | 00:59 |
| @lisitsyn | HeikoS: cold war :) | 01:00 |
| @lisitsyn | well actually this list is doable | 01:00 |
| @HeikoS | lisitsyn: it IS doable! | 01:01 |
| @lisitsyn | HeikoS: there is a wisdom I've learnt at some point | 01:01 |
| @HeikoS | which is? | 01:01 |
| @lisitsyn | to develop something it is usually enough to set up a cycle | 01:01 |
| @lisitsyn | when you change, compile and test | 01:01 |
| @lisitsyn | and to make people use software you just have to make it installable :) | 01:02 |
| @HeikoS | hehe | 01:02 |
| @HeikoS | yes | 01:02 |
| @lisitsyn | even with scarce functionality | 01:02 |
| @lisitsyn | it will be used | 01:02 |
| @lisitsyn | if it just works | 01:02 |
| @lisitsyn | you know people use VW so much | 01:03 |
| @HeikoS | yes I agree | 01:03 |
| @HeikoS | it is cool | 01:03 |
| @HeikoS | shogun is not installable | 01:03 |
| @HeikoS | people fail all the time | 01:03 |
| @HeikoS | half my institude tried it | 01:03 |
| @HeikoS | institute | 01:03 |
| @lisitsyn | haha | 01:03 |
| @HeikoS | smart people you know | 01:03 |
| @lisitsyn | yeah that's the priority one thing | 01:03 |
| @HeikoS | not linux experts but know things | 01:03 |
| @HeikoS | but they just are overwhealmed | 01:03 |
| @HeikoS | so the install project is a must | 01:04 |
| @lisitsyn | we can even have two students on that | 01:04 |
| @HeikoS | if we dont get a student for that, I refuse to do GSoC :D | 01:04 |
| @lisitsyn | you're still phd | 01:04 |
| @lisitsyn | student I mean | 01:04 |
| @HeikoS | I cannot do it | 01:04 |
| @lisitsyn | :) | 01:04 |
| @lisitsyn | why? :D | 01:04 |
| @lisitsyn | admin? | 01:04 |
| @lisitsyn | :) | 01:04 |
| @HeikoS | I dont know these things | 01:04 |
| @lisitsyn | ah | 01:04 |
| @HeikoS | also no time | 01:04 |
| @lisitsyn | yeah sure :) | 01:04 |
| @HeikoS | viktor could do it | 01:05 |
| @lisitsyn | I am not student and corporate slave | 01:05 |
| @HeikoS | but he also doesnt have time | 01:05 |
| @HeikoS | you could do it | 01:05 |
| @HeikoS | but also no time | 01:05 |
| @HeikoS | so we need a student who somehow knows | 01:05 |
| @lisitsyn | remember I am not a student :) | 01:05 |
| @HeikoS | I do | 01:05 |
| @lisitsyn | good time to sleep | 01:06 |
| @HeikoS | same here | 01:06 |
| @HeikoS | good night! | 01:06 |
| @lisitsyn | HeikoS: I'll extend these gsoc ideas tomorrow | 01:06 |
| @lisitsyn | see you! gntie | 01:06 |
| -!- vortex_ape [~vortex_ap@120.59.205.250] has quit [Ping timeout: 246 seconds] | 01:20 | |
| -!- vortex_ape [~vortex_ap@120.59.77.110] has joined #shogun | 01:36 | |
| -!- vortex_ape [~vortex_ap@120.59.77.110] has quit [Quit: Leaving] | 02:02 | |
| -!- HeikoS [~heiko@2.221.116.151] has quit [Ping timeout: 245 seconds] | 02:10 | |
| -!- HeikoS [~heiko@2.221.116.151] has joined #shogun | 02:15 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 02:15 | |
| -!- HeikoS [~heiko@2.221.116.151] has quit [Quit: Leaving.] | 03:54 | |
| -!- HeikoS [~heiko@2.221.116.151] has joined #shogun | 03:57 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 03:57 | |
| -!- rajul [~rajul@node-wns.pool-125-24.dynamic.totbb.net] has joined #shogun | 04:04 | |
| -!- HeikoS [~heiko@2.221.116.151] has quit [Quit: Leaving.] | 04:06 | |
| -!- curiousguy13_ [~curiousgu@120.59.228.160] has quit [Ping timeout: 256 seconds] | 04:38 | |
| -!- curiousguy13_ [~curiousgu@59.177.205.235] has joined #shogun | 04:51 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has joined #shogun | 04:58 | |
