--- Log opened Tue Jul 11 00:00:50 2017 | ||
micmn | wiking: lol you sleep like 4 hours? XD | 00:22 |
---|---|---|
-!- sukey1 [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection] | 03:40 | |
-!- sukey [~nodebot@ks312251.kimsufi.com] has joined #shogun | 03:40 | |
-!- mode/#shogun [+o sukey] by ChanServ | 03:40 | |
-shogun-buildbot:#shogun- Build nightly_all #14 is complete: Failure [failed test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/22/builds/14 | 03:52 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3920 merged by vigsterkr | 08:03 |
@sukey | [https://github.com/shogun-toolbox/shogun] vigsterkr pushed 2 commits: | 08:03 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/a26f9f142925af720d081550b6ea749532d03b22 | 08:03 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/ed1bb04b5aa0483616b38be02f81a5e6ce97d79f | 08:03 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-jathdrlvlcdzbkyj] has quit [Quit: Connection closed for inactivity] | 08:48 | |
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun | 09:11 | |
-!- geektoni_ [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun | 09:26 | |
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds] | 09:27 | |
-!- iglesiasg [~iglesiasg@217.119.234.214] has joined #shogun | 09:46 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 09:46 | |
@iglesiasg | qué pasa | 09:47 |
lisitsyn | what are reasonable answers? never taught yet | 10:12 |
@wiking | nada | 10:15 |
@wiking | bien | 10:15 |
@wiking | mierda | 10:15 |
@wiking | sukey, flip | 10:15 |
@sukey | (ノಥДಥ)ノ︵┻━┻・/ | 10:15 |
@wiking | any of these are fine :) | 10:15 |
@wiking | joder :) | 10:16 |
lisitsyn | wiking: mierda! | 10:26 |
@iglesiasg | shit is happening?! | 10:26 |
lisitsyn | haha | 10:28 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3906 synchronized by MikeLing | 10:28 |
@iglesiasg | any hint why swig would be including a class that has been removed? | 10:31 |
lisitsyn | removed how? | 10:32 |
@iglesiasg | it is not in libshogun.so, but dist-packages/_modshogun.so has it | 10:32 |
@wiking | ig | 10:32 |
@iglesiasg | remove the header and implementation file | 10:32 |
@wiking | template | 10:32 |
@wiking | ? | 10:32 |
@iglesiasg | no, it is not a template - I double check | 10:32 |
@iglesiasg | class CSumOne : public CDensePreprocessor<float64_t> | 10:33 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 10:33 |
@iglesiasg | I also removed the include and the rename from the Preprocessor.i | 10:34 |
@wiking | have you checked | 10:35 |
@wiking | that the in the build/src/interface.... the right verison of Preprocessor.i | 10:35 |
@wiking | is there | 10:35 |
@wiking | i.e. it's updated? | 10:35 |
@wiking | but i mean | 10:36 |
@wiking | fuck | 10:36 |
@wiking | i need to work | 10:36 |
@wiking | so you just do it | 10:36 |
@iglesiasg | ? | 10:36 |
@wiking | dunno :d | 10:37 |
@wiking | that's in short | 10:37 |
@iglesiasg | take it easy | 10:37 |
@iglesiasg | I am just having this doubt, tried a couple of things and run out of ideas for now | 10:37 |
@iglesiasg | so I asked for help | 10:37 |
@iglesiasg | I am not telling none to fix nothing | 10:37 |
micmn | iglesiasg: did you try deleting the old swig build dir? (happened to me :p) | 10:40 |
lisitsyn | wiking: don't worry be happy (c) v.v. putin | 10:40 |
lisitsyn | :P | 10:40 |
@wiking | hahah | 10:41 |
@wiking | that was so fucking good | 10:41 |
@wiking | https://www.youtube.com/watch?v=yfLx6X6461E | 10:42 |
@wiking | :D | 10:42 |
@wiking | [16:37] * moriarty (~dtijd@unaffiliated/moriarty) has joined #tensorflow | 10:42 |
@wiking | [16:37] <Trioxin> need help: https://hastebin.com/itakumibiw.sql | 10:42 |
@wiking | [16:38] <moriarty> That will be 10 bitcoins thank you | 10:42 |
