--- Log opened Fri May 24 00:00:15 2013 | ||
@sonney2k | lisitsyn, you are close - just one test fails now converter_tdistributedstochasticneighborembedding_modular | 00:09 |
---|---|---|
lisitsyn | sonney2k: hmm | 00:10 |
gsomix | good night, guys | 00:11 |
@sonney2k | night everyone | 00:28 |
-!- gsomix [~Miranda@83.234.54.102] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 00:41 | |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has quit [Quit: Leaving] | 01:09 | |
-!- lisitsyn [~blackburn@83.234.54.102] has quit [Ping timeout: 256 seconds] | 01:15 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 01:58 | |
-!- foulwall [~foulwall@2001:da8:215:503:218b:289d:d11c:3fac] has joined #shogun | 03:41 | |
shogun-buildbot | build #404 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/404 | 04:10 |
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has joined #shogun | 04:58 | |
-!- sumit_ [75dc1241@gateway/web/freenode/ip.117.220.18.65] has joined #shogun | 06:27 | |
-!- nube [~rho@116.90.239.3] has joined #shogun | 06:34 | |
-!- sumit_ [75dc1241@gateway/web/freenode/ip.117.220.18.65] has quit [Ping timeout: 250 seconds] | 06:53 | |
-!- abinashpanda [7ab11c92@gateway/web/freenode/ip.122.177.28.146] has joined #shogun | 07:39 | |
-!- nube [~rho@116.90.239.3] has quit [Read error: Connection reset by peer] | 07:57 | |
-!- nube1 [~rho@116.90.239.3] has joined #shogun | 07:57 | |
-!- lisitsyn [~blackburn@37.61.181.21] has joined #shogun | 07:57 | |
-!- foulwall [~foulwall@2001:da8:215:503:218b:289d:d11c:3fac] has quit [Remote host closed the connection] | 08:07 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:49f1:bed:e381:e0b2] has joined #shogun | 08:10 | |
-!- lisitsyn [~blackburn@37.61.181.21] has quit [Quit: Leaving.] | 08:26 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 08:29 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 571ea98 / tests/integration/python_modular/generator.py: https://github.com/shogun-toolbox/shogun/commit/571ea9864dbb19e3f4b2de929739eeb6b3ce47fb | 08:29 |
shogun-notifier- | shogun: disable failing tests | 08:29 |
shogun-notifier- | shogun: | 08:29 |
shogun-notifier- | shogun: - converter test fails for real | 08:29 |
shogun-notifier- | shogun: - disable other tests failing only on travis | 08:29 |
shogun-buildbot | build #869 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/869 | 08:52 |
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has left #shogun [] | 08:53 | |
-!- sijin [~smuxi@144.214.222.109] has joined #shogun | 08:58 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:06 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 09:06 | |
shogun-buildbot | build #1162 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1162 | 09:09 |
-!- travis-ci [~travis-ci@ec2-54-224-241-7.compute-1.amazonaws.com] has joined #shogun | 09:11 | |
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/7452306 | 09:11 |
-!- travis-ci [~travis-ci@ec2-54-224-241-7.compute-1.amazonaws.com] has left #shogun [] | 09:11 | |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-duyxqpfymnsbvjzy] has joined #shogun | 09:50 | |
lambday | Guten morgen, Доброе утро. Buenos días. | 09:52 |
@lisitsyn | lambday: haha | 09:58 |
@lisitsyn | going multilingual | 09:58 |
lambday | :) | 09:58 |
@lisitsyn | lambday: morgen, wie gehts? | 09:58 |
@lisitsyn | lambday: ich spreche etwas deutsch | 09:58 |
lambday | shit.. (urgent help needed @google translator) | 09:59 |
@lisitsyn | hahaha | 09:59 |
lambday | Hablo un poco de español. | 10:00 |
lambday | HeikoS: hi.. | 10:05 |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-duyxqpfymnsbvjzy] has quit [Quit: lambday] | 10:13 | |
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has joined #shogun | 10:18 | |
sonne|work | shogun-buildbot: force build nightly_default | 10:19 |
shogun-buildbot | build forced [ETA 50m45s] | 10:19 |
shogun-buildbot | I'll give a shout when the build finishes | 10:19 |
@lisitsyn | sonne|work: I will have one hour buffer before that meeting so I can make it for sure | 10:27 |
sonne|work | lisitsyn: sry what? | 10:28 |
@lisitsyn | sonne|work: #gsoc | 10:28 |
sonne|work | ahh oki | 10:28 |
@lisitsyn | I had a chance I get to that meeting just in time because of being in bus | 10:29 |
sonne|work | I might not need to do 'fix my pc' support tonight so I should be around too | 10:29 |