| -!- rajul [~rajul@node-wns.pool-125-24.dynamic.totbb.net] has quit [Ping timeout: 272 seconds] | 05:06 | |
| -!- Floatingman [~Floatingm@c-68-52-34-232.hsd1.tn.comcast.net] has quit [Remote host closed the connection] | 05:19 | |
| -!- rajul [~rajul@node-wns.pool-125-24.dynamic.totbb.net] has joined #shogun | 05:20 | |
| -!- Floatingman [~Floatingm@c-68-52-34-232.hsd1.tn.comcast.net] has joined #shogun | 05:23 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 06:05 | |
| -!- rajul [~rajul@node-wns.pool-125-24.dynamic.totbb.net] has quit [Read error: Connection reset by peer] | 06:13 | |
| -!- rajul [~rajul@node-6gf.pool-180-180.dynamic.totbb.net] has joined #shogun | 06:29 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has quit [Ping timeout: 255 seconds] | 06:43 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 256 seconds] | 07:49 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has joined #shogun | 07:50 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 08:11 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 244 seconds] | 08:19 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has quit [Ping timeout: 240 seconds] | 08:34 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 08:50 | |
| -!- joergkurtwegner [~joergkurt@d54c5db3e.access.telenet.be] has joined #shogun | 09:41 | |
| joergkurtwegner | hi, does anyone have a clean ubuntu 14.04 install routine? | 09:42 |
| -!- joergkurtwegner [~joergkurt@d54c5db3e.access.telenet.be] has quit [Client Quit] | 09:42 | |
| -!- joergkurt [~joergkurt@178.162.199.93] has joined #shogun | 09:45 | |
| joergkurt | ping is online... | 09:46 |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has joined #shogun | 09:50 | |
| -!- smallgraph [~textual@vpn-cl-194-80.scc.kit.edu] has joined #shogun | 10:07 | |
| joergkurt | hi, I am looking for a clean ubuntu install instruction for python-modular, will be back in 2-3h | 10:12 |
| -!- curiousguy13_ [~curiousgu@59.177.205.235] has quit [Ping timeout: 245 seconds] | 10:39 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 264 seconds] | 10:40 | |
| -!- vortex_ape [~vortex_ap@59.177.140.38] has joined #shogun | 10:44 | |
| joergkurt | with clean I mean fresh Ubuntu14.04 in VirtualBox and then a native from GIT source, if all dependencies are properly defined in the install docu, all should just compile through, I guess we neeed some refinement on this process, since my naive attempt does not work right away | 10:48 |
| -!- curiousguy13_ [~curiousgu@120.59.226.121] has joined #shogun | 10:52 | |
| -!- kingsley_ [~kingsley@174-21-235-71.tukw.qwest.net] has joined #shogun | 11:07 | |
| -!- kingsley [~kingsley@174-21-171-80.tukw.qwest.net] has quit [Ping timeout: 255 seconds] | 11:10 | |
| @lisitsyn | joergkurt: hey | 11:21 |
| -!- smallgraph [~textual@vpn-cl-194-80.scc.kit.edu] has quit [Ping timeout: 256 seconds] | 11:22 | |
| joergkurt | hoi | 11:22 |
| @lisitsyn | joergkurt: looks like you've already installed shogun :) | 11:22 |
| @lisitsyn | I mean from you mail I see you've run an example and it worked | 11:23 |
| joergkurt | nope, its compiled, but not running :( | 11:23 |
| @lisitsyn | what do you mean? | 11:23 |
| @lisitsyn | what if you run python and write 'from modshogun import KMeans'? | 11:23 |
| joergkurt | import works, kinda busy right now, can I get back to you in 2h...? | 11:25 |
| joergkurt | sorry, just have to do something else ...will get back to it later... | 11:25 |