@wiking | :DDDDDDDDDDDDDDDDDDDDDDDDDD | 10:42 |
@wiking | too good | 10:42 |
@iglesiasg | micmn: I removed the complete build directory | 10:43 |
@iglesiasg | found it! | 11:19 |
@iglesiasg | 'import modshogun' no no no | 11:19 |
@iglesiasg | 'import shogun' xD | 11:19 |
lisitsyn | import shogun!!!!!!!!!!!!! | 11:20 |
@iglesiasg | oh yeah | 11:20 |
@wiking | :) | 11:21 |
@wiking | crazy biatch :) | 11:21 |
lisitsyn | wiking: should've been swagshogun | 11:25 |
@wiking | import !@#$shogun | 11:26 |
@wiking | thug life | 11:26 |
lisitsyn | import if you can | 11:26 |
lisitsyn | fucker | 11:26 |
lisitsyn | wiking: it should also curse quite a lot | 11:26 |
@wiking | yes | 11:26 |
@wiking | bazdmeg | 11:26 |
@wiking | every 2nd __get__ | 11:26 |
lisitsyn | wiking: if you don't do anything it should shout at you like what are you waiting for | 11:26 |
@wiking | in the python interface | 11:26 |
@wiking | :) | 11:26 |
@sukey | [https://github.com/shogun-toolbox/shogun] vigsterkr pushed 2 commits: | 11:29 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/706872f70a813288ae5a6c4e9763b385c854a356 | 11:29 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/da08025401ee9d42405bdc9f45dfe63e725634f5 | 11:29 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 11:37 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 11:37 | |
@HeikoS | micmn: hi! | 11:39 |
micmn | HeikoS: hi, do you have an answer for me? :P | 11:40 |
@HeikoS | micmn: answer to what? :D | 11:40 |
@HeikoS | switching on/off the gpl? | 11:40 |
micmn | yeah | 11:40 |
@HeikoS | i see | 11:40 |
micmn | my last mail | 11:40 |
@HeikoS | so there is two mechanics | 11:41 |
@HeikoS | 1.) the src/gpl folder. | 11:41 |
@HeikoS | 2.) the cmake switch | 11:41 |
@HeikoS | the gpl folder contents can be deleted | 11:41 |
@HeikoS | but then the switch (currently) also has to be turned off | 11:42 |
@HeikoS | (because of the guards in unit tests) | 11:42 |
@HeikoS | I will later remove the switch and set USE_GPL_SHOGUN based on the presence of the gpl folder | 11:42 |
@HeikoS | so for now, set the cmake switch to off, and delete gpl folder to enable a full BSD build | 11:42 |
@HeikoS | one thing | 11:42 |
@wiking | HeikoS, i wouldnt do auto switching based on USE_GPL_SHOGUN | 11:43 |
@HeikoS | the build currently even fails if the folder is present and the switch is turned off (trained_model) | 11:43 |
@wiking | i mean baed on a presence of a folder or not | 11:43 |
@wiking | let it fail | 11:43 |
@wiking | if the folder is not and USE_GPL_SHOGUN | 11:43 |
@HeikoS | wiking: whatever, we are not there yet | 11:43 |
@HeikoS | yeah also ok | 11:43 |
@HeikoS | point is | 11:43 |
@wiking | i'm just saying | 11:43 |
@HeikoS | build is broken :) | 11:43 |
@HeikoS | wiking: yeah ok I take the point | 11:43 |
@HeikoS | so the switch stays . | 11:44 |
micmn | yeah the problem is just that | 11:44 |
@HeikoS | micmn: so can just guard the parts in the test that include files from gpl folder | 11:44 |
@HeikoS | with USE_GPL_SHOGUN | 11:44 |
micmn | yep that was my question, if I could rely on USE_GPL_SHOGUN or not | 11:44 |
@HeikoS | micmn: so if the folder is removed, the test actually compiles | 11:45 |
micmn | or it'll get removed sooner or later | 11:45 |
@HeikoS | but oif the folder is there and the switch is OFF then it doesnt | 11:45 |
@HeikoS | yeah you can rely on USE_GPL_SHOGUN | 11:45 |
@HeikoS | but not in the classes in the src/gpl folder | 11:45 |
@HeikoS | no guards are necessary in the folder | 11:45 |
@HeikoS | in the files in the folder I mean | 11:45 |
micmn | yeah yeah I get it | 11:45 |
@HeikoS | cool! | 11:45 |
@sukey | [https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/4950bfa9a47003988a971a3f4c1840d6529d6d9a by karlnapf | 11:46 |