@lisitsyn | ahh okay | 10:29 |
sonne|work | btw 7 am UTC was the deadline right? | 10:29 |
@lisitsyn | yes | 10:29 |
sonne|work | did you check for last minute conflicts? | 10:29 |
@lisitsyn | we have passed it I guess | 10:29 |
@lisitsyn | sonne|work: I see no conflicts | 10:30 |
@lisitsyn | sonne|work: everything seemed to be correct yes | 10:30 |
sonne|work | smooth :) | 10:31 |
sonne|work | alright then it should all be no problem tonight | 10:31 |
-!- abinashpanda [7ab11c92@gateway/web/freenode/ip.122.177.28.146] has quit [Ping timeout: 250 seconds] | 10:39 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 10:45 | |
vgorbati | sonney2k: hi, are you here? | 10:59 |
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has quit [Ping timeout: 248 seconds] | 11:05 | |
shogun-buildbot | build #405 of nightly_default is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/405 | 11:14 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 11:30 | |
@wiking | sonne|work: did u merge the prng improvement branch? | 11:49 |
@wiking | oh yeah i see u did | 11:49 |
-!- nube1 [~rho@116.90.239.3] has quit [Quit: Leaving.] | 12:09 | |
sonne|work | wiking: yeah yeah | 12:09 |
@wiking | great :) | 12:29 |
foulwall_ | Hi sonne|work | 12:35 |
-!- foulwall_ [~foulwall@2001:da8:215:503:49f1:bed:e381:e0b2] has quit [Remote host closed the connection] | 12:36 | |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 12:44 | |
-!- foulwall [~foulwall@2001:da8:215:6901:5eb:a14a:3610:fc59] has joined #shogun | 12:49 | |
foulwall | did you see my yesterday's work? | 12:50 |
-!- sumit_ [75d69e3e@gateway/web/freenode/ip.117.214.158.62] has joined #shogun | 13:01 | |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-jlfnpdgkvoahjmxy] has joined #shogun | 13:05 | |
sonne|work | foulwall: yes I tried it even :) | 13:05 |
sonne|work | foulwall: I think I will create a couple of reasonable datasets and put them in some repository - maybe even not shogun-data (slightly big already...) | 13:06 |
foulwall | sonne|work: any suggests? | 13:06 |
foulwall | sonne|work: Ok | 13:06 |
sonne|work | foulwall: no just assume that you can load all *.h5 in some given dir and present them for loading | 13:07 |
sonne|work | foulwall: maybe we even have a few data sets per category - e.g. some for binary classification, multiclass, regression | 13:08 |
foulwall | sonne|work: hmm | 13:08 |
-!- sumit_ [75d69e3e@gateway/web/freenode/ip.117.214.158.62] has quit [Ping timeout: 250 seconds] | 13:08 | |
sonne|work | so you would then only show the ones from that respective dir | 13:08 |
foulwall | sonne|work: ok , gotcha | 13:08 |
foulwall | sonne|work: I'll do some fix this night ~~~ | 13:10 |
-!- foulwall [~foulwall@2001:da8:215:6901:5eb:a14a:3610:fc59] has quit [Remote host closed the connection] | 13:11 | |
@wiking | lisitsyn: ping | 13:12 |
@lisitsyn | wiking: pong | 13:13 |
@wiking | lisitsyn: openmp implementation...? :) | 13:13 |
@wiking | lisitsyn: i can haz openmp with clang finally | 13:13 |
lambday | HeikoS: hi | 13:14 |
@HeikoS | lambday: hi! | 13:14 |
@wiking | vgorbati: hey you've /msg -ed me about prng | 13:14 |
@wiking | vgorbati: shoot what you need if you still need it | 13:14 |
lambday | HeikoS: the initial class structure.. seems okay to you? | 13:14 |
@HeikoS | lambday: yes | 13:14 |
@HeikoS | so one thing I would maybe change: | 13:15 |
@HeikoS | sampler as a base class | 13:15 |
lambday | HeikoS: hmm... | 13:15 |
@lisitsyn | wiking: openmp of? | 13:15 |
@HeikoS | actually I would do this everywhere where there might be different implementations of things | 13:15 |
lambday | okay.. not the enum there... | 13:15 |
@HeikoS | then you call abstract methods and impleent them in subclasses | 13:15 |
@HeikoS | only enum if you definitely know there are just a small number of cases and not more | 13:16 |
@wiking | lisitsyn: i'm writing now bagging machien and it would be nice that i could do the training of the n machines in parallel | 13:16 |
vgorbati | wiking: yes. I did work on one integration test for random DR method (setting seed to some specified value on the beginning), and on my Mac OS it gave one result, while on build servers (travis and shogun-buildbot) the results were different | 13:16 |
lambday | HeikoS: alright... so for matrix function things, enum is fine, right? | 13:16 |
@HeikoS | otherwise you will have to do all this switch(type) stuff which makes code hard to read | 13:16 |