| @lisitsyn | joergkurt: sure. but if import works probably everything is probably ok anyway :) | 11:26 |
| joergkurt | ok, lets figure it later :) | 11:26 |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 12:54 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has joined #shogun | 12:58 | |
| joergkurt | ok, my bet, it works, I was kinda expecting distance_tanimoto_modular.py returns some console printout, its not | 13:06 |
| joergkurt | so, seems to works just fine, one other question | 13:06 |
| joergkurt | if loading a dat set like feats_train=RealFeatures(CSVFile(train_fname)) | 13:06 |
| joergkurt | how can I get this into a python object I can also process on the Python side, and vice versa? | 13:07 |
| joergkurt | Second question, how can I load a sparse file, and which formats are accepted, also for usage in Python? | 13:07 |
| -!- vortex_ape [~vortex_ap@59.177.140.38] has quit [Ping timeout: 264 seconds] | 13:23 | |
| @lisitsyn | ok first | 13:43 |
| @lisitsyn | joergkurt: what do you mean python object? | 13:44 |
| @lisitsyn | I am not sure I get what you want to do with that | 13:44 |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 250 seconds] | 14:15 | |
| -!- exploreshaifali [~shaifali@182.70.196.201] has joined #shogun | 14:22 | |
| joergkurt | maybe I just dont know all details, the CSVFile is loaded, how to I access the elements in this object? | 14:25 |
| joergkurt | feats_train=RealFeatures(CSVFile(train_fname)), aka how does a data object look like and what are access functions? | 14:26 |
| joergkurt | since its SWIG I assume the data is staying at the C++ side and Python is just pointing to it | 14:34 |
| @lisitsyn | joergkurt: yes | 14:36 |
| @lisitsyn | joergkurt: it depends what you want to do | 14:37 |
| @lisitsyn | simple access like per vector is easy | 14:38 |
| @lisitsyn | use get_feature_vector | 14:38 |
| @lisitsyn | if you want to change data you'd probably have to get vector, modify it and set_feature_vector | 14:38 |
| joergkurt | the fecture vector requires an index, so do I need to do this element-wise? | 14:42 |
| @lisitsyn | joergkurt: that's the index of vector | 14:43 |
| @lisitsyn | so vector-wise | 14:43 |
| joergkurt | iterator or loop 0length ? | 14:43 |
| joergkurt | vector is here a feature, right? (not an instance) | 14:44 |
| @lisitsyn | might be confusing but it is an instance :) | 14:44 |
| @lisitsyn | or example | 14:44 |
| @lisitsyn | joergkurt: you can access whole matrix as well | 14:44 |
| @lisitsyn | get_feature_matrix | 14:44 |
| @lisitsyn | http://shogun-toolbox.org/doc/en/latest/classshogun_1_1CDenseFeatures.html | 14:45 |
| joergkurt | (checking...) | 14:46 |
| joergkurt | vectors=features | 14:48 |
| joergkurt | features=instances | 14:48 |
| joergkurt | right? | 14:48 |
| @lisitsyn | joergkurt: example is feature vector | 14:50 |
| @lisitsyn | its components are features | 14:50 |
| @wiking | morning | 14:58 |
| @lisitsyn | wiking: hmm morning.. are you in us? :D | 15:00 |
| @wiking | heheh no | 15:05 |
| @wiking | actually it's middle of night here | 15:05 |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has quit [Ping timeout: 246 seconds] | 15:06 | |
| @lisitsyn | wiking: looks like you've been having busy times lately? | 15:08 |
| @wiking | yes | 15:09 |
| @wiking | soon over | 15:09 |
| joergkurt | is there an overview about data formats, especially sparse ones, have seen libSVM is supported... | 15:17 |
