--- Log opened Thu Mar 23 00:00:16 2017 | ||
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 00:01 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 260 seconds] | 00:06 | |
-!- travis-ci [~travis-ci@54.198.124.107] has joined #shogun | 00:10 | |
travis-ci | it's Sergey Lisitsyn'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/214028117 | 00:10 |
---|---|---|
-!- travis-ci [~travis-ci@54.198.124.107] has left #shogun [] | 00:10 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 00:26 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 00:30 | |
-!- neha [ca8e6c97@gateway/web/freenode/ip.202.142.108.151] has quit [Ping timeout: 260 seconds] | 00:56 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 01:04 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 264 seconds] | 01:09 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 01:45 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 260 seconds] | 01:49 | |
-!- nagdev [dce39546@gateway/web/freenode/ip.220.227.149.70] has joined #shogun | 01:53 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 02:25 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 02:29 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-rncdiloslifjixkx] has joined #shogun | 02:32 | |
-!- Killin [~Killin-A1@2602:304:68a6:ddc0:79ce:1107:f8a6:7335] has joined #shogun | 02:45 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 02:49 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 02:54 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 03:23 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 03:27 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 04:02 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 04:07 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 04:56 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 05:10 | |
-!- rads [3a92880d@gateway/web/freenode/ip.58.146.136.13] has joined #shogun | 07:15 | |
mikeling | I'm still wondering how to make DataGenerator as a global fixture and use it in the all the tests :/ | 07:20 |
mikeling | let's assume we have a global fixture like this https://gist.github.com/MikeLing/5c19e062e7a689c010f0427d4c329ec7 and registered it by ::testing::AddGlobalTestEnvironment(new DataGenerator) | 07:21 |
mikeling | what should I do next? Can I use the features_train and labels_train in it directly? use something like TEST(CSVMLight, train)? or TEST_F(DataGenerator, tainSVMLight?) | 07:22 |
mikeling | wiking: ^ | 07:22 |
-!- rads [3a92880d@gateway/web/freenode/ip.58.146.136.13] has quit [Ping timeout: 260 seconds] | 07:30 | |
-!- rushali [1b7c2602@gateway/web/freenode/ip.27.124.38.2] has joined #shogun | 08:07 | |
rushali | hey, i am new here ! does shogun have an OCR module ? | 08:08 |
-!- lambday|work [c40f170b@gateway/web/freenode/ip.196.15.23.11] has joined #shogun | 08:15 | |
-!- mode/#shogun [+o lambday|work] by ChanServ | 08:15 | |
-!- micmn [uid216702@gateway/web/irccloud.com/x-ejkkdlrwrsyfsjpe] has joined #shogun | 08:36 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 08:38 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 08:43 | |
-!- rushali [1b7c2602@gateway/web/freenode/ip.27.124.38.2] has quit [Ping timeout: 260 seconds] | 08:52 | |
-!- slayerjain [uid117577@gateway/web/irccloud.com/x-rkafcakpjapscpps] has joined #shogun | 08:59 | |
slayerjain | Hi folks! | 08:59 |
@wiking | mikeling, just a sec | 09:00 |
@wiking | mikeling, i need some time :) | 09:00 |
@wiking | slayerjain, hi ! | 09:00 |
slayerjain | I'm a machine learning enthusiast, but haven't really used a lot of shogun toolbox though. Wanted to know the status of gsoc here :) | 09:01 |
@wiking | what exactly would like to know regarding gsoc? | 09:01 |
