--- Log opened Fri Jul 07 00:00:44 2017 | ||
-!- sukey [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection] | 00:29 | |
-!- sukey1 [~nodebot@ks312251.kimsufi.com] has joined #shogun | 00:29 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3908 merged by vigsterkr | 02:11 |
---|---|---|
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/41129f8ecb3955d37ed9dbba9643abca5495a493 by vigsterkr | 02:11 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-eiulobmjkkcvxlcr] has joined #shogun | 03:47 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3906 synchronized by MikeLing | 05:03 |
@wiking | mikeling, here? | 07:26 |
mikeling | wiking: yes, I'm checking your email :) | 07:27 |
mikeling | thank you | 07:27 |
@wiking | ok cool | 07:27 |
@wiking | ok i have some questions regarding the random story | 07:27 |
mikeling | ok | 07:27 |
@wiking | as on travis i'm getting weird errors | 07:27 |
@wiking | the integration tests for you locally | 07:27 |
@wiking | all of them fail? | 07:27 |
mikeling | not all, some of them failed | 07:28 |
@wiking | ok | 07:28 |
@wiking | i suspect the story there the same | 07:28 |
@wiking | i.e. fails due to different randoms | 07:28 |
mikeling | The following tests FAILED: | 07:28 |
mikeling | 212 - unit-NeuralLinearLayer (Failed) | 07:28 |
mikeling | 246 - unit-QuadraticTimeMMD (Failed) | 07:28 |
mikeling | 247 - unit-TwoDistributionTest (Failed) | 07:28 |
mikeling | 278 - integration_meta_cpp-clustering-gmm (Failed) | 07:28 |
mikeling | 282 - integration_meta_cpp-distance-cosine (Failed) | 07:28 |
mikeling | 299 - integration_meta_cpp-multiclass_classifier-random_forest (Failed) | 07:28 |
mikeling | 303 - integration_meta_cpp-neural_nets-feedforward_net_classification (Failed) | 07:28 |
mikeling | 304 - integration_meta_cpp-neural_nets-feedforward_net_regression (Failed) | 07:28 |
mikeling | that's all | 07:28 |
@wiking | mmmhmm | 07:28 |
@wiking | can you pastebin me the integration_meta_cpp-multiclass_classifier-random_forest output | 07:28 |
@wiking | ctest -R integration_meta_cpp-multiclass_classifier-random_forest -VV | 07:28 |
mikeling | sure, just a second | 07:29 |
mikeling | wiking: here https://pastebin.mozilla.org/9026480 | 07:29 |
@wiking | mmm | 07:31 |
@wiking | could you do the following | 07:31 |
@wiking | cd /Users/mikeling/shogun/build/examples/meta/cpp/multiclass_classifier | 07:32 |
@wiking | and copy paste the output of ls | 07:32 |
mikeling | wiking: https://pastebin.mozilla.org/9026481 | 07:33 |
@wiking | ok cool | 07:33 |
@wiking | do you have something like | 07:33 |
@wiking | random_forest.dat in ../data ? | 07:34 |
mikeling | let me see.. | 07:34 |
@wiking | or do a find ~/shogun/data -name 'random_forest.data' | 07:35 |
@wiking | or do a find ~/shogun/data -name 'random_forest.dat' | 07:35 |
mikeling | wiking: yes, https://pastebin.mozilla.org/9026482 | 07:35 |
mikeling | it's in '/Users/mikeling/shogun/data/testsuite/meta/multiclass_classifier/random_forest.dat' I guess | 07:36 |
@wiking | that's the only location? | 07:36 |
@wiking | aaah no | 07:36 |
@wiking | ok yesh | 07:36 |
mikeling | yes | 07:36 |
@wiking | so what you would need now is | 07:36 |
@wiking | pastebin | 07:36 |
@wiking | diff -uN /Users/mikeling/shogun/data/testsuite/meta/multiclass_classifier/random_forest.dat random_forest.dat | 07:36 |
mikeling | https://pastebin.mozilla.org/9026485 | 07:37 |
@wiking | what's in random_forest.dat | 07:38 |
@wiking | ? | 07:38 |
mikeling | https://pastebin.mozilla.org/9026486 | 07:39 |
mikeling | here is the content of random_forest.dat I think | 07:39 |
@wiking | mmmm | 07:39 |
@wiking | ok it's totally different | 07:40 |
mikeling | so, it means I mess something up in random_forest during work on the random? | 07:41 |
@wiking | mmmm | 07:49 |
@wiking | yeah there's definitely something different | 07:49 |
@wiking | i'm wondering | 07:49 |
@wiking | can you pastebin the random_forest.cpp | 07:49 |
@wiking | mikeling, ^ | 07:51 |
mikeling | sure, just a second | 07:51 |
mikeling | https://pastebin.mozilla.org/9026489 | 07:52 |
@wiking | nono | 07:52 |
@wiking | in the meta example | 07:52 |
@wiking | there should be a random_forest.cpp | 07:52 |
@wiking | mikeling, in cd /Users/mikeling/shogun/build/examples/meta/cpp/multiclass_classifier | 07:53 |
mikeling | https://pastebin.mozilla.org/9026490 here | 07:53 |
@wiking | oh ok | 07:53 |
@wiking | so this one had a seed 1 | 07:53 |
@wiking | mmm | 07:54 |
@wiking | can you try to run it | 07:54 |
@wiking | the test | 07:54 |
@wiking | like | 07:54 |
@wiking | SHOGUN_NUM_THREADS=1 ctest -R integration_meta_cpp-multiclass_classifier-random_forest | 07:55 |