| @lisitsyn | joergkurt: http://shogun-toolbox.org/doc/en/4.0.0/classshogun_1_1CLibSVMFile.html | 15:23 |
| @lisitsyn | that's for libsvm | 15:23 |
| @lisitsyn | joergkurt: hdf5 and protobuf are also supported | 15:23 |
| -!- iglesiasg [~iglesias@524B8E0B.cm-4-4c.dynamic.ziggo.nl] has joined #shogun | 15:29 | |
| -!- mode/#shogun [+o iglesiasg] by ChanServ | 15:29 | |
| @lisitsyn | iglesiasg: who's there | 15:30 |
| joergkurt | hdf5 support spare and dense right? | 15:30 |
| joergkurt | sparse and dense? | 15:30 |
| @lisitsyn | joergkurt: looks like only dense | 15:31 |
| joergkurt | imports SparseFEatures.h | 15:31 |
| joergkurt | ehm, includes | 15:32 |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has joined #shogun | 15:32 | |
| joergkurt | probably last question for today, if I wanna bind from C++ directly, any algorithm and property parser available, or do we need to hard bind all algorithms and algo properties? | 15:33 |
| @lisitsyn | joergkurt: something like using classifier based on some configuration file? | 15:34 |
| joergkurt | yes, something just defining data+config file | 15:34 |
| @lisitsyn | if yes then the answer is no but sounds like a feature request | 15:34 |
| joergkurt | :) | 15:35 |
| @iglesiasg | lisitsyn: surprise! xD | 15:35 |
| @lisitsyn | iglesiasg: how is it going? | 15:35 |
| @iglesiasg | lisitsyn: very well, with you? | 15:36 |
| joergkurt | kinda factory pattern on C++ side or compiler-compiler for some to be defined algo+property language | 15:36 |
| joergkurt | BTW, where are you folks from? | 15:36 |
| @lisitsyn | joergkurt: yeah I'll think about it and describe as an issue | 15:36 |
| @lisitsyn | joergkurt: different places | 15:36 |
| @lisitsyn | I am from the most wild country in the world | 15:37 |
| joergkurt | me:German:nowBelgium | 15:37 |
| @lisitsyn | joergkurt: you're close to iglesiasg which is now in netherlands | 15:37 |
| @iglesiasg | yep, me:Spanish:inNL :) | 15:38 |
| joergkurt | ah, hi, iglesiasg, hoe gaat het? | 15:38 |
| @iglesiasg | het gaat heel goed, en met jou? | 15:38 |
| joergkurt | ah, hola :) | 15:38 |
| @iglesiasg | hahaha hola, hola | 15:38 |
| joergkurt | hihi :P | 15:39 |
| @iglesiasg | where is Belgium are you? | 15:40 |
| @iglesiasg | in* | 15:40 |
| joergkurt | living in Mechelen (between Brussels and Antwerp), working in Beerse/Turnhout (Janssen Pharma) | 15:40 |
| @lisitsyn | you have bad weather there, do you? | 15:41 |
| joergkurt | never :P | 15:41 |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has quit [Ping timeout: 250 seconds] | 15:42 | |
| @lisitsyn | though I am not a person to speak about bad weather :D | 15:42 |
| joergkurt | lisisyn-whats the wildest country? | 15:42 |
| @lisitsyn | joergkurt: this big bear to the east of you | 15:42 |
| joergkurt | russia? | 15:43 |
| @lisitsyn | yeap | 15:43 |
| joergkurt | ah, cool, some of our developers are from Russi ... | 15:44 |
| joergkurt | Russia | 15:44 |
| @lisitsyn | joergkurt: somehow bad environment produces good math and software guys :) | 15:44 |
| joergkurt | (best dev of the world, honestly!) | 15:44 |
| joergkurt | ok, sold now, we will integrate shogun as soon as we get to it ;) | 15:45 |
| @lisitsyn | haha | 15:45 |
| @lisitsyn | joergkurt: what do you want to apply it to? | 15:45 |
| joergkurt | where are you located (please tell me StPetersburg ;) | 15:45 |
| @lisitsyn | joergkurt: unfortunately moscow :) | 15:46 |
| joergkurt | hihi, ok :P | 15:46 |
| joergkurt | kinda big compute ML stuff, with tendency to multi-class and MKL | 15:46 |