@sukey | Pull Request #3674 "Use weight vector *only* through getters/setters" merged by lisitsyn - https://github.com/shogun-toolbox/shogun/pull/3674 | 09:12 |
@sukey | New Commit "Merge pull request #3674 from shogun-toolbox/feature/private-wector | 09:12 |
@sukey | Use weight vector *only* through getters/setters" to shogun-toolbox/shogun by lisitsyn: https://github.com/shogun-toolbox/shogun/commit/0582d1518708e101ccf653885046c81f2b09959a | 09:12 |
lisitsyn | wiking: honey | 09:37 |
lisitsyn | how are you feeling today | 09:37 |
lisitsyn | :D | 09:37 |
@wiking | ? | 09:37 |
@wiking | too much stuff | 09:37 |
lisitsyn | wiking: did you get microservices hangover? | 09:37 |
lisitsyn | :P | 09:37 |
lisitsyn | okok just do it | 09:37 |
@wiking | mikeling, i dont have an answer to you... you need to read the manuals and find it out yourself | 09:38 |
@wiking | and try which one actually works | 09:38 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-rncdiloslifjixkx] has quit [] | 09:46 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-ifvusntgxvzlnccg] has joined #shogun | 09:50 | |
slayerjain | Deep learning fascinates me because all the jazzy stuff its used on. I love ML and would like to work on something related to it. | 10:03 |
slayerjain | @wiking sorry for the late reply. got into something.. | 10:04 |
slayerjain | just checked the Wiki page, project "fundamental usual aspects" looks good to me! :D | 10:05 |
slayerjain | i think i'll get started with the entrance issues :) | 10:07 |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has joined #shogun | 10:10 | |
travis-ci | it's Sergey Lisitsyn'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/214152806 | 10:10 |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has left #shogun [] | 10:10 | |
slayerjain | actually "Applying Shogun to the real world" looks a lot more fun!! | 10:11 |
slayerjain | can someone please tell me the entrance issues for "Applying Shogun to the real world" project? | 10:12 |
-!- nagdev [dce39546@gateway/web/freenode/ip.220.227.149.70] has quit [Quit: Page closed] | 10:28 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 10:39 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 10:43 | |
-!- abhinav_ [~abhinav@103.211.18.27] has joined #shogun | 10:51 | |
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun | 11:00 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 11:00 | |
CaBa | jojo | 11:00 |
-!- abhinav_ [~abhinav@103.211.18.27] has quit [Ping timeout: 260 seconds] | 11:02 | |
-!- ptizoom [~user@79-70-50-227.dynamic.dsl.as9105.com] has joined #shogun | 11:11 | |
@sukey | Wiki page: GSoC_2015_project_fundamental edited on shogun-toolbox/shogun by karlnapf | 11:12 |
@sukey | Wiki page: GSoC_2015_project_fundamental edited on shogun-toolbox/shogun by karlnapf | 11:13 |
lisitsyn | HeikoS: ok I confirm references are wrong in meta cpp | 11:13 |
slayerjain | Can anyone help me with this? (I'm on macOS) https://www.irccloud.com/pastebin/KdAPgE3O/ | 11:15 |
@HeikoS | lisitsyn: details? | 11:15 |
@HeikoS | slayerjain: that is an outdated compiler I think | 11:15 |
slayerjain | i'll update it then.. | 11:16 |
lisitsyn | HeikoS: well rc is apparently wrong | 11:17 |
lisitsyn | it is 2 for features on exit | 11:17 |
lisitsyn | then 1 | 11:17 |
lisitsyn | but not deleted | 11:17 |
@HeikoS | ah I see | 11:18 |
@HeikoS | you checked the listings? | 11:18 |
@HeikoS | it has to do with wrap stuff | 11:18 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 11:20 | |
@lambday|work | xD | 11:20 |
@lambday|work | wiking: there? | 11:20 |
@lambday|work | wiking: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/compiler/xla/array3d.h#L55 | 11:21 |