lambday | cause that method works for only sqrt, log and A^alpha | 13:16 |
@lisitsyn | wiking: so openmp is in clang? | 13:16 |
@HeikoS | lambday: exactly, we will only have a few there so thats fine | 13:16 |
@wiking | vgorbati: that is rather fucking intersting :) | 13:16 |
@HeikoS | let me check the rest of the mail ... | 13:16 |
lambday | HeikoS: nice idea... okay :) | 13:16 |
vgorbati | wiking: are you using linux? | 13:16 |
@HeikoS | lambday: sorry for not responding too fast, I have a deadline coming btw ... | 13:16 |
@wiking | vgorbati: osx | 13:17 |
@HeikoS | lambday: could you maybe create a diagram for these things? | 13:17 |
@HeikoS | just with inheritance and who references who | 13:17 |
lambday | I've declared the enum in CContourIntegral.h instead, to avoid including unnecessary headers, in case one just want to use ContourIntegral and not RationalApproximation | 13:17 |
@wiking | vgorbati: can u plz tell me where exactly you had the problem | 13:17 |
lambday | HeikoS: sure... I'll mail it, changing the sampler as the base class | 13:17 |
@wiking | vgorbati: the exact reason to have our own PRNG is that we dont have this problem anymore | 13:18 |
@HeikoS | lambday: and maybe a small description for every class, like one sentence | 13:18 |
-!- gsomix [~Miranda@r206-10.smr.ru] has joined #shogun | 13:18 | |
gsomix | hello | 13:18 |
@HeikoS | and method | 13:18 |
@wiking | vgorbati: i.e. the seeded random is always the same, regardless of OS and arch | 13:18 |
@HeikoS | so what does register do for example | 13:18 |
lambday | HeikoS: alright... | 13:18 |
vgorbati | wiking: ok, can you please try running "python generator.py converter_tdistributedstochasticneighborembedding_modular.py" in the "tests/integration/python_modular" and paste the file "data/testsuite/tests/converter_tdistributedstochasticneighborembedding_modular0.txt" somewhere? | 13:18 |
@HeikoS | CComputation for example will also be a base class | 13:18 |
@HeikoS | which is just used for registering problems (maybe make problem a class also) | 13:19 |
@wiking | vgorbati: will do in a second | 13:19 |
@HeikoS | only the implementation knows that there are shifts in the systems | 13:19 |
vgorbati | wiking: you will need to un-blacklist the file 'converter_tdistributedstochasticneighborembedding_modular.py' in generator.py | 13:19 |
@HeikoS | try to think as general as possible | 13:19 |
@HeikoS | once we have the diagram, we can start moving things around (maybe attach the dia file so that I can modify it) | 13:19 |
vgorbati | wiking: it is actually blacklisted 2 times :) | 13:19 |
@HeikoS | lambday, we should spend some time on this, even before any code is written | 13:20 |
lambday | HeikoS: alright... I'll send you by tonight... | 13:20 |
@HeikoS | and then we should start adding concrete implementations in the diagram | 13:20 |
@HeikoS | also before writing code | 13:20 |
@HeikoS | also, keep in mind that everything has to be defined in terms of linear operators | 13:21 |
@HeikoS | so when computing the matrix logarithm for example | 13:21 |
@HeikoS | we always compute log(M)*v | 13:21 |
@HeikoS | never log(M) alone (not possible sometimes) | 13:21 |
lambday | yes... | 13:21 |
lambday | hmm.. | 13:21 |
@HeikoS | even for log-dets | 13:21 |
@HeikoS | sometimes one cannot even store the matrix one wants to compute the log-det of | 13:21 |
@HeikoS | but one can compute matrix-vector products of the parts for example | 13:22 |
@HeikoS | so always linear operator apply, never a matrix on its own | 13:22 |
lambday | hmmm... | 13:22 |
sonne|work | wiking: if you want to use multiple threads for bagging | 13:24 |
sonne|work | you should assign new SGParallel objects to the internal machine | 13:24 |
sonne|work | not that it uses multiple threads | 13:24 |
sonne|work | with nthreads = 1 | 13:25 |
lambday | HeikoS: since computing shifts weights etc don't depend on the rest of the things, that things can be done now, may be? | 13:27 |
@HeikoS | lambday: yes, if you want to code, then this can be done now :) | 13:27 |
lambday | HeikoS: I hope I'll finish the coding for ContourIntegral thing by tonight/tomorrow.. and spend some time on designing the whole thing... | 13:28 |
@HeikoS | lambday: okay, | 13:29 |
gsomix | sonne|work: hi, working? :) | 13:29 |