| @lisitsyn | joergkurt: svms? | 15:47 |
| joergkurt | ...possible...its still cubic, and we will see if it can handle our data sizes | 15:47 |
| @lisitsyn | joergkurt: shogun is good but we lack ease of installation and may be better API | 15:48 |
| @lisitsyn | but if you push us enough and tell us what's better be fixed it would be fixed ;) | 15:48 |
| joergkurt | never mind, so far things look good | 15:48 |
| @lisitsyn | joergkurt: one thing is that we have pointer based api | 15:49 |
| @lisitsyn | but that's going to be fixed soon | 15:49 |
| joergkurt | right now we are aiming at writing single train-test instances and just wanna call them via algoXZY with parameter setting XYZ | 15:50 |
| joergkurt | then we fetch the results and estimate profiling (time,space) and scientific performance | 15:50 |
| @lisitsyn | I see | 15:52 |
| -!- exploreshaifali [~shaifali@182.70.196.201] has quit [Quit: Leaving] | 15:52 | |
| joergkurt | ok, will poke around a bit more and sketch some plans ... we will need a while ... ttyl | 15:54 |
| joergkurt | cheers, moving out now ... have a nice weekend | 15:55 |
| -!- joergkurt [~joergkurt@178.162.199.93] has quit [Quit: WeeChat 1.1.1] | 15:58 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 15:58 | |
| -!- prakhar2511 [~prakhar25@202.78.175.199] has joined #shogun | 16:08 | |
| -!- curiousguy13_ [~curiousgu@120.59.226.121] has quit [Ping timeout: 246 seconds] | 16:40 | |
| -!- exploreshaifali [~shaifali@182.70.196.201] has joined #shogun | 16:42 | |
| -!- sanuj [0e8bc402@gateway/web/freenode/ip.14.139.196.2] has joined #shogun | 16:44 | |
| -!- iglesiasg [~iglesias@524B8E0B.cm-4-4c.dynamic.ziggo.nl] has quit [Ping timeout: 246 seconds] | 16:48 | |
| -!- curiousguy13_ [~curiousgu@59.177.203.40] has joined #shogun | 16:53 | |
| -!- smallgraph [~textual@scc-wkit-clx-208-236.scc.kit.edu] has joined #shogun | 16:59 | |
| smallgraph | Hi, I'm trying to build shogun from the develop branch and I get errors on "symbol(s) not found for architecture x86_64" for several methods (first is "shogun::CGaussianARDFITCKernel::CGaussianARDFITCKernel()") when linking libshogun.dylib. Building from master worked fine and I'm working on Yosemite with clang 3.5.0. Did anyone had those problems before? | 17:00 |
| -!- sanuj [0e8bc402@gateway/web/freenode/ip.14.139.196.2] has quit [Ping timeout: 246 seconds] | 17:15 | |
| -!- sanuj [0e8bc402@gateway/web/freenode/ip.14.139.196.2] has joined #shogun | 17:25 | |
| -!- sanuj [0e8bc402@gateway/web/freenode/ip.14.139.196.2] has quit [Ping timeout: 246 seconds] | 17:35 | |
| -!- joergkurt [~joergkurt@2a02:1810:1c19:ea00:b9e8:4dc2:fd9:1e69] has joined #shogun | 17:44 | |
| joergkurt | feature questions: I see numeric, boolean, and String as features, if we want other nominal features they are represented as? | 17:45 |
| -!- prakhar2511 [~prakhar25@202.78.175.199] has quit [Ping timeout: 246 seconds] | 17:51 | |
| joergkurt | rephrasing question, in CDense the types are listed bool, 8bit, etc | 17:59 |
| joergkurt | http://www.shogun-toolbox.org/doc/en/latest/classshogun_1_1CDenseFeatures.html | 18:00 |
| joergkurt | what are the equivalent sparse features being allowed in CSparse | 18:00 |
| joergkurt | http://www.shogun-toolbox.org/doc/en/latest/classshogun_1_1CSparseFeatures.html | 18:00 |
| joergkurt | if I interpret this correctly the template allows largely numeric, and boolean as input | 18:01 |