@lambday|work | we can't pass std::vector there man.. even if you manage to convert our memory to std::initializer_list, it will still alloc memory | 11:22 |
@lambday|work | sorry I gave you the wrong file.. here it is : https://github.com/tensorflow/tensorflow/blob/master/tensorflow/compiler/xla/array2d.h#L57 | 11:22 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 11:24 | |
slayerjain | HeikoS: updated the compiler. same results! :( | 11:24 |
@HeikoS | slayerjain: try googling the error messages | 11:24 |
@HeikoS | e.g. the one about the template double >> | 11:25 |
@HeikoS | that your compiler wants to be a > > (with space) | 11:25 |
slayerjain | all i found online was, adding "-std=c++11" | 11:26 |
slayerjain | then i get this nasty error log https://www.irccloud.com/pastebin/u1R4C7xj/ | 11:27 |
-!- ptizoom [~user@79-70-50-227.dynamic.dsl.as9105.com] has quit [Remote host closed the connection] | 11:27 | |
@HeikoS | that seems like a setup issue of your compiler | 11:28 |
@HeikoS | slayerjain: if you enable testing, does make build the examples correctly? | 11:28 |
slayerjain | yes | 11:29 |
@HeikoS | i.e. make build_cpp_meta_examples | 11:29 |
slayerjain | the examples build correctly | 11:29 |
slayerjain | that's what doesn't make sense! | 11:29 |
@HeikoS | it does | 11:29 |
@HeikoS | the detected compiler and libc is porbably different to the one you use when you build manually | 11:29 |
slayerjain | oh! | 11:30 |
@HeikoS | check the CMakeCache.txt output to check which compiler cmake deted | 11:30 |
slayerjain | on it! | 11:30 |
@HeikoS | clang probably | 11:30 |
slayerjain | yeah! | 11:30 |
slayerjain | 'AppleClang 8.0.0.8000042' to be precise | 11:31 |
@HeikoS | slayerjain: so there you go | 11:31 |
@HeikoS | you are trying to compile with gcc there | 11:31 |
@HeikoS | use the same compiler and it should work | 11:31 |
@HeikoS | (feel free to send a PR that adds a sentence on that in the readme you followed, helpful!) | 11:31 |
slayerjain | definitely! :) | 11:31 |
slayerjain | thanks a lot! :D | 11:32 |
-!- ptizoom [~user@79-70-50-227.dynamic.dsl.as9105.com] has joined #shogun | 11:32 | |
slayerjain | HeikoS: similar error with clang! :( https://www.irccloud.com/pastebin/v5TcGoLA/ | 11:36 |
@HeikoS | slayerjain: you have to make sure the compiler matches the one detected by cmake | 11:37 |
CaBa | HeikoS: heya | 11:37 |
@HeikoS | CaBa: jojo | 11:37 |
CaBa | HeikoS: cmakelists has changed since branch? | 11:37 |
slayerjain | HeikoS: let me check.. | 11:38 |
@HeikoS | CaBa: check the diff of your PR | 11:38 |
slayerjain | HeikoS: they are the same. https://www.irccloud.com/pastebin/nZGMenuV/ | 11:39 |
CaBa | HeikoS: yes. i renamed the autogenerated test to "..._generated" and used the original name for the new test, as we discussed | 11:39 |
@HeikoS | slayerjain: your LD path might cause problems as well | 11:39 |
@HeikoS | CaBa: ah this was on purpose? | 11:40 |
CaBa | HeikoS: yes | 11:40 |
@HeikoS | lisitsyn: you still there? | 11:40 |
@HeikoS | CaBa: ok let me check again then... | 11:40 |
@HeikoS | CaBa: for the future ... rather send multiple PRs if things are not related | 11:40 |
@HeikoS | lisitsyn: https://github.com/shogun-toolbox/shogun/pull/3734/files#diff-c0c82dce2edf38481495103b1ecc4174R303 | 11:40 |
@HeikoS | lisitsyn: this is nuts! Storing the variables locally, obscures the code!!! | 11:40 |
slayerjain | HeikoS: this looks good? https://www.irccloud.com/pastebin/RJZAbUnh/ | 11:40 |
slayerjain | i set it manually as per the installation instructions | 11:41 |
@HeikoS | CaBa: I didnt see a sep commit for the CMakeLists.txt change ... | 11:42 |
@HeikoS | aaaah | 11:42 |