@HeikoS | lambday: again, I am slowing down things for the class structure to make it as flexible as possible | 13:29 |
lambday | HeikoS: yes.. that's the best idea.. :) | 13:29 |
sonne|work | gsomix: sure :D | 13:30 |
gsomix | sonne|work: me too, sitting in the Lab. | 13:30 |
lambday | HeikoS: could you please tell me what you have in mind about base CComputation class? What I thought was CSolver as a base class and then its subclasses implementing different solvers | 13:31 |
@HeikoS | Computation and solver are different | 13:31 |
@HeikoS | computation class has a register method where one can register subproblems (i.e. one linear system) in some structured way (i.e. all linear systems with indices) | 13:32 |
@HeikoS | the computation class then has an abstract method compute which then solves all registered problems somehow | 13:32 |
@HeikoS | note that registering might differ in subclasses (i.e. register a shifts vector) | 13:32 |
@HeikoS | the solver class just solves a linear system | 13:33 |
@HeikoS | and might be used by the computation class impleentations | 13:33 |
@HeikoS | does that make sense? | 13:33 |
lambday | HeikoS: yes... okay | 13:34 |
lambday | HeikoS: alright... this helped a lot... I'll mail you the diagram... | 13:37 |
@HeikoS | good :) | 13:37 |
thoralf | HeikoS, sonne|work: You wrote lib/DynamicObjectArray.h, right? | 13:42 |
thoralf | Can you explain, why you always do SG_REF before returning a reference, e.g. in get_element()? | 13:42 |
@HeikoS | thoralf: this is done everywhere in shohun | 13:43 |
@HeikoS | shogun | 13:43 |
@HeikoS | whenever you get an element, you store it, so the ref-counter has to be incremented | 13:43 |
-!- sumit [75cf9448@gateway/web/freenode/ip.117.207.148.72] has joined #shogun | 13:44 | |
sonne|work | thoralf: in python & swig you need a way to tell whether an object is still used | 13:44 |
thoralf | HeikoS: Okay, that means I can rely that a returned reference has been SG_REF and I need to care about SG_UNREF? | 13:44 |
sonne|work | thoralf: this is done via reference counts | 13:44 |
sonne|work | so whenever we return an object | 13:45 |
sonne|work | python & swig calls ref() | 13:45 |
@HeikoS | thoralf, yes you have to unref to prevent leaks | 13:45 |
sonne|work | when swig is done using the object | 13:45 |
sonne|work | it calls unref() | 13:46 |
@HeikoS | thoralf: this is done most of the time, sometimes, we forgot SG_REF, in this case, send a bug report :) | 13:46 |
sonne|work | once the refcount (via unref reaches 0) | 13:46 |
@wiking | sonne|work: will check | 13:46 |
thoralf | sonne|work: Maybe I don't store the reference, but only pass then to a method which cares about ref/unref by itself. | 13:46 |
thoralf | sonne|work: So I always need to store the reference in a variable and UNREF it afterwards. | 13:46 |
thoralf | sonne|work: Yeah, I know. I just wasn't aware of this kind of "convention". :) | 13:48 |
sonne|work | thoralf: yes | 13:48 |
sonne|work | this also works the other way round | 13:48 |
sonne|work | if you create say some feature object | 13:49 |
sonne|work | you can pass this directly into some machine | 13:49 |
sonne|work | then an unref of the machine will also erase the feature object | 13:49 |
thoralf | sonne|work: So I can't do calls like method_eats_ref( object->method_returns_ref() ) | 13:49 |
thoralf | sonne|work: I always need addional code just for ref-counting. :) | 13:50 |
thoralf | sonne|work: One example: kernel->set_features( ref_to_my_features ); | 13:51 |
thoralf | sonne|work: Do I need to care about SG_UNREF of the passed object or not? | 13:51 |
@wiking | sonne|work: for feature/label selection with replacement, can i use the subset thingy? | 13:51 |
thoralf | sonne|work: I always do SG_REF() after creation and SG_UNREF() in the end - but after what you told, this is not neccessary? Just pass into ->set_features() and stop caring? | 13:53 |
sonne|work | thoralf: for kernel->set_features( ref_to_my_features ); you don't need to SG_REF(ref_to_my_features) after you created them and if you don't touch them later | 13:57 |
sonne|work | so e.g. kernel->set_features(new DenseFeatures<float64_t>()) | 13:58 |
sonne|work | but for your method_eats_ref( object->method_returns_ref() ) you need to store the object and unref | 13:59 |
thoralf | sonne|work: Thanks, that answers my questions. | 14:00 |
sonne|work | wiking: I think not - hmmh HeikoS - maybe we have conflicts then right? | 14:00 |