@HeikoS | shogun-buildbot: force build 'nighly_bsd_license' | 11:47 |
-shogun-buildbot:#shogun- no such builder 'nighly_bsd_license' | 11:47 | |
@HeikoS | wiking: ah cool the webinterface works again for force building? | 11:47 |
@wiking | ? | 11:47 |
@wiking | what? | 11:47 |
@HeikoS | I just pressed the button "rebuild" | 11:49 |
@HeikoS | on buildbot website | 11:49 |
@HeikoS | and now a build is queued | 11:49 |
@wiking | ah cool | 11:51 |
@wiking | :D | 11:51 |
@wiking | didnt know | 11:51 |
@wiking | :DDDDDDDDD | 11:51 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 11:56 |
-shogun-buildbot:#shogun- Build osx1 - libshogun #77 is complete: Failure [failed test (failure)] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/77 | 12:03 | |
Trixis | wiking: any idea what could be causing this error? "MKL.cpp line 901: Assertion R >= 0 failed!" the code at that location doesnt seem to make it immediately obvious what R is... | 12:07 |
@wiking | mmmm | 12:07 |
@wiking | Trixis, for( p=0; p<num_kernels; ++p ) | 12:15 |
@wiking | thi sis my line 901 in MKL.cpp | 12:15 |
Trixis | wiking: odd. sec | 12:15 |
@wiking | ok anyhow couple of lines above | 12:15 |
@wiking | i'm wondeirng hat might have happened here | 12:15 |
@wiking | mmm | 12:16 |
@wiking | const float64_t R = CMath::sqrt( preR / mkl_norm ) * epsRegul; | 12:16 |
@wiking | and afaik neighter of the vars in elements in sqrt are complex | 12:16 |
@wiking | only way is that R is negative | 12:16 |
Trixis | http://www.shogun-toolbox.org/api/latest/MKL_8cpp_source.html seems to be line 901 in this | 12:16 |
@wiking | that epsRegul is negative | 12:16 |
@wiking | right? | 12:16 |
@wiking | but then again | 12:17 |
@wiking | const float64_t epsRegul = 0.01; // fraction of root mean squared deviation | 12:17 |
@wiking | :D | 12:17 |
Trixis | yeah | 12:17 |
@wiking | and i dont see anywhere else epsRegul being defined differently | 12:17 |
@wiking | in compute_optimal_betas_elasticnet | 12:17 |
Trixis | i mean it could be NaN as well, right? | 12:18 |
@wiking | mmm | 12:18 |
Trixis | although im not sure how NaN behaves with comparisons in c++ | 12:18 |
@wiking | you mean preR / mkl_norm | 12:18 |
@wiking | ? | 12:18 |
@wiking | i mean whatever f(NaN ) = NaN afaik | 12:18 |
Trixis | yes | 12:19 |
@wiking | and float64_t preR; | 12:19 |
Trixis | NaN is unordered as per specification | 12:19 |
Trixis | so | 12:19 |
Trixis | youd have false | 12:19 |
Trixis | on any comparison | 12:19 |
@wiking | yeah but | 12:19 |
@wiking | ah yeah | 12:19 |
@wiking | ok | 12:19 |
@wiking | so you get somehow there a nan? | 12:20 |
Trixis | i'd assume so it seems the only realistic scenario? | 12:20 |
@wiking | ok | 12:20 |
@wiking | so | 12:20 |
@wiking | ++ the logging level | 12:20 |
@wiking | of shogun | 12:20 |
@wiking | set_loglevel( | 12:20 |
@wiking | since there are many debugging info there | 12:20 |
@wiking | printed | 12:20 |
@wiking | like | 12:21 |
@wiking | SG_PRINT("MKL-direct: R = %e\n", R ) | 12:21 |
@wiking | so you should have an idea wtf is happening there | 12:21 |
Trixis | actually no, cant be nan, forgot sqrt of nefative doesnt return nan but throws an error | 12:21 |
@wiking | yeah but | 12:22 |
@wiking | sqrt(NaN) | 12:22 |
@wiking | ? | 12:22 |
Trixis | hm true | 12:24 |
@wiking | Trixis, from so: " What's the square root of -9? It's an imaginary number, but sqrt doesn't support imaginary results, so the result is nan (not a number). It's a domain error. The solution: Don't pass negative arguments to sqrt." | 12:24 |
@wiking | so actually sqrt(R < 0) = NaN | 12:24 |
Trixis | yeah | 12:24 |
Trixis | oh | 12:24 |
Trixis | right | 12:24 |
@wiking | would be great to have finally a patch on MKL | 12:25 |
@wiking | CMath::isnan(R) | 12:25 |