@HeikoS | sorry | 11:42 |
@HeikoS | remember now | 11:42 |
CaBa | HeikoS: https://github.com/shogun-toolbox/shogun/pull/3721/commits/2d532ddcfc54d86c2f41acdc2cf775cd9d5b8275 | 11:42 |
@HeikoS | too many things | 11:42 |
CaBa | HeikoS: it's the unit test commit | 11:42 |
CaBa | HeikoS: renaming the jinja file was not enough, had to change cmakelists | 11:43 |
@HeikoS | yeah | 11:43 |
@HeikoS | ok then | 11:43 |
@HeikoS | finally** | 11:43 |
@HeikoS | :) | 11:43 |
CaBa | \o/ | 11:43 |
@sukey | Pull Request #3721 "Alternative fix for Dynamic(Object)Array::clone()" merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3721 | 11:43 |
@sukey | New Commit "Merge pull request #3721 from lkuchenb/fixDynObjArrayClone2 | 11:43 |
@sukey | Alternative fix for Dynamic(Object)Array::clone()" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/8884cf78699a292f52e3c274cc132d8e6835b31c | 11:43 |
@HeikoS | slayerjain: you have some weird configuration issues, this is not shogun related. Just make sure that you are using the same c++ environment as cmake detected when compiling things by hand. You might have to alter the readme instructions in that case. Also ,you could just use cmake to build your example and not bother doing it manually (I do that) | 11:45 |
CaBa | HeikoS: let's clean up some small stuff when we're at it. what about #3689 - i'll remove the comment now, i suggest du leave the signature consistent with the other class though | 11:45 |
lisitsyn | HeikoS: back | 11:46 |
lisitsyn | sup | 11:46 |
lisitsyn | HeikoS: yes I did check ref_counts | 11:46 |
@HeikoS | lisitsyn: see comment, this storing of the get() variables | 11:46 |
@HeikoS | thats weird | 11:46 |
@HeikoS | making local copies | 11:46 |
lisitsyn | which comment sorry? | 11:46 |
@sukey | Pull Request #3689 "Add CLabels::get_subset_stack()" synchronized by lkuchenb - https://github.com/shogun-toolbox/shogun/pull/3689 | 11:46 |
lisitsyn | ah | 11:46 |
lisitsyn | HeikoS: yeah why to do that? | 11:47 |
@HeikoS | CaBa: const definitely, no? | 11:47 |
@HeikoS | lisitsyn: so doing multiple get() calls might be slow no? | 11:47 |
@HeikoS | lisitsyn: so in some PRs people started putting them in local variables | 11:48 |
@HeikoS | m_member_name_value | 11:48 |
@HeikoS | and then use tha | 11:48 |
slayerjain | HeikoS: sounds good to me.. | 11:48 |
@HeikoS | but thats weird style | 11:48 |
lisitsyn | HeikoS: slow? | 11:48 |
@HeikoS | CaBa: yeah ok leave the signature | 11:48 |
lisitsyn | why? | 11:48 |
@HeikoS | lisitsyn: well the get is not exactly returning a const reference | 11:49 |
lisitsyn | HeikoS: come on it is just a function call | 11:49 |
lisitsyn | :) | 11:49 |
@HeikoS | lisitsyn: https://github.com/shogun-toolbox/shogun/pull/3734/files#diff-c0c82dce2edf38481495103b1ecc4174R303 | 11:50 |
@HeikoS | this style is weird anyways | 11:50 |
lisitsyn | HeikoS: max_l1_norm -> get_max_l1_norm()? | 11:51 |
lisitsyn | what's weird? :) | 11:51 |
@HeikoS | no that is is stored in a variable | 11:51 |
@HeikoS | and then the variable is used | 11:51 |
@HeikoS | in the places below | 11:52 |
lisitsyn | HeikoS: well it is not what I suggested :) | 11:52 |
@HeikoS | lisitsyn: ok then, pls comment in that case | 11:52 |
lisitsyn | HeikoS: sure | 11:53 |
lisitsyn | HeikoS: I should reduce your function call anxiety | 11:53 |
lisitsyn | some massage may be | 11:53 |
lisitsyn | :D :D | 11:53 |
@HeikoS | just concerned if there is a loop that should move fast and is using a member a lot | 11:54 |
lisitsyn | HeikoS: psshh relax it is compiler job | 11:54 |
lisitsyn | :P | 11:54 |
@HeikoS | yeah | 11:54 |
@HeikoS | but not for tags | 11:54 |
@HeikoS | nothing can be done there | 11:55 |