mikeling | mmmm still failed | 07:55 |
mikeling | https://pastebin.mozilla.org/9026491 | 07:56 |
mikeling | here is the result | 07:56 |
@wiking | SHOGUN_NUM_THREADS=3 ? | 07:56 |
@wiking | although i guess you are on osx | 07:56 |
@wiking | so it doesn't matter :d | 07:56 |
mikeling | no, still not pass | 07:56 |
@wiking | ok nvermind | 07:57 |
@wiking | mikeling, can you ping heiko and lambday about the other unit tests failing | 07:57 |
@wiking | in the PR | 07:57 |
mikeling | ok, I will do that right away | 07:57 |
@wiking | and now we should talk about how to do the change to c++11 random | 07:58 |
mikeling | wiking: ok, I just make a description for these failed test and @Heikos and lambday in there | 08:06 |
mikeling | hope I describe it clear enough | 08:07 |
@wiking | cool | 08:09 |
@wiking | dunno if mater lisitsyn is here ? | 08:09 |
-!- rcurtin [~ryan@knife.lugatgt.org] has quit [Ping timeout: 240 seconds] | 08:41 | |
-!- rcurtin [~ryan@knife.lugatgt.org] has joined #shogun | 08:42 | |
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has joined #shogun | 09:09 | |
lisitsyn | wiking: hey | 10:10 |
@wiking | lisitsyn, ho | 10:10 |
@iglesiasg | I am tired of using technology | 10:22 |
@wiking | :D | 10:24 |
@wiking | iglesiasg, wood and stone | 10:24 |
@wiking | go! | 10:24 |
@wiking | lisitsyn, ok so here's another question | 10:24 |
@wiking | lisitsyn, i'm thiknig about dropping the whole CRandom object and just add some macros for lambdas to generate random | 10:25 |
@wiking | as the problem is the following | 10:25 |
@wiking | imagine we take the planned c++11 random engine | 10:25 |
@iglesiasg | wiking: the | 10:25 |
lisitsyn | a set of lambdas sounds like a class | 10:25 |
lisitsyn | :P | 10:25 |
@iglesiasg | wiking: the "hey ho" reminded me of the ayo in that song :P | 10:26 |
@wiking | but listen | 10:26 |
lisitsyn | yes | 10:26 |
@wiking | lisitsyn, so check here http://en.cppreference.com/w/cpp/numeric/random | 10:26 |
@wiking | you have like 10+ different distribution | 10:26 |
@wiking | do we really wanna wrap those in CRandom? | 10:26 |
lisitsyn | yes | 10:26 |
lisitsyn | ahh | 10:26 |
lisitsyn | ok agree just provide seed | 10:26 |
lisitsyn | and let people sample | 10:27 |
lisitsyn | LET MY PEOPLE SAMPLE! | 10:27 |
@wiking | :D | 10:27 |
@wiking | but yeah i mean the thing is for lambda | 10:27 |
lisitsyn | what this lambda is for? | 10:28 |
@wiking | nothing | 10:28 |
@wiking | i mean | 10:28 |
@wiking | we could have it for case yo uneed a simple random | 10:29 |
@wiking | so that you dont need to do this | 10:29 |
lisitsyn | make using random hard again | 10:29 |
@wiking | std::mt19937 prng(global_seed); | 10:29 |
@wiking | because if somebody | 10:30 |
lisitsyn | yeah but such lambda would produce the same thing all the time as we don't change the seed | 10:30 |
@wiking | i meant only the ctor of the prng :) | 10:31 |
@wiking | everything else (see distribuation etc) is up to the implementation | 10:31 |
@wiking | the only thing if we dont have an interface | 10:31 |
lisitsyn | generator<std::mt19937>() | 10:31 |
@wiking | or like a macro | 10:31 |
lisitsyn | I like it | 10:31 |
@wiking | to generate the prrng | 10:31 |
lisitsyn | template function | 10:31 |
@wiking | then users might mistakenly | 10:31 |
@wiking | not provide the seed for the engine | 10:31 |
@wiking | and then we are fucked | 10:32 |
@wiking | lisitsyn, get_prng() ? | 10:32 |
@wiking | and you can pass various PRNG engines? | 10:32 |
@wiking | i mean the problem with generator is that what the fuck it does generate? :) | 10:33 |
@wiking | and each time you ctor a new engine? | 10:33 |
lisitsyn | yes | 10:34 |
lisitsyn | but that's what you wanted no? | 10:34 |
@wiking | no what i wanted is just to get a prng set with the global seed :) | 10:34 |
@wiking | so i would like to have some thing | 10:34 |
lisitsyn | yes | 10:34 |
lisitsyn | get_prng<std::mt19937>(); | 10:35 |
@wiking | auto prng = get_prng(); // | 10:35 |
@wiking | std::normal_distribution<> normal_dist(mean, 2); | 10:35 |
@wiking | normal_dist(prng) | 10:35 |
@wiking | right? | 10:35 |
lisitsyn | ah there are also parametric | 10:35 |
lisitsyn | I'd avoid macro at all cost | 10:35 |
lisitsyn | they infect scope | 10:36 |
@wiking | yeah | 10:36 |
@wiking | we can then have a | 10:36 |
lisitsyn | but if possible with templates sounds good to me | 10:36 |
@wiking | template<typename T = std::mt19937> shogun::get_prng() {return T(global_seed);} | 10:37 |
lisitsyn | exactly | 10:37 |
@wiking | k | 10:37 |
lisitsyn | but parameters of normal | 10:37 |
lisitsyn | don't know how to provided that | 10:37 |
@wiking | what do you mean? | 10:38 |