@wiking | :D | 12:25 |
@wiking | and print a different error | 12:25 |
@wiking | :P | 12:25 |
-!- geektoni_ [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds] | 12:25 | |
Trixis | im trying to find out right now how to increase the logging level via modular interfaces | 12:26 |
@wiking | so every SGObject | 12:26 |
@wiking | so like MKL | 12:26 |
Trixis | yeah | 12:27 |
@wiking | has get_global_io() | 12:27 |
@wiking | MKL.get_global_io().set_loglevel(MSG_DEBUG); | 12:27 |
@wiking | for example | 12:27 |
Trixis | so swig doesnt seem to be generating full interfaces for the SGIO object, there's only "SWIGTYPE_p_shogun__SGIO" which is has no methods | 12:30 |
-!- travis-ci [~travis-ci@ec2-54-144-106-53.compute-1.amazonaws.com] has joined #shogun | 12:30 | |
travis-ci | it's MikeLing's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/252328604 | 12:30 |
-!- travis-ci [~travis-ci@ec2-54-144-106-53.compute-1.amazonaws.com] has left #shogun [] | 12:30 | |
Trixis | nvm managed to get the output using init_shogun_with_defaults | 12:32 |
Trixis | yeah R is NaN | 12:32 |
Trixis | https://pastebin.com/SMLNSMkn this is the output | 12:32 |
Trixis | that Z doesnt look good either lol | 12:34 |
@HeikoS | Trixis, wiking probably some initialization problem | 12:35 |
@wiking | Trixis, PRs are MOST welcome | 12:35 |
@wiking | especially for MKL | 12:35 |
@wiking | :D | 12:35 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3923 opened by micmn | 12:52 |
micmn | HeikoS: ^ | 12:53 |
Trixis | wiking: judging by the code id assume beta[] is all 0s | 12:53 |
@HeikoS | micmn: chceking | 12:54 |
@HeikoS | micmn: so what about other guards? | 12:54 |
@HeikoS | micmn: this works for GPL | 12:54 |
@HeikoS | but what if a class is guarded otherwise? | 12:55 |
Trixis | that way you'd get Z to inf, then on line 878 you'd get 0*inf which indeed is a NaN. The only issue is i dont see why it would be 0 :\ | 12:55 |
micmn | HeikoS: other guards are already handled | 12:55 |
@HeikoS | ah sorry where? | 12:55 |
@HeikoS | line91? | 12:55 |
micmn | def is_guarded(include, defined_guards): | 12:55 |
micmn | yes | 12:56 |
@HeikoS | https://github.com/micmn/shogun/blob/aaa9c760e44d7ee1780da3eceb8a1661de1168be/tests/unit/base/trained_model_serialization_unittest.cc.py#L25 | 12:56 |
@HeikoS | micmn: mmmmh | 12:56 |
@HeikoS | micmn: couldnt we also use ctags here? | 12:56 |
@HeikoS | cant that be used to find out whether a class is defined or not? | 12:56 |
micmn | uhm no | 12:56 |
@HeikoS | I dont like this parsing preprocesor code too much :( | 12:56 |
@HeikoS | micmn: I wonder whether there is a ctags option that you can pass all the #defines so that it ignores certain classes if they are guarded? | 12:57 |
micmn | checking | 12:58 |
micmn | but I don't think so | 12:58 |
@HeikoS | micmn: if not, it is fine as well .... just would be a bit nicer | 12:58 |
@HeikoS | https://github.com/micmn/shogun/blob/aaa9c760e44d7ee1780da3eceb8a1661de1168be/tests/unit/base/trained_model_serialization_unittest.cc.py#L36 | 12:58 |
@HeikoS | micmn: this is quite nice | 12:58 |
@HeikoS | micmn: Ok i will wait for CI and then merge | 12:59 |
@HeikoS | micmn: can also be improved later on | 13:00 |
@HeikoS | micmn: thanks for the patch | 13:00 |
@HeikoS | I think I broke the nightly_bsd_license tests with my last commit, but at least all GPL code is now in that folder and shogun compiles | 13:01 |
micmn | HeikoS: can't find anything anyway I already checked when writing the script and resort to that as last option, I think that if we want to make it cleaner we should use clang's frontend | 13:04 |
@HeikoS | ok | 13:07 |
@HeikoS | micmn: nevermind then | 13:07 |
@HeikoS | I think this might cause problems at some point .... but maybe it doesnt :) | 13:07 |