| joergkurt | this means all integers are in a numeric relationship, right? Nominal classes are not allowed as input space, just as multiclass labels for prediction? | 18:02 |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has quit [Quit: sonne|osx] | 18:12 | |
| -!- smallgraph [~textual@scc-wkit-clx-208-236.scc.kit.edu] has quit [Quit: Textual IRC Client: www.textualapp.com] | 18:15 | |
| -!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has joined #shogun | 18:20 | |
| -!- joergkurt [~joergkurt@2a02:1810:1c19:ea00:b9e8:4dc2:fd9:1e69] has quit [Quit: WeeChat 1.1.1] | 18:31 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 255 seconds] | 18:44 | |
| -!- abinashm [uid66179@gateway/web/irccloud.com/x-hkeeqnqajahrlnip] has joined #shogun | 19:01 | |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has joined #shogun | 19:27 | |
| -!- exploreshaifali [~shaifali@182.70.196.201] has quit [Quit: Leaving] | 19:33 | |
| -!- smallgraph [~textual@p4FDCCD96.dip0.t-ipconnect.de] has joined #shogun | 19:43 | |
| -!- PirosB3 [~pirosb3@host72-217-dynamic.42-79-r.retail.telecomitalia.it] has quit [Quit: PirosB3] | 20:04 | |
| smallgraph | Hi, can someone help me, please? I'm having problems to compile shogun. | 20:05 |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has joined #shogun | 20:14 | |
| -!- rajul [~rajul@node-6gf.pool-180-180.dynamic.totbb.net] has quit [Quit: Leaving] | 20:16 | |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has quit [Client Quit] | 20:19 | |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has joined #shogun | 20:20 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has joined #shogun | 20:34 | |
| -!- kshitijk [~kshitijk@115.249.18.25] has quit [Ping timeout: 246 seconds] | 20:46 | |
| -!- exploreshaifali [~shaifali@182.70.164.213] has joined #shogun | 20:52 | |
| -!- HeikoS [~heiko@2.221.116.151] has joined #shogun | 21:03 | |
| -!- mode/#shogun [+o HeikoS] by ChanServ | 21:04 | |
| -!- pickle27 [~pickle27@192-0-134-235.cpe.teksavvy.com] has quit [Remote host closed the connection] | 21:14 | |
| -!- exploreshaifali [~shaifali@182.70.164.213] has quit [Quit: Leaving] | 21:19 | |
| -!- f0z [67157d4e@gateway/web/freenode/ip.103.21.125.78] has joined #shogun | 21:25 | |
| f0z | hey | 21:26 |
| f0z | Can anybody guide me through how debugging is done in shogun? | 21:26 |
| f0z | Any...body? | 21:28 |
| f0z | @lisitsyn Do you mind? | 21:31 |
| @lisitsyn | f0z: hey | 21:32 |
| f0z | hey | 21:32 |
| @lisitsyn | f0z: well, we mostly use gdb and valgrind | 21:32 |
| @lisitsyn | what do you want to debug? | 21:32 |
| f0z | No no | 21:32 |
| f0z | I am not talking about that | 21:32 |
| f0z | I am talking about SG_DEBUG flags and the likes | 21:32 |
| @lisitsyn | sg_debug is basically a printf | 21:32 |
| @lisitsyn | that shows something is the level is set to debug | 21:33 |
| -!- abinashm [uid66179@gateway/web/irccloud.com/x-hkeeqnqajahrlnip] has quit [Quit: Connection closed for inactivity] | 21:33 | |
| @lisitsyn | if you are asking how to see sg_debug messages then the answer is | 21:33 |
| @lisitsyn | just get global io | 21:33 |
| @lisitsyn | anyshogunobject->io | 21:33 |
| @lisitsyn | and call set_loglevel | 21:33 |
| @lisitsyn | anyshogunobject->io->set_loglevel(MSG_DEBUG) | 21:33 |
| f0z | what's the default state? | 21:34 |
| @lisitsyn | iirc it is msg_info | 21:34 |
| @lisitsyn | so only SG_INFO are shown | 21:35 |
| f0z | so this isssue | 21:35 |
| f0z | https://github.com/shogun-toolbox/shogun/issues/2723 | 21:35 |