lisitsyn | normal_distribution(mean, std) | 10:39 |
lisitsyn | we need to fix seed but not everything else | 10:39 |
@wiking | that is up to the user | 10:39 |
@wiking | this you'll have to write in the code yourself | 10:39 |
@wiking | we dont provide you a helper method to create a distribution over the prng | 10:40 |
@wiking | because actually there's a lot | 10:40 |
@wiking | and we dont want to provide wrappers for all those things | 10:40 |
@wiking | ok? | 10:40 |
@wiking | mikeling, ok so 2 things for the c++11 switch: you need to add the following global function to init.cpp: template<typename T = std::mt19937> shogun::get_prng() {return T(sg_random_seed);} | 10:42 |
@wiking | and replace uint32_t sg_random_seed = generate_seed(); | 10:44 |
@wiking | 10:44 | |
@wiking | bgw | 10:44 |
@wiking | ok | 10:44 |
@wiking | so the btw | 10:44 |
@wiking | nevemering | 10:45 |
@wiking | * | 10:45 |
@wiking | i'll comment | 10:45 |
@wiking | mikeling, check my comments... | 10:49 |
@wiking | lisitsyn, https://github.com/shogun-toolbox/shogun/pull/3902 merge? | 10:52 |
lisitsyn | wiking: yeah probably | 10:54 |
lisitsyn | wiking: it wants me to change codestyle | 10:54 |
lisitsyn | but better not | 10:54 |
@wiking | yeyey leave it as is | 10:56 |
@wiking | merge it! | 10:56 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3751 synchronized by micmn | 10:58 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 11:06 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3902 merged by lisitsyn | 11:18 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/a9235abd19af360880a3d2761826d2af6428a6b8 by lisitsyn | 11:18 |
@wiking | yey | 11:20 |
mikeling | wiking: oh hey! Sorry for the late reply. | 11:25 |
mikeling | thank you for your comment, I will address them asap | 11:26 |
@wiking | kk | 11:27 |
@wiking | thnx | 11:27 |
-!- iglesiasg [~iglesiasg@217.119.234.214] has quit [Quit: leaving] | 11:51 | |
micmn | wiking: when you have a minute please check the updated pr, ci pass should be ok | 12:08 |
@wiking | which one? | 12:08 |
micmn | Trained models tests | 12:09 |
@wiking | kk | 12:15 |
@wiking | checking | 12:15 |
@wiking | srry | 12:15 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 12:18 |
@wiking | micmn, this is great! | 12:40 |
@wiking | merging | 12:40 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3751 merged by vigsterkr | 12:41 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/370a66b3473b492cbcdc942179290414a7dc938a by vigsterkr | 12:41 |
micmn | Cool thx! | 12:41 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has joined #shogun | 12:45 | |
-shogun-buildbot:#shogun- Build bsd1 - libshogun #60 is complete: Failure [failed test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/31/builds/60 | 12:46 | |
-shogun-buildbot:#shogun- Build osx1 - libshogun #60 is complete: Failure [failed test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/60 | 12:46 | |
dougal | Hey, anyone knowledgeable about shogun build system around? I'm trying to add shogun to conda-forge, which builds Anaconda packages that would make it far easier+faster, at least for me, to get a release version of shogun installed. Partially working recipe: https://github.com/conda-forge/staged-recipes/pull/3237 | 12:48 |
dougal | Things that don't work yet: | 12:48 |
dougal | - How do I configure it to link against OpenBLAS (or MKL) instead of the system BLAS, especially getting Eigen to use them? | 12:48 |
dougal | - Is it possible to take a built libshogun.so / etc and then build the python interface separately? Re-building the full library for each python/numpy combination is super-wasteful (and times out on the conda-forge infrastructure), but conda's build system doesn't easily allow you to use partial build artifacts. A base shogun package plus a python interface package that builds based on libshogun.so / whatever else the base package wo | 12:50 |
dougal | - My Windows builds aren't working, though I pretty closely followed https://github.com/shogun-toolbox/shogun/blob/develop/appveyor.yml (which btw is different from the website instructions for building on windows, which also didn't work...). Any suggestions? | 12:51 |
dougal | Specifically, the VC 9 builds for python 2.7 are complaining that "Project file does not exist.", while the higher versions of VC for python 3.5/3.6 are warning about exception semantics not being turned on and also giving errors like: error C2039: 'max': is not a member of 'std' (???). Logs here: https://ci.appveyor.com/project/conda-forge/staged-recipes/build/1.0.14823 | 12:54 |