sonne|work | thoralf: btw we cannot return lazy copies (not SG_REF'd) | 14:01 |
sonne|work | or swig interfaces will have memory leaks | 14:01 |
thoralf | sonne|work: Thats alright. I just assumed, you only count references when copying objects and I have to care about references by myself. | 14:05 |
sonne|work | wiking: actually you can do it! | 14:07 |
sonne|work | wiking: just do add_subset | 14:07 |
sonne|work | and once done remove_subset | 14:07 |
thoralf | sonne|work: Btw, I think it would make sense to initialize "new SGVector<T>(len)" with zeros in the constructor, i.e. zero(); right after the SG_MALLOC. | 14:09 |
@wiking | sonne|work: mmm but can i have replacement there? | 14:10 |
sonne|work | thoralf: we could use SG_CALLOC for that but it is not done to have a speed gain in case you just want to alloc some vector and assign values... | 14:10 |
thoralf | sonne|work: valgrind told me something about uninitialized variables - and I think the fix would be trivial | 14:10 |
sonne|work | wiking: replacement? | 14:10 |
@wiking | sonne|work: or the index vector can be whatever, i.e. i can have in index vector the same index as many times as i want | 14:10 |
@wiking | sonne|work: "each of size n′ < n, by sampling from D uniformly and with replacement" | 14:11 |
sonne|work | wiking: sure why not! | 14:11 |
sonne|work | just the same size! | 14:11 |
@wiking | ok cool | 14:11 |
@wiking | yeah sure ;) | 14:11 |
sonne|work | no actually any n' <= n yes | 14:11 |
sonne|work | wiking: if you do that better extract the sampling strategies | 14:12 |
sonne|work | like random subset / bagging etc | 14:12 |
@wiking | ? | 14:12 |
sonne|work | no idea what the official name for sampling strategy is | 14:12 |
sonne|work | maybe heiko knows | 14:12 |
sonne|work | HeikoS: ^? | 14:12 |
@wiking | well i'll just do random selection | 14:12 |
sonne|work | wiking: well there are various strategies for sampling | 14:13 |
@HeikoS | ehm what? | 14:13 |
@wiking | mmm | 14:13 |
@HeikoS | sampling what? | 14:13 |
@wiking | sonne|work: really? | 14:13 |
@wiking | sonne|work: here it said only that uniform random with replacement | 14:13 |
@wiking | HeikoS: bagging | 14:13 |
sonne|work | HeikoS: wiking is building some classifier that is trained a couple of times on subsets of the data | 14:13 |
@HeikoS | random forests? | 14:13 |
sonne|work | the way this subset is generated should be generalized | 14:13 |
sonne|work | yes for that | 14:14 |
sonne|work | but it will work for anything | 14:14 |
@HeikoS | well then the samples have to be independent | 14:14 |
sonne|work | and there are various strategies for sampling | 14:14 |
@HeikoS | so with repetitions | 14:14 |
sonne|work | like uniform / stratified / repetitions etc | 14:14 |
sonne|work | so we might want to generalize this somehow | 14:14 |
@wiking | now the only question is that whether all our CMachine implementations has clone() function implemented | 14:19 |
sonne|work | wiking: surely not :D | 14:23 |
sonne|work | wiking: but what do you need clone for? | 14:23 |
sonne|work | ohh I see | 14:24 |
@wiking | sonne|work: how else i'm going to create a bag of defined machines | 14:24 |
thoralf | What's the difference between left_features and right_features in kernel->init(left, right)? | 14:32 |
-!- nube [~rho@49.244.98.80] has joined #shogun | 14:36 | |
sonne|work | thoralf: well rows, columns | 14:39 |
sonne|work | thoralf: so for training you do k(X,X) and for testing k(X,X_test) | 14:39 |
thoralf | sonne|work: Thanks. | 14:39 |
thoralf | sonne|work: That explains why the cache is only initialized for lhs. | 14:40 |
thoralf | (As far as i know) | 14:40 |
lambday | quit | 14:49 |
-!- lambday [e8b6149@gateway/web/cgi-irc/kiwiirc.com/x-jlfnpdgkvoahjmxy] has quit [Quit: lambday] | 14:49 | |
sonne|work | thoralf: it is only inited for svmlight IIRC and only when training | 14:50 |
-!- foulwall [~foulwall@2001:da8:215:6901:6804:84d2:5c2c:d726] has joined #shogun | 14:51 | |
@wiking | sonne|work: any other ideas for creating a bunch of *same* machine? :) | 14:52 |
-!- foulwall [~foulwall@2001:da8:215:6901:6804:84d2:5c2c:d726] has quit [Remote host closed the connection] | 14:52 | |
@wiking | sonne|work: and i really wonder how this should work in case of a kernel machine? : | 14:53 |
@wiking | :P | 14:53 |