micmn | mmm it'll do for sure :) | 13:09 |
lisitsyn | HeikoS: can I use some MMD-thingy to reweight dataset to match distribution? | 13:10 |
@HeikoS | lisitsyn: can you explain | 13:10 |
@HeikoS | lisitsyn: I am about to get lunch though | 13:10 |
lisitsyn | HeikoS: I have biased train and my test is everything else | 13:10 |
@HeikoS | but yes I guess! | 13:10 |
lisitsyn | :D | 13:10 |
lisitsyn | ok | 13:10 |
lisitsyn | can you give me any reference later? | 13:10 |
@HeikoS | ill chekc log in an hour | 13:10 |
lisitsyn | kk | 13:10 |
@HeikoS | "A kernel two sample test" JMLR | 13:10 |
@HeikoS | but I mean this is just nonparametic two-sample testing | 13:11 |
lisitsyn | although I don't need to test | 13:11 |
@HeikoS | lisitsyn: explain problem, and I can help | 13:11 |
lisitsyn | I need to reweight them | 13:11 |
lisitsyn | or do something to make my train unbiased | 13:11 |
@HeikoS | so that they are more similar? | 13:11 |
lisitsyn | yes | 13:11 |
@HeikoS | so you want a distance measure on distributions? | 13:11 |
lisitsyn | I want to change distribution of train to match the distribution of test | 13:11 |
@HeikoS | i.e. try to find a re-weighting such that distance is smaller | 13:11 |
Trixis | wiking: that way you'd get Z to inf, then on line 878 you'd get 0*inf which indeed is a NaN. The only way beta can be all 0 if sumw is all zero. guess ill have to try to get the c++ code into a debugger, or use another way to track the sumw computation | 13:11 |
lisitsyn | yes | 13:12 |
@HeikoS | sure that can be done | 13:12 |
@HeikoS | simplest idea: | 13:12 |
@HeikoS | choose a couple of weighting schemes | 13:12 |
@HeikoS | run MMD test between test and train for each | 13:12 |
@HeikoS | pick the one with largest p-value | 13:12 |
@HeikoS | there is details, but something like this has been done before | 13:13 |
@HeikoS | i.e. when training GANs :D | 13:13 |
@HeikoS | ok lunch | 13:13 |
@HeikoS | talk later | 13:13 |
lisitsyn | surer | 13:13 |
lisitsyn | thanks a lot | 13:13 |
Trixis | wiking: ... which would be because the svm returns 0 as the number of support vectors -- i made a typo and used SVRLight instead of SVMLight. checks out. I might make a PR later on adding an if() to check if betas/sumw are all 0 and print a meaningful error later on. | 13:18 |
Trixis | is parallel computation enabled by default in shogun? | 13:57 |
@wiking | Trixis, should be | 13:58 |
@wiking | Trixis, but you can query that as well | 13:58 |
@wiking | note that not everything is using parallel computation ;) | 13:58 |
@HeikoS | lisitsyn: back | 14:01 |
Trixis | wiking: yeah gotcha, i dont expect everything running in parallel, lol | 14:02 |
Trixis | wiking: at the very least the matrix computations should, lol | 14:03 |
Trixis | hm the training time looks reasonable for a low dimensional training sample where log10(n) = ~4, about 19minutes | 14:06 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3923 merged by karlnapf | 14:25 |
@sukey | [https://github.com/shogun-toolbox/shogun] karlnapf pushed 2 commits: | 14:25 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/aaa9c760e44d7ee1780da3eceb8a1661de1168be | 14:25 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/f027fb3a4bcb8dc42156826974ce9e53c1ef46ca | 14:25 |
-shogun-buildbot:#shogun- Build osx1 - libshogun #78 is complete: Success [build successful] - http://buildbot.shogun-toolbox.org:8080/#builders/25/builds/78 | 14:30 | |
@HeikoS | shogun-buildbot: force build nightly_bsd_licensy | 14:33 |
-shogun-buildbot:#shogun- no such builder 'nightly_bsd_licensy' | 14:33 | |
@HeikoS | shogun-buildbot: force build nightly_bsd_license | 14:34 |
@HeikoS | micmn: | 14:35 |
@HeikoS | this compares signed and unsigned | 14:36 |