dougal | Oh actually, maybe https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/classifier/mkl/MKLMulticlassGLPK.cpp needs to #include <algorithm> on Windows to use std::min / std::max? Not sure why it would build on your version then, though | 13:01 |
dougal | Also getting: fatal error C1128: number of sections exceeded object file format limit: compile with /bigobj | 13:01 |
@wiking | dougal, hello hello | 13:06 |
@wiking | lemme read thisi one | 13:06 |
@wiking | as this is rather long | 13:06 |
dougal | yep! | 13:06 |
@wiking | but i need to finish something quickly | 13:06 |
@wiking | is it ok if i get back to you in 10 | 13:06 |
@wiking | ? | 13:06 |
dougal | sure, no problem | 13:06 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has quit [Quit: Page closed] | 13:17 | |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has joined #shogun | 13:17 | |
@wiking | geektoni, here/ | 13:53 |
@wiking | ? | 13:53 |
geektoni | wiking: yes | 13:53 |
@wiking | geektoni, https://github.com/shogun-toolbox/tflogger | 13:58 |
@wiking | viola | 13:58 |
@wiking | sorry for the delay | 13:58 |
geektoni | wiking: \o/ np | 13:58 |
@wiking | there's gonna be several pushes so that i get travis working | 13:58 |
geektoni | wiking: kk, btw, can we add tflogger to the docker image? so that I can finally test on Travis the cod I wrote :P | 14:00 |
@wiking | yes | 14:00 |
@wiking | do you wanna add a pr for that? | 14:00 |
geektoni | wiking: sure | 14:02 |
geektoni | wiking: the `make install` command fails because when you set TFLOGGER_HEADER into the CMakeLists file you still use tensorflow_logger.h instead of event_logger.h. | 14:12 |
@wiking | ah i havent tested that | 14:12 |
@wiking | idiot me | 14:12 |
@wiking | sorry | 14:12 |
@wiking | ok i'm fixing + trying to get protobufv3 for the stupid travis image | 14:13 |
@wiking | so the CI works | 14:13 |
geektoni | wiking: np, when refactoring+renaming there are almost 1000 things which slip through ;) | 14:15 |
@wiking | geektoni, HEAD has the fix | 14:19 |
geektoni | wiking: cool | 14:20 |
@wiking | geektoni, ok look i did a change | 14:26 |
@wiking | geektoni, can you test a full cycle with 554aad3353fc4d36f6943334f98bb3424f28e4e3 | 14:26 |
@wiking | ? | 14:26 |
@wiking | geektoni, so basically make sure that you use that and then run it with any example... and see if you can load it in into TB | 14:26 |
@wiking | plz | 14:26 |
@wiking | much appreciated! | 14:27 |
@wiking | if that works we'll make a 0.1.0 release | 14:27 |
geektoni | ok | 14:27 |
geektoni | give me some minutes | 14:27 |
@wiking | kk thnx heaps | 14:29 |
@wiking | dougal, ok lemme read now your questions :) | 14:38 |
@wiking | dougal, ok so about being able to compile libshogun and interface separatelu | 14:42 |
@wiking | yes it's possible | 14:42 |
@wiking | by default libshogun.so building is turned on | 14:42 |
@wiking | dougal, the LIBSHOGUN cmake parameter controls whether to turn on or off libshogun.so build | 14:43 |
@wiking | namely if you do cmake -DLIBSHOGUN=OFF -DPythonModular=On .. | 14:43 |
@wiking | then it'll try to find on your system the installed libshogun and create an interface against that | 14:44 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3906 synchronized by MikeLing | 14:46 |
dougal | @wiking aha! thanks, will try | 14:52 |
dougal | wiking: any thoughts on the windows build issues? | 14:52 |
@wiking | dougal, regarding MKL/Openblas | 14:54 |
@wiking | is there a special thingy with eigen + openblas ? i know about MKL but not about openblas. if you have info regarding this could you link it to me plz | 14:54 |
dougal | i don't think there's anything special about it, should be the same as MKL | 14:55 |
@wiking | mmmm | 14:55 |
dougal | (as far as i know) | 14:55 |
@wiking | nono | 14:55 |
@wiking | i mean MKL has some special things with | 14:55 |
@wiking | Eigen | 14:55 |
@wiking | openblas is just openblas afaik | 14:55 |
@wiking | but can you give me the runtime env | 14:55 |
@wiking | or how i can reproduce | 14:55 |
@wiking | on which/where should detect the custom location | 14:55 |
@wiking | of the openblas? | 14:55 |
@wiking | either i can help you with giving you the right cmake flags | 14:55 |
@wiking | or simply fix the cmake of shogun | 14:55 |
@wiking | MKL is a different story... there some extra env vars needs to be set | 14:56 |
dougal | https://eigen.tuxfamily.org/dox/TopicUsingBlasLapack.html says it works with openblas, as long as you set those macros | 14:56 |
@wiking | in order that eigen starts using that given MKL | 14:56 |
@wiking | dougal, ok i'll check into it | 14:56 |
@wiking | ah ok | 14:56 |
dougal | i haven't fiddled with pointing cmake at it directly yet because i thought that would be an issue too | 14:56 |