sonne|work | wiking: ohh I realize parameters of the machine need to be set already | 14:54 |
sonne|work | I mean fully | 14:54 |
sonne|work | wiking: so you need in case of a kernelmachine, the kernel, its parameters and the kernel machine and its parameters already set | 14:55 |
sonne|work | then take that machine and train with different subsets | 14:55 |
sonne|work | wiking: for now just do it this way but not parallel at all | 14:56 |
-!- sumit [75cf9448@gateway/web/freenode/ip.117.207.148.72] has quit [Quit: Page closed] | 15:10 | |
-!- foulwall [~foulwall@2001:da8:215:6901:19a3:8a30:a2c8:3045] has joined #shogun | 15:11 | |
@HeikoS | sonne|work, wiking clone should be done via parameter framework, please dont start writing custom clone methods for all classes, this wil be a mess | 15:11 |
sonne|work | HeikoS, wiking - well wikign doesn't need clone as long as he runs everything *single* threaded | 15:12 |
@HeikoS | okey | 15:12 |
@HeikoS | I see | 15:12 |
@HeikoS | similar to grid-search stuff | 15:12 |
sonne|work | yeah | 15:13 |
gsomix | <offtop>https://dl.dropboxusercontent.com/u/19029407/results_of_raytracing.png<\offtop> | 15:20 |
gsomix | ahh, braces... :) | 15:21 |
@wiking | sonne|work: ok but how do i save the different models? | 15:44 |
@wiking | sonne|work: because later i'll have to apply the features on those different models | 15:44 |
sonne|work | wiking: ohh indeed | 15:46 |
sonne|work | HeikoS: any suggestions? I mean this certainly needs store_model_features | 15:47 |
@HeikoS | what about serialising? | 15:48 |
@HeikoS | to a byte stream | 15:48 |
@HeikoS | ah no | 15:48 |
@HeikoS | sorry | 15:48 |
@HeikoS | so multple subsets on the same machine | 15:48 |
@HeikoS | thats not trivial | 15:49 |
@HeikoS | store_model_features has to be extended | 15:49 |
@wiking | sonne|work: btw: random forest is not a simple bagging machine. as it splits the data set vertically as well not only horizontally | 15:51 |
@wiking | i.e. bagging just splits the dataset into smaller bags... but random forest splits the data set by feature dim | 15:52 |
sonne|work | wiking: both at once? | 15:53 |
@wiking | each tree gets n < m features where m is the feature dim of the original dataset | 15:53 |
@wiking | "Each tree is trained on a bootstrapped sample of the original data set." | 15:54 |
@wiking | "Each time a leaf is split, only a randomly chosen subset of the dimensions are considered for splitting." | 15:54 |
@wiking | anyhow for bagging we need to be able to clone machines or something like that | 15:54 |
-!- foulwall [~foulwall@2001:da8:215:6901:19a3:8a30:a2c8:3045] has quit [Remote host closed the connection] | 15:58 | |
@lisitsyn | yes RF does both | 16:02 |
@wiking | anybody expert in VS? | 16:03 |
@wiking | i would need to link a static library into a dll | 16:03 |
@wiking | how the hell do i do it? | 16:04 |
@lisitsyn | wiking: visual studio? | 16:04 |
-!- gsomix [~Miranda@r206-10.smr.ru] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 16:06 | |
@wiking | lisitsyn: yes | 16:07 |
@wiking | lisitsyn: any ideas? | 16:09 |
@lisitsyn | wiking: no not really | 16:12 |
@wiking | :S | 16:12 |
@wiking | yeah i'm having trouble to figure out this shit as well | 16:12 |
@lisitsyn | wiking: I did something about it while ago | 16:12 |
@lisitsyn | but can't remember any way out | 16:12 |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has joined #shogun | 16:18 | |
* wiking hates vs | 16:26 | |
@wiking | i cannot believe that doing this can be so fucking complicated... | 16:26 |
@wiking | ok i'll use cmake now ;) | 16:37 |
-!- foulwall [~foulwall@2001:da8:215:c252:45e7:800e:3e38:f907] has joined #shogun | 16:41 | |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Read error: Connection reset by peer] | 16:47 | |
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has quit [Quit: Leaving.] | 16:50 | |
@sonney2k | wiking, life is tough | 17:07 |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has joined #shogun | 17:32 | |
pickle27 | sonney2k: working on QDA now | 17:50 |
pickle27 | sonney2k: going through and making it consistent with MCLDA and updating the styling | 17:51 |
pickle27 | sonney2k: Eigen3 can also be used in more places that in currently is | 17:53 |
-!- thoralf [~thoralf@enki.zib.de] has quit [Quit: Konversation terminated!] | 18:00 | |