@HeikoS | https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/preprocessor/KernelPCA_unittest.cc#L23 | 14:36 |
@HeikoS | probably should change something as it is not really consistent | 14:36 |
@wiking | muuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu | 14:37 |
@wiking | Trixis, mmmsure thing | 14:37 |
@wiking | Trixis, as said all PRs are most welcome | 14:38 |
@wiking | we can give creditz and fame :D | 14:38 |
@wiking | mmm i haven't seen olivier around lately | 14:39 |
@HeikoS | wiking: he is away | 14:40 |
@wiking | last week? | 14:41 |
@HeikoS | almost two weeks | 14:41 |
@wiking | kk | 14:41 |
micmn | HeikoS: what do you mean by not consistent? I'll add a cast to index_t to get rid of the warning but other than that should be ok | 14:41 |
@wiking | then that's why | 14:41 |
@wiking | HeikoS, have u seen this | 14:41 |
@HeikoS | micmn: the cast is not safe ;) | 14:41 |
@wiking | HeikoS, https://github.com/ericniebler/range-v3 | 14:41 |
@HeikoS | so in shogun, we should either use unsigned for positive quantities such as size | 14:42 |
@wiking | HeikoS, especially note view:: | 14:42 |
@HeikoS | or signed (seems better sometimes) | 14:42 |
@HeikoS | but not both | 14:42 |
@wiking | HeikoS, https://github.com/ericniebler/range-v3/blob/master/doc/index.md#views | 14:42 |
micmn | so that's a problem with index_t | 14:42 |
@HeikoS | micmn: yeah not your fault | 14:43 |
@HeikoS | I think the return type of size() should be index_t | 14:43 |
@HeikoS | or int64_t | 14:43 |
@HeikoS | wiking: uh that is good stuff | 14:43 |
@HeikoS | Ill check | 14:43 |
micmn | but why have index_t signed? | 14:43 |
@wiking | micmn, ok so there's huge discission about this | 14:44 |
@wiking | micmn, google et.al. :) | 14:44 |
@HeikoS | problem with signed is that STL is not | 14:44 |
@HeikoS | so we cannot nicely nix SG* and std:: | 14:44 |
@HeikoS | that is annoying | 14:44 |
@HeikoS | I'd rather follow STL | 14:45 |
@HeikoS | but no hard feelings | 14:45 |
micmn | ok thx :) | 14:45 |
@wiking | micmn, You should not use the unsigned integer types such as uint32_t, unless there is a valid reason such as representing a bit pattern rather than a number, or you need defined overflow modulo 2^N. In particular, do not use unsigned types to say a number will never be negative. Instead, use assertions for this. | 14:45 |
@HeikoS | wiking: how to talk to STL? Want to put assertions everywhere in code? | 14:46 |
@wiking | micmn, since for example... what if you wanna represent relative index ;) | 14:46 |
micmn | yeah... | 14:46 |
@wiking | micmn, that could and should be possible to be negative | 14:46 |
@wiking | so that's the idea about why index actually should be a signed value | 14:46 |
@wiking | instead of an unsigned | 14:46 |
@wiking | HeikoS, well size cannnot be ever negative | 14:47 |
@wiking | right? :) | 14:47 |
@wiking | i mean if we are in non-loco mode | 14:47 |
@wiking | HeikoS, i presume you are talking about size_t container::size() const; | 14:47 |
@wiking | right? | 14:47 |
@HeikoS | yes | 14:47 |
@wiking | cast-o machine assertation | 14:47 |
@HeikoS | so if you iterate over a std::vector to touch SG* then you get these warnings | 14:47 |
@wiking | life's a bitch | 14:47 |
@wiking | HeikoS, of course! | 14:47 |
@wiking | i mean | 14:48 |
@wiking | of course, nobody | 14:48 |
@wiking | should ever use | 14:48 |
@wiking | index for iterating over a data | 14:48 |
@wiking | :) | 14:48 |
@wiking | only for random accessing it | 14:48 |
@wiking | :P | 14:48 |
@HeikoS | that is true, BUT | 14:48 |
@HeikoS | that is what shogun does | 14:48 |
@wiking | yeye of ocurse | 14:48 |
@HeikoS | and currently if you start using std:: that sucks | 14:48 |
@wiking | it's a bad practice | 14:48 |