@wiking | this assume 3.3.x eigen | 14:56 |
dougal | but will try | 14:56 |
dougal | does shogun use blas directly outside of eigen also? | 14:56 |
@wiking | yes | 14:57 |
@wiking | less and less | 14:57 |
@wiking | but still it does in some places | 14:57 |
dougal | okay, so i'll try fiddling with the cmake variables there, shouldn't be har | 14:57 |
dougal | d | 14:57 |
@wiking | k | 14:57 |
@wiking | how do you acquire eigen? | 14:57 |
@wiking | do you make shogun to use it's own eigen? | 14:57 |
@wiking | or you have some system eigne? | 14:57 |
@wiking | *eigen | 14:57 |
dougal | been using conda's eigen | 14:57 |
@wiking | mmm can you help me understand what is the env | 14:58 |
dougal | which is 3.3.3 | 14:58 |
@wiking | that you have | 14:58 |
@wiking | that has blas etc | 14:58 |
dougal | installing everything via conda | 14:58 |
@wiking | so i could locally reproduce some stuff? | 14:58 |
@wiking | like a set of command will do for me | 14:58 |
dougal | are you familiar with conda stuff? | 14:58 |
@wiking | very brief | 14:59 |
@wiking | it'd be very helpful for me if you just give the set of commands | 14:59 |
dougal | yep | 14:59 |
dougal | so | 14:59 |
@wiking | that you have for your conda env | 14:59 |
@wiking | and then i can start experimenting | 14:59 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3906 synchronized by MikeLing | 14:59 |
dougal | 1 - install conda (https://conda.io/miniconda.html) | 14:59 |
dougal | 2 - conda config --prepend channels conda-forge | 15:00 |
@wiking | regarding windows: you've seen our appveyor? there's only Visual Studio 14 2015 and older support for shogun | 15:00 |
dougal | 3 - conda install conda-build | 15:00 |
@wiking | i meant to say newer | 15:00 |
@wiking | as older visual studios does not have a c++11 compiler | 15:00 |
@wiking | afaik | 15:00 |
@wiking | ok testing | 15:00 |
@wiking | dougal, python3 or 2? | 15:00 |
dougal | either | 15:01 |
dougal | 4 - git clone https://github.com/dougalsutherland/staged-recipes/ -b shogun | 15:01 |
dougal | cd staged-recipes/recipes/shogun | 15:01 |
dougal | conda build . --python 2.7 --numpy 1.13 | 15:01 |
dougal | that will build the recipe | 15:01 |
dougal | if you want to fiddle with it in there, easiest thing is to edit build.sh to add a "bash" call or whatever | 15:02 |
dougal | or whatever equivalent with bld.bat on windows | 15:02 |
@wiking | k lemme try | 15:02 |
dougal | re: windows: does the python interface for that actually work? my impression was that the ABI is not compatible across visual studio versions, and you need to use same version as python is compiled with | 15:03 |
dougal | python 2.7 is compiled with VC 9 | 15:03 |
@wiking | i've honestly actually never tried | 15:04 |
@wiking | SWIG + windows :) | 15:04 |
@wiking | there's a ticket for it | 15:04 |
@wiking | but nobody did that yet | 15:04 |
dougal | heh okay | 15:04 |
@wiking | but again | 15:04 |
@wiking | VC 9 | 15:04 |
@wiking | i'm not so sure if that's gonna fly | 15:04 |
dougal | okay, i think that means that python interface to shogun is not possible with windows + python 2.7 | 15:05 |
dougal | unless i'm misunderstanding and it actually works for some reason | 15:05 |
dougal | python >= 3.5 uses VC 14, which is still giving me compile errors | 15:06 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3909 opened by romainviard | 15:07 |
@wiking | geektoni, so we release? :) | 15:08 |
geektoni | wiking: mmm Tensorboard works but shouts this error: WARNING:tensorflow:Invalid event.proto file_version. Defaulting to use of out-of-order event.step logic for purging expired events. | 15:11 |
@wiking | yep yep | 15:11 |
@wiking | that's what i suspected | 15:11 |
@wiking | fucker | 15:11 |
@wiking | geektoni, if you step one commit back c88805798127bcf0420f7c3cad8331d1052358a5 | 15:12 |
@wiking | that should not have this error i suspect | 15:12 |
geektoni | let me check | 15:12 |
geektoni | yeah, with the previous commit, the error disappears. | 15:18 |
geektoni | wiking: ^ | 15:18 |
@wiking | mmm | 15:18 |
@wiking | i dont understand the fucking ostringstream :) | 15:18 |
@wiking | ok | 15:20 |
@wiking | i think i have a soltuon | 15:20 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 15:22 |
dougal | wiking: building interface separately seems to be working, thanks. :) | 15:49 |
@wiking | dougal, nw | 15:49 |
dougal | re openblas: it looks like the problem is this section, https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/CMakeLists.txt#L278 | 15:49 |
dougal | https://github.com/Kitware/CMake/blob/master/Modules/FindLAPACK.cmake will find openblas | 15:49 |
dougal | but since it's not accelerate or mkl, this code then looks for atlas or cblas | 15:50 |