-!- zxtx [~zv@host-216-220-114-135.dsl.bway.net] has joined #shogun | 18:08 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 18:23 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 18:43 | |
shogun-notifier- | shogun: Roman Votyakov :develop * fdac479 / src/shogun/regression/gp/ (4 files): https://github.com/shogun-toolbox/shogun/commit/fdac479aac844e368aa977caf0be7b87e5568230 | 18:43 |
shogun-notifier- | shogun: Removed unused fields from inference method classes | 18:43 |
shogun-notifier- | shogun: Heiko Strathmann :develop * ffa6cd0 / src/shogun/regression/gp/ (4 files): https://github.com/shogun-toolbox/shogun/commit/ffa6cd015103ad6003f8758887788be996dbbda6 | 18:43 |
shogun-notifier- | shogun: Merge pull request #1128 from votjakovr/develop | 18:43 |
shogun-notifier- | shogun: | 18:43 |
shogun-notifier- | shogun: Removed unused fields from inference method classes | 18:43 |
votjakovr | HeikoS: Very fast merge, thanks :) | 18:47 |
@HeikoS | welcome :) | 18:47 |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-oldqzgtrfytyuyka] has quit [Ping timeout: 264 seconds] | 18:52 | |
pickle27 | sonney2k: sent the qda pr, let me know if its all good | 18:57 |
pickle27 | sonney2k: there is still room for better use of Eigen3 there which I may play with in the next bit but theres some things I need to wrap up for school atm | 18:57 |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has quit [Quit: Leaving] | 18:57 | |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has joined #shogun | 19:03 | |
-!- foulwall [~foulwall@2001:da8:215:c252:45e7:800e:3e38:f907] has quit [Remote host closed the connection] | 19:06 | |
-!- zxtx [~zv@host-216-220-114-135.dsl.bway.net] has quit [Ping timeout: 256 seconds] | 19:09 | |
@sonney2k | pickle27, yeah one can *always* improve | 19:21 |
pickle27 | sonney2k: very true! | 19:21 |
@sonney2k | pickle27, could you please remove all the SG_PRINT functions still? | 19:23 |
@sonney2k | or is this covered in debug macros? | 19:23 |
pickle27 | sonney2k: sure, why do we not want the print statements? | 19:23 |
@sonney2k | ohh it is | 19:23 |
pickle27 | sonney2k: yeah the alg doesn't print unless the debug flag is set | 19:24 |
@sonney2k | pickle27, it is annoying for people using it | 19:24 |
@sonney2k | so all good! | 19:24 |
@sonney2k | lets merge | 19:24 |
pickle27 | sonney2k: let the fun begin ... jk should be smooth | 19:24 |
shogun-notifier- | shogun: Kevin Hughes :develop * 0c4934b / / (6 files): https://github.com/shogun-toolbox/shogun/commit/0c4934b5cd89e9ecd896b57b425b469149ca05c5 | 19:25 |
shogun-notifier- | shogun: fixed QDA with Eigen3 for merge, also added a unit test for QDA and fixed a bad tab in the MCLDA code | 19:25 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 6255686 / / (6 files): https://github.com/shogun-toolbox/shogun/commit/6255686e857d4d044e6c241168f45914e89cf3ae | 19:25 |
shogun-notifier- | shogun: Merge pull request #1129 from pickle27/qda | 19:25 |
shogun-notifier- | shogun: | 19:25 |
shogun-notifier- | shogun: QDA with Eigen3 for merge thanks monalisag & ozanser for their work! | 19:25 |
-!- travis-ci [~travis-ci@ec2-54-235-58-237.compute-1.amazonaws.com] has joined #shogun | 19:36 | |
travis-ci | [travis-ci] it's Heiko Strathmann'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/7468608 | 19:36 |
-!- travis-ci [~travis-ci@ec2-54-235-58-237.compute-1.amazonaws.com] has left #shogun [] | 19:36 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 19:41 | |
shogun-buildbot | build #1164 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/1164 blamelist: Roman Votyakov <votjakovr@gmail.com> | 19:48 |
shogun-buildbot | build #1165 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/1165 blamelist: Soeren Sonnenburg <sonne@debian.org>, Kevin Hughes <kevin@kevin-M3920.(none)> | 20:15 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Leaving] | 20:25 | |
lambday | HeikoS: hi... | 20:44 |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-myercfhnlmkikdyu] has joined #shogun | 20:50 | |
-!- naywhayare is now known as mlpack_naywhayar | 20:53 | |
-!- mlpack_naywhayar is now known as mlpack_ryan | 20:53 | |
-!- sonney2k is now known as shogun-toolbox_s | 21:01 | |
-!- shogun-toolbox_s is now known as shogun_sonney2k | 21:01 | |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has joined #shogun | 21:05 | |