@HeikoS | unlike for inline get_w() const | 11:55 |
@HeikoS | w is needle in heap | 11:55 |
slayerjain | does shogun use GPUs for stuff like DNNs? | 11:55 |
@HeikoS | anyways | 11:55 |
lisitsyn | slayerjain: no | 11:55 |
@HeikoS | coffee break | 11:55 |
@HeikoS | slayerjain: no, and you are more than welcome to do something about that | 11:56 |
lisitsyn | HeikoS: yeah stop worrying and call the function ;P | 11:56 |
slayerjain | that's interesting! :D | 11:56 |
@HeikoS | slayerjain: it doesnt even use vectorized CPU operations | 11:56 |
@HeikoS | mostly loops | 11:56 |
lisitsyn | well indeed it uses vectorized | 11:56 |
lisitsyn | ;) | 11:56 |
@HeikoS | good thing would be to go over and use linalg | 11:56 |
@HeikoS | lisitsyn: nope, not everywhere | 11:56 |
lisitsyn | compilers do that nowadays | 11:56 |
@HeikoS | I saw some hand implemented convolutions | 11:56 |
lisitsyn | convolutions are tough | 11:56 |
slayerjain | oh | 11:56 |
lisitsyn | but for loop is vectorized quite often | 11:57 |
@HeikoS | in any case, lots of room for improvement | 11:57 |
slayerjain | i would love to implement convolution on GPUs | 11:57 |
@HeikoS | slayerjain: check the linalg framework | 11:57 |
slayerjain | cool | 11:57 |
@HeikoS | would be a cool first task to implement a convolution in linalg CPU and GPU and then make use of it in the dnn code | 11:57 |
@HeikoS | Ill be back in 15m | 11:58 |
slayerjain | cool | 11:58 |
slayerjain | cya | 11:58 |
lisitsyn | HeikoS: there is a thing called flame graph | 11:58 |
lisitsyn | we should integrate it | 11:58 |
lisitsyn | so you see most of the time is spent not in accessing some small bits | 11:58 |
-!- lambday|work [c40f170b@gateway/web/freenode/ip.196.15.23.11] has quit [Quit: Page closed] | 12:09 | |
@wiking | lisitsyn, :D | 12:21 |
CaBa | jo wiking | 12:21 |
lisitsyn | wiking: sup honey | 12:21 |
lisitsyn | :P | 12:21 |
@wiking | y0y0 | 12:21 |
@wiking | long day | 12:21 |
@wiking | and not even near to end | 12:21 |
@wiking | sup sup | 12:22 |
@wiking | :) | 12:22 |
@wiking | all's good | 12:22 |
@wiking | ? | 12:22 |
@wiking | we solved all problems | 12:22 |
@wiking | ?D: | 12:22 |
@wiking | :D | 12:22 |
lisitsyn | wiking: yes microservices | 12:22 |
@wiking | \o/ | 12:22 |
@wiking | btw i've used spdlog | 12:22 |
@wiking | for another project | 12:22 |
@wiking | quite nice | 12:22 |
@wiking | hahah yeah this is the best! | 12:23 |
@wiking | http://www.aaronbell.com/secret-colours-of-the-commodore-64/ | 12:23 |
CaBa | wiking: wanna do the honor on #3731? | 12:26 |
@sukey | Pull Request #3731 "LibSVM parameters" merged by vigsterkr - https://github.com/shogun-toolbox/shogun/pull/3731 | 12:27 |
@wiking | dumdidum | 12:27 |
@sukey | New Commit "Merge pull request #3731 from lkuchenb/libSVMParameters | 12:27 |
@sukey | LibSVM parameters" to shogun-toolbox/shogun by vigsterkr: https://github.com/shogun-toolbox/shogun/commit/f17638fe5124b63e2188859186f2367114395213 | 12:27 |
CaBa | ... and #3730 | 12:27 |
CaBa | ^^ | 12:27 |
@sukey | Pull Request #3730 "Register parameters for CMKL and override clone()" merged by vigsterkr - https://github.com/shogun-toolbox/shogun/pull/3730 | 12:29 |
@sukey | New Commit "Merge pull request #3730 from lkuchenb/addMKLParameters | 12:29 |
@sukey | Register parameters for CMKL and override clone()" to shogun-toolbox/shogun by vigsterkr: https://github.com/shogun-toolbox/shogun/commit/750d3800d2fb411425fca428fbd22acb208973ec | 12:29 |
@wiking | CaBa, you should teach sukey to do merges | 12:29 |
@wiking | sukey flp | 12:30 |
@wiking | sukey fli | 12:30 |
@wiking | sukey flip | 12:30 |
@sukey | (ノ ゜Д゜)ノ ︵ ┻━┻ | 12:30 |
@wiking | :) | 12:30 |
CaBa | O.o | 12:30 |