dougal | and openblas provides a cblas interface but not libcblas.so.3, so it uses system blas | 15:50 |
@wiking | geektoni, so good? | 15:53 |
@wiking | because then i release 0.1.0 and then we can add that to the Docker | 15:53 |
geektoni | wiking: building takes some time on my machine. Give my 5 minutes. | 15:54 |
@wiking | :))))))) | 15:54 |
@wiking | build what? tflogger? | 15:54 |
@wiking | you use ccache right? | 15:54 |
geektoni | wiking: yes yes, of course. | 15:56 |
@wiking | geektoni, mmm are you using -G"Ninja" | 16:01 |
@wiking | ? | 16:01 |
@wiking | for cmake | 16:01 |
geektoni | wiking: mmh nope | 16:02 |
@wiking | hooo | 16:03 |
@wiking | try it | 16:03 |
geektoni | btw, now the file generated by TFLogger are empty :/ | 16:03 |
@wiking | micmn, have you seen this: http://buildbot.shogun-toolbox.org:8080/#/builders/25/builds/60/steps/6/logs/stdio | 16:03 |
@wiking | geektoni, what? :D | 16:04 |
@wiking | geektoni, the test generates files :DDD | 16:04 |
@wiking | (unit test) | 16:04 |
@wiking | micmn, same with bsd http://buildbot.shogun-toolbox.org:8080/#/builders/31/builds/60/steps/6/logs/stdio | 16:04 |
geektoni | wiking: nono, after the TFLogger update, the output file generated for TB are empty :/ there are no data inside. | 16:05 |
@wiking | geektoni, try this locally | 16:05 |
@wiking | for tflogger | 16:06 |
@wiking | cmake -DBUILD_TESTS=ON .. | 16:06 |
@wiking | make | 16:06 |
@wiking | and then | 16:06 |
@wiking | ./test/tfloggerTest | 16:06 |
@wiking | tensorboard --logdir . | 16:06 |
@wiking | the test will generate a file | 16:07 |
@wiking | oh nice | 16:07 |
@wiking | wtf happened now | 16:07 |
@wiking | :DDDDDDDDD | 16:07 |
@wiking | ok no | 16:08 |
@wiking | it worksforme | 16:08 |
@wiking | geektoni, u? | 16:08 |
@wiking | mmmm | 16:08 |
@wiking | actually weird stuff is happening | 16:08 |
geektoni | mmh just a sec | 16:08 |
@wiking | on the first run | 16:08 |
@wiking | (the first file) will not be valid | 16:09 |
@wiking | cannot even read it | 16:09 |
@wiking | any consecutive test run that generates a file | 16:09 |
@wiking | seems to be ok | 16:09 |
geektoni | ?? lol | 16:09 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3910 opened by dougalsutherland | 16:09 |
@wiking | yeah pretty much lol | 16:09 |
dougal | (^ that issue was me, about openblas) | 16:10 |
@wiking | yep yep | 16:10 |
@wiking | geektoni, idk | 16:10 |
@wiking | fucking string concatenation man :DDDD | 16:10 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3910 | 16:10 |
geektoni | wiking: ahahah | 16:11 |
micmn | wiking: yeah, dont know wtf is going on I guess different random training data | 16:12 |
@wiking | micmn, fixed seed? | 16:14 |
@wiking | geektoni, mmm this is sooooo weird | 16:14 |
micmn | Are those builders easily replicabile locally? | 16:14 |
@wiking | micmn, one is osx the other is freebsd | 16:15 |
@wiking | micmn, i can test on osx | 16:16 |
@wiking | geektoni, works for u | 16:17 |
@wiking | ? | 16:17 |
micmn | So they are not virtualized | 16:17 |
@wiking | noup | 16:17 |
geektoni | wiking: mmh, for me works | 16:18 |
@wiking | so wtf | 16:19 |
@wiking | :))) | 16:19 |
geektoni | wiking: mmh I'm checking | 16:21 |
@wiking | REQUIRE_THAT( actual.file_version(), Catch::Equals(getCurrentFileVersion()) ) | 16:23 |
@wiking | with expansion: | 16:23 |
@wiking | "brain.Event:2" equals: "brain.Event:2" | 16:23 |
@wiking | arg | 16:23 |
@wiking | i guess there's a \0 somewhere | 16:23 |
geektoni | wiking: lol | 16:24 |
geektoni | wiking: I think it is a problem on the Shogun-side | 16:32 |
@wiking | geektoni,mmm | 16:33 |
geektoni | wiking: I mean, a problem of my code | 16:33 |
@wiking | maok | 16:37 |
@wiking | lets see | 16:37 |
@wiking | because i cannot reproduce any bad things with the last commit | 16:37 |
geektoni | wiking: I'm an idiot | 16:39 |
geektoni | I figured out the problem | 16:39 |
@wiking | hehhe what was it? | 16:39 |
@wiking | geektoni, so it works? | 16:40 |
geektoni | If a class doesn't emit any object, of course the Event file will be empty. | 16:40 |
geektoni | yes yes | 16:40 |
@wiking | ehhe ok | 16:40 |
@wiking | i release it then | 16:40 |
geektoni | you can finally mark tflogger as 0.1.0 | 16:40 |
geektoni | ;) | 16:40 |
@wiking | https://github.com/shogun-toolbox/tflogger/archive/v0.1.0.tar.gz | 16:41 |
@wiking | this needs to go into shogun-dev docker | 16:41 |
@wiking | and finally then your PR will be testable :) | 16:42 |
@wiking | geektoni, you want me to do the Dockerfile? | 16:42 |
geektoni | wiking: nono, I've already done it | 16:43 |
geektoni | thanks anyway :D | 16:43 |
@wiking | geektoni, where's the dockerfile | 16:44 |