@wiking | not only in shogun | 14:48 |
@wiking | but many other places | 14:48 |
@wiking | but at least | 14:48 |
@wiking | stl stuff | 14:48 |
@HeikoS | so what to do right now? | 14:48 |
@wiking | has .begin() and .end() | 14:48 |
@wiking | so you could actually iterate through STLs | 14:49 |
@HeikoS | I guess we should give SG* a begin() and end() as well | 14:49 |
@wiking | with iterator | 14:49 |
micmn | SGMatrix::size() is not related to STL | 14:49 |
@wiking | HeikoS, yeah but we know this :) | 14:49 |
@HeikoS | still need count if you want to put stuff in SG* | 14:49 |
@wiking | HeikoS, count? | 14:49 |
@wiking | HeikoS, like size()? | 14:49 |
@HeikoS | vector[i] | 14:49 |
@HeikoS | the i | 14:49 |
@wiking | i = index_t | 14:49 |
@HeikoS | whatever | 14:49 |
@HeikoS | anyways | 14:49 |
@wiking | but size = can be unsgined :D | 14:49 |
@wiking | it's just shitty | 14:49 |
@HeikoS | I think we should change return type of SGVector::size | 14:50 |
@HeikoS | to index_t | 14:50 |
@HeikoS | currently it is unsigned | 14:50 |
@HeikoS | or change to int64_t | 14:50 |
@HeikoS | wiking: complaints on that? | 14:50 |
@wiking | HeikoS, index_t should actually be | 14:50 |
@wiking | int64_t :D | 14:50 |
@HeikoS | whatever | 14:50 |
@wiking | because i's a bit of a shitty stuff atm | 14:51 |
@wiking | that we can only index | 14:51 |
@HeikoS | but size() method should return index_t | 14:51 |
@wiking | 2^31 | 14:51 |
@HeikoS | micmn: so that return type can be changed and all good with that warning | 14:51 |
@HeikoS | wiking: and yes let's change it to 64 | 14:51 |
@HeikoS | i mean index_t | 14:51 |
@wiking | HeikoS, feature/BigInteger | 14:51 |
@wiking | ;) | 14:51 |
@HeikoS | ah yes I forgot | 14:52 |
@wiking | it's crazy fucking shit | 14:52 |
@HeikoS | in that case | 14:52 |
@wiking | many many places | 14:52 |
@HeikoS | micmn: it needs to be changed to int64_t for now :D | 14:52 |
@HeikoS | as otherwise it might well overflow | 14:52 |
@HeikoS | for big matrices | 14:52 |
@HeikoS | SG_MALLOC(8*1024*size); | 14:53 |
@HeikoS | wiking: I think I once saw that somewhere | 14:53 |
@wiking | HeikoS, what ? | 14:53 |
@HeikoS | wiking: hard-coded sizeof | 14:53 |
@wiking | hehehe | 14:53 |
@wiking | when i started that branch | 14:54 |
@wiking | and turned on overflow detection as error | 14:54 |
@wiking | laaaaaaaaaaalalala land | 14:54 |
@wiking | but yeah needs to be done one day | 14:54 |
@wiking | as it is stupid that | 14:54 |
@wiking | nowadays you could have a memory of 256G | 14:54 |
@HeikoS | lala land welcomes you | 14:55 |
@wiking | and then you cannot index | 14:55 |
@wiking | 200G of doubles | 14:55 |
@wiking | so yeah :) i mean of course | 14:55 |
@wiking | we know what we know :P | 14:55 |
@wiking | (never admit in public) | 14:55 |
@wiking | but yeah theoretically if we would see and SGMatrix of 200g | 14:56 |
@wiking | we could index it :) | 14:56 |
@wiking | *we couldn't index it | 14:56 |
@wiking | buut yeah stuff like this is crazy nice | 14:57 |
@wiking | auto rng = vi | view::remove_if([](int i){return i % 2 == 1;}) | 14:57 |
@wiking | | view::transform([](int i){return std::to_string(i);}); | 14:57 |
@wiking | boom | 14:57 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-jfouepcimcfulsss] has joined #shogun | 15:12 | |
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun | 15:20 | |
mikeling | HeikoS: ping | 15:26 |
@HeikoS | mikeling: pong | 15:26 |
mikeling | Hi , after I add set_global_seed(1) to calls.sg, the set_global_seed(1) been translated to set_global_seed() | 15:26 |
mikeling | which missing the parameter '1' | 15:26 |
@HeikoS | mikeling: oh | 15:26 |
@HeikoS | mikeling: I might have screwed something up | 15:26 |
@HeikoS | checking! | 15:27 |