* shogun_sonney2k dedup meeting in gsoc going on... | 21:09 | |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has quit [Ping timeout: 250 seconds] | 21:16 | |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 21:20 | |
-!- gsomix [~gsomix@188.168.2.209] has joined #shogun | 21:21 | |
gsomix | good evening | 21:21 |
@shogun_sonney2k | gsomix, any idea where lisitsyn is? | 21:21 |
gsomix | shogun_sonney2k, on the road to home. probably. but it's too late... | 21:22 |
@shogun_sonney2k | thought he will join the meeting... | 21:23 |
-!- HeikoS [~heiko@nat-166-123.internal.eduroam.ucl.ac.uk] has left #shogun [] | 21:23 | |
gsomix | shogun_sonney2k, what's the nick? | 21:23 |
gsomix | shogun_sonney2k, I can call him. | 21:24 |
@shogun_sonney2k | it is close to over | 21:24 |
gsomix | shogun_sonney2k, ok | 21:25 |
-!- shogun_sonney2k is now known as sonney2k | 21:26 | |
-!- mlpack_ryan is now known as naywhayare | 21:26 | |
-!- lisitsyn [~blackburn@213.87.143.137] has joined #shogun | 21:26 | |
lisitsyn | damn! | 21:26 |
lisitsyn | sonney2k: is it ok? | 21:26 |
@sonney2k | lisitsyn, jsut over | 21:26 |
lisitsyn | my ISP went crazy | 21:26 |
@sonney2k | all good | 21:26 |
lisitsyn | sonney2k: no conflicts? | 21:27 |
@sonney2k | yup | 21:27 |
lisitsyn | sonney2k: I am using 3G :D | 21:27 |
lisitsyn | ping is 8s hah | 21:28 |
lisitsyn | ok will wait my ISP' repairing of my connection | 21:29 |
-!- lisitsyn [~blackburn@213.87.143.137] has quit [Client Quit] | 21:30 | |
@sonney2k | pickle27, have you seen? | 21:30 |
@sonney2k | classifier_qda_modular.py setting 1/2 ERROR | 21:30 |
@sonney2k | python: /usr/include/eigen3/Eigen/src/Core/CwiseBinaryOp.h:146: Eigen::CwiseBinaryOp<BinaryOp, Lhs, Rhs>::CwiseBinaryOp(const Lhs&, const Rhs&, const BinaryOp&) [with BinaryOp = Eigen::internal::scalar_product_op<double, double>; Lhs = const Eigen::ArrayWrapper<Eigen::Block<Eigen::Map<Eigen::Matrix<double, -1, -1>, 0, Eigen::Stride<0, 0> >, 1, -1, false, true> >; Rhs = const Eigen::Map<Eigen::Array<double, -1, 1> >]: Assertion `lhs.rows() == rhs.rows() && lhs.co | 21:30 |
@sonney2k | ls() == rhs.cols()' failed. | 21:30 |
@sonney2k | Aborted | 21:30 |
pickle27 | sonney2k: I did see that | 21:31 |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/session] has joined #shogun | 21:31 | |
pickle27 | sonney2k: I wasn't totally sure what it meant | 21:31 |
@sonney2k | pickle27, well qda crashes :D | 21:31 |
pickle27 | sonney2k: was waiting for it to pop up in here so we could discuss | 21:31 |
@sonney2k | pickle27, you should be able to reproduce it by running the python_modular example | 21:32 |
pickle27 | sonney2k: the crash came from python, but the c++ example didn't crash, I bet there is a leak in the c++ code too then | 21:33 |
pickle27 | sonney2k: there was some code I wasn't crazy about around where the svd is done | 21:33 |
@sonney2k | doesn't look like a leak more like some shape of some matrix is wrong | 21:33 |
pickle27 | sonney2k: thats sort of what I meant I guess, c++ is getting away with it though | 21:34 |
pickle27 | sonney2k: I know roughly where this has got be from | 21:34 |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/session] has quit [Quit: lambday] | 21:37 | |
pickle27 | sonney2k: curious, the example just ran on my system | 21:37 |
@sonney2k | pickle27, go to tests/integration/python_modular | 21:48 |
@sonney2k | then run ./tester.py classifier_qda_modular.py | 21:48 |
pickle27 | sonney2k: I had the debug flag on so I rebuilt | 21:52 |
pickle27 | sonney2k: but it looks like I lost the python interface some how | 21:52 |
pickle27 | I'll rebuild and take a look into this | 21:52 |
pickle27 | sonney2k: I need to head out for a bit though so I'll get into this a bit later | 21:53 |
-!- zxtx [~zv@ool-44c1360a.dyn.optonline.net] has joined #shogun | 22:21 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 22:25 | |
gsomix | need to sleep | 22:52 |
gsomix | good night, guys | 22:52 |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has joined #shogun | 23:07 | |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has quit [Client Quit] | 23:08 | |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has joined #shogun | 23:19 | |
-!- katia [d557869c@gateway/web/freenode/ip.213.87.134.156] has quit [Quit: Page closed] | 23:34 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds] | 23:54 | |
--- Log closed Sat May 25 00:00:16 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!