@wiking | i dont see it | 16:44 |
geektoni | updating it with the new link | 16:44 |
@wiking | kk | 16:45 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3911 opened by geektoni | 16:46 |
geektoni | wiking: ^ | 16:46 |
@wiking | ok | 16:46 |
@wiking | merging | 16:46 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3911 merged by vigsterkr | 16:47 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/78e5ef5d62fb36b14af09121c2838a6e0dadcf24 by vigsterkr | 16:47 |
@wiking | ok it's in queue https://hub.docker.com/r/shogun/shogun-dev/builds/by6wfedsnztxkzqdjdtdjpj/ | 16:48 |
geektoni | cool | 16:49 |
@wiking | once ready i'll restart the travis build for this pr https://github.com/shogun-toolbox/shogun/pull/3877 | 16:49 |
geektoni | wiking: thanks! | 16:49 |
@wiking | and then if that's green i'll merge it into the feature brnach | 16:49 |
@wiking | do we miss something ?:) | 16:49 |
@wiking | or this feature branch can land in develop i guess? :) | 16:50 |
geektoni | wiking: mmh, new files still need a license, I guess | 16:50 |
@wiking | which new files? | 16:50 |
geektoni | but yeah, I think it can land in develop ;) | 16:50 |
geektoni | ParameterObserver* | 16:51 |
@wiking | ah yeah | 16:51 |
@wiking | add BSD license plz | 16:51 |
@wiking | ah wait | 16:51 |
@wiking | we have a lot of the TODOs | 16:51 |
@wiking | we need to add the new signal handler | 16:52 |
@wiking | right? | 16:52 |
@wiking | we'll ask mike's collaboration for that | 16:52 |
-shogun-buildbot:#shogun- Build osx1 - libshogun #61 is complete: Success [build successful] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/61 | 16:52 | |
geektoni | wiking: some TODOs are placed in classes which do not derive directly from CMachine | 16:53 |
@wiking | ohooooooooooo | 16:53 |
@wiking | i see this | 16:53 |
@wiking | https://github.com/shogun-toolbox/shogun/pull/3877/files#diff-3ea04b3b4f57c2582e62770948450c74R39 | 16:53 |
geektoni | so we'll need to plan carefully what to do | 16:53 |
@wiking | geektoni, cant we actually make this templated? | 16:53 |
geektoni | mmmh it is not actual templates, because we are dealing with Any | 16:54 |
geektoni | google test offers utilities only for template | 16:54 |
@wiking | geektoni, https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/classifier/LDA_unittest.cc#L95 | 16:54 |
@wiking | mmm | 16:55 |
@wiking | micmn, lemme test locally what's happening with that unit test | 16:56 |
micmn | Ok thx | 16:57 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has quit [Quit: Page closed] | 17:04 | |
@wiking | geektoni, error :D | 17:08 |
geektoni | wiking: docker? | 17:08 |
@wiking | geektoni, yep | 17:09 |
@wiking | missing dependency | 17:09 |
@wiking | Could NOT find Protobuf (missing: PROTOBUF_LIBRARY PROTOBUF_INCLUDE_DIR) | 17:09 |
@wiking | geektoni, yeah you need to add it to the apt | 17:09 |
geektoni | ahh crap | 17:10 |
geektoni | ok, time for a new PR! | 17:10 |
@wiking | protobuf-compiler libprotobuf-dev | 17:10 |
@wiking | these 2 | 17:10 |
geektoni | thnx | 17:11 |
@wiking | micmn, uhoh unit-TrainedModelSerializationTest ... passed on my machine | 17:12 |
micmn | lol | 17:12 |
@wiking | micmn, interestingly... even on buildbot that's the case | 17:12 |
@wiking | bsd still failing | 17:12 |
@wiking | lemme check it there | 17:13 |
@wiking | :) | 17:13 |
@wiking | mikeling, here? | 17:15 |
mikeling | wiking: yes | 17:16 |
@wiking | ok so have you seen our next approach? | 17:16 |
@wiking | for the global | 17:16 |
mikeling | seen? you mean the global seed, right? or the get_prng()? | 17:17 |
@wiking | get_prng | 17:17 |
@wiking | mikeling, but dont push that into this pr | 17:18 |
@wiking | yet | 17:18 |
mikeling | yeah, and it to the pr https://github.com/shogun-toolbox/shogun/pull/3906/commits/48aa9bcb21c35fd03a223ee5bb91a41bd688942f | 17:18 |
mikeling | oh | 17:18 |
@wiking | aaah | 17:18 |
@wiking | i see | 17:18 |
@wiking | cool | 17:18 |
@wiking | it's there | 17:18 |
@wiking | okok | 17:18 |
mikeling | sorry, | 17:18 |
@wiking | it's good | 17:18 |
@wiking | nono | 17:18 |
@wiking | it's fine | 17:18 |
@wiking | it's there but we dont use it yet | 17:18 |
mikeling | ok | 17:19 |
@wiking | mikeling, lets see if we get an answer from the people | 17:19 |
@wiking | for failures | 17:20 |
mikeling | wiking: and I' checking the the sg_io things, so I found the spdlog already have info, error, debug in its logger | 17:20 |
mikeling | do we need to replace it with sg_io or io? | 17:20 |
mikeling | So we don't need SGIO anymore? | 17:21 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3912 opened by geektoni | 17:22 |