| f0z | I get how to compute the avg. gradient and all | 21:36 |
| f0z | i mean absolute value | 21:36 |
| f0z | but do I have to put explicitly that info somewhere using these flags? | 21:36 |
| @lisitsyn | hmm not sure I get the question | 21:37 |
| @lisitsyn | but currently SG_INFO is used here | 21:37 |
| @lisitsyn | so the gradient stuff should be output as info as well | 21:37 |
| f0z | by SG_DEBUG or SG_INFO? the latter one right? | 21:38 |
| @lisitsyn | SG_INFO yeah | 21:38 |
| @lisitsyn | it is not debug info I think as watching at gradients is normal for neural networks training | 21:39 |
| -!- prakhar2511 [~prakhar25@202.78.172.162] has quit [Ping timeout: 246 seconds] | 21:39 | |
| f0z | thanks@@ | 21:40 |
| f0z | i meant thanks!!! | 21:40 |
| @lisitsyn | you're welcome :) | 21:40 |
| smallgraph | yeah, if I can't see those numbers change I nearly die because of impatience^^ | 21:40 |
| smallgraph | lisitsyn, do you also have time for my question? | 21:41 |
| @lisitsyn | smallgraph: sure | 21:42 |
| smallgraph | I'm trying to build from develop branch but get this error: | 21:42 |
| smallgraph | Linking CXX shared library libshogun.dylib | 21:42 |
| smallgraph | Undefined symbols for architecture x86_64: | 21:42 |
| smallgraph | "shogun::CGaussianARDFITCKernel::get_parameter_gradient(shogun::TParameter const*, int)", referenced from: | 21:42 |
| smallgraph | vtable for shogun::CGaussianARDFITCKernel in GaussianARDFITCKernel.cpp.o | 21:43 |
| @lisitsyn | smallgraph: oh | 21:43 |
| @lisitsyn | should be easy to fix | 21:43 |
| @lisitsyn | smallgraph: let me update and reproduce | 21:43 |
| smallgraph | well the method is there, so I don't see why the compiler does not generate it | 21:44 |
| @lisitsyn | smallgraph: could be due to no eigen available | 21:44 |
| @lisitsyn | do you have eigen3? | 21:45 |
| smallgraph | ok, no, not yet^^ | 21:46 |
| smallgraph | Oh, I did install it, but cmake did not find it^^ | 21:47 |
| smallgraph | let me check why | 21:47 |
| @lisitsyn | smallgraph: could be the reason | 21:47 |
| -!- HeikoS [~heiko@2.221.116.151] has quit [Quit: Leaving.] | 21:57 | |
| -!- f0z [67157d4e@gateway/web/freenode/ip.103.21.125.78] has quit [Ping timeout: 246 seconds] | 21:59 | |
| smallgraph | ok, now it finds it and compiles but it still dosen't find OpenCL | 22:05 |
| smallgraph | thanks :) | 22:07 |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has quit [Quit: sonne|osx] | 22:18 | |
| -!- curiousguy13_ [~curiousgu@59.177.203.40] has quit [Ping timeout: 264 seconds] | 22:40 | |
| -!- smallgraph [~textual@p4FDCCD96.dip0.t-ipconnect.de] has quit [Quit: My Mac has gone to sleep. ZZZzzz...] | 22:49 | |
| -!- curiousguy13_ [~curiousgu@120.59.227.130] has joined #shogun | 22:54 | |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has joined #shogun | 23:12 | |
| -!- vortex_ape [~vortex_ap@120.59.74.178] has joined #shogun | 23:13 | |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has quit [Quit: sonne|osx] | 23:44 | |
| -!- sonne|osx [~sonne@f049254171.adsl.alicedsl.de] has joined #shogun | 23:50 | |
| -!- PirosB3 [~pirosb3@host72-217-dynamic.42-79-r.retail.telecomitalia.it] has joined #shogun | 23:52 | |
| -!- xAndreLopesx [ThrashIRC@179.34.60.135] has joined #shogun | 23:59 | |
| -!- xAndreLopesx [ThrashIRC@179.34.60.135] has left #shogun [] | 23:59 | |
| --- Log closed Mon Mar 02 00:00:41 2015 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!