-!- geektoni_ [c1cdd221@gateway/web/freenode/ip.193.205.210.33] has joined #shogun | 15:53 | |
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds] | 15:54 | |
@wiking | geektoni_, wadaaaaaaaaaaaaaapppppppppp | 15:58 |
geektoni_ | wiking: it still takes too much time | 15:58 |
@wiking | nooooooooooooooooouuuuuuuu | 15:59 |
geektoni_ | and it also seg fault now | 15:59 |
@wiking | ah buuutt | 15:59 |
@wiking | buuut | 15:59 |
geektoni_ | but it is a minor problem :P | 15:59 |
@wiking | geektoni_, cc5628d262fb09eac56b6bd8fa5d82df3776886f | 15:59 |
@wiking | interractive rebase and drop this | 15:59 |
@wiking | i told | 15:59 |
@wiking | u | 15:59 |
@wiking | :) | 15:59 |
@wiking | as this removes -j2 from make | 16:00 |
@wiking | so that's for sure one reason why it's slower than usual | 16:00 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 16:00 |
geektoni_ | boom | 16:01 |
@wiking | zuuuuuuuuuuupzuuuuuuuuuup | 16:01 |
@wiking | btw honsly | 16:06 |
@wiking | instead of making insane amount of traffic for our loved github/aws | 16:06 |
@wiking | maybe we should cache gtest source :)))) | 16:07 |
@wiking | geektoni_, but yead there's some weird memory error | 16:10 |
geektoni_ | wiking: yep. I'm working on it. | 16:14 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3859 synchronized by MikeLing | 16:15 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3859 synchronized by MikeLing | 16:17 |
@wiking | mmm | 16:21 |
@wiking | HeikoS, can we drop all linalg/ | 16:22 |
@wiking | and gpuvector/matrix | 16:22 |
@wiking | or at least on the cleanup | 16:22 |
@wiking | to drop it | 16:22 |
@wiking | ? | 16:22 |
-!- iglesiasg [~iglesiasg@217.119.234.214] has quit [Quit: leaving] | 16:37 | |
@wiking | geektoni_, we almost managed to do it :))) | 16:51 |
@wiking | but ccache is populated so i guess next time it'll be faster | 16:52 |
@wiking | HeikoS, can we drop serialization format tests? | 16:52 |
@wiking | like json/xml/ascii? | 16:52 |
geektoni_ | wiking: I've managed to fix the memory error | 16:52 |
@wiking | cool | 16:53 |
@wiking | push | 16:53 |
geektoni_ | the next push will (probably) make travis green | 16:53 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 17:06 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 17:07 |
@HeikoS | wiking: so dropping linalg, what do you mean exactly? | 17:24 |
@HeikoS | wiking: and drop serialization format tests, yes do it, though we havent got the new stuff in yet, so maybe wait a bit more | 17:25 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 17:54 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni | 17:56 |
lisitsyn | drop linalg? :D | 17:57 |
-!- geektoni_ [c1cdd221@gateway/web/freenode/ip.193.205.210.33] has quit [Ping timeout: 260 seconds] | 18:14 | |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 240 seconds] | 18:45 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-jfouepcimcfulsss] has quit [Quit: Connection closed for inactivity] | 19:08 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 merged by vigsterkr | 22:48 |
@sukey | [https://github.com/shogun-toolbox/shogun] vigsterkr pushed 5 commits: | 22:48 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/e3dde552813cb04bd1fc79b76c6415ac0ae778cc | 22:48 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/8ac2b028f00560346d282cc4b7bd1fe5861ee815 | 22:48 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/6f1603a08822d224528e43f09a5d0c27fbcdb107 | 22:48 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/4ad2814cdac331de6fbbdda15eda950e019dded5 | 22:48 |
@sukey | https://github.com/shogun-toolbox/shogun/commit/9be1025f596e1325fbc478a93647a93fe75474ee | 22:48 |
--- Log closed Wed Jul 12 00:00:51 2017 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!