@wiking | mikeling, well the idea is that we just put the spdlog | 17:22 |
@wiking | into SGIO class | 17:22 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3912 merged by vigsterkr | 17:23 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/4ac8bfaab6259c94df4282c0c260d5e7e9a44d78 by vigsterkr | 17:23 |
-!- OXPHOS [92bd305b@gateway/web/freenode/ip.146.189.48.91] has joined #shogun | 17:24 | |
-shogun-buildbot:#shogun- Build osx1 - libshogun #62 is complete: Failure [failed test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/62 | 17:28 | |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has joined #shogun | 17:41 | |
@wiking | geektoni, sad days | 17:48 |
geektoni | wiking: docker still fails :/ summary.proto:1:10: Unrecognized syntax identifier "proto3". This parser only recognizes "proto2". | 17:48 |
@wiking | yes | 17:48 |
geektoni | yes, sad days. | 17:49 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/6344c7496484b83241a43b6f828023f59082ee80 by vigsterkr | 17:51 |
@wiking | hopefully this'll fix it | 17:51 |
geektoni | cool! | 17:52 |
dougal | still getting windows errors, using VC14 on x64 (I think same environment as shogun's appveyor build...): lots of warnings about wanting /EHsc, a few errors in MKLMulticlassGLPK about std::min, std::max, and a fatal error asking for /bigobj | 17:52 |
dougal | https://ci.appveyor.com/project/conda-forge/staged-recipes/build/1.0.14839/job/vcnqd6krwqenb7d6 | 17:52 |
dougal | :| | 17:53 |
dougal | linux/mac are working though, except for the openblas problems (and no openmp on mac because of compiler issues) | 17:55 |
@wiking | dougal, checking | 17:56 |
@wiking | dougal, openmp on osx only possible if u would have gcc | 17:56 |
dougal | yeah i know | 17:56 |
dougal | (or homebrew version of clang) | 17:56 |
-shogun-buildbot:#shogun- Build osx1 - libshogun #63 is complete: Success [build successful] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/63 | 17:57 | |
dougal | (conda packages gcc, but the version is too old; some comments about that at https://github.com/conda-forge/staged-recipes/pull/3237) | 17:57 |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/8bf4aed8252e1c3772d249eb8b854de3286dc2e0 by vigsterkr | 18:03 |
@wiking | dougal, ^ this should fix the problem of max/min | 18:03 |
dougal | cool, yeah | 18:03 |
dougal | looking at your appveyor builds, e.g. https://ci.appveyor.com/project/vigsterkr/shogun/build/1014: it seems like you don't have GLPK installed there, so that's why not getting the error | 18:03 |
dougal | and they're also spitting out the same /EHsc warnings | 18:04 |
@wiking | yes | 18:04 |
@wiking | that warning is 'fine' | 18:04 |
@wiking | i'm just wondering what's the difference of LDA | 18:04 |
@wiking | i guess eigen | 18:04 |
dougal | yeah | 18:04 |
@wiking | verison | 18:04 |
dougal | (for reference, which maybe you know: your appveyor builds download eigen 3.2.8, mine is using system 3.3.3) | 18:09 |
@wiking | probably that's the story | 18:10 |
dougal | i'll try using system 3.2.8 | 18:11 |
dougal | see if that works | 18:11 |
dougal | okay, builds queued, i'm heading out for a while | 18:13 |
dougal | let me know on the github PR or whatever if you happen to figure anything else out :) | 18:13 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has quit [Quit: Page closed] | 18:13 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/72cfeb15639d4b39565d61700dce11c0d839c4da by vigsterkr | 18:14 |
-shogun-buildbot:#shogun- Build osx1 - libshogun #64 is complete: Failure [test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/64 | 18:16 | |
-shogun-buildbot:#shogun- Build osx1 - libshogun #65 is complete: Success [build successful] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/65 | 18:42 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 18:50 |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 20:06 |
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has quit [Quit: Leaving.] | 20:35 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-eiulobmjkkcvxlcr] has quit [Quit: Connection closed for inactivity] | 20:35 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3877 synchronized by geektoni | 21:51 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has joined #shogun | 22:42 | |
dougal | wiking: FYI, still getting the same error with eigen 3.2.8, so i guess that's not it... | 22:42 |
-!- dougal [56adb86b@gateway/web/freenode/ip.86.173.184.107] has quit [Ping timeout: 260 seconds] | 22:49 | |
sukey1 | [https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3913 opened by MaximilianLomardo | 23:52 |
--- Log closed Sat Jul 08 00:00:46 2017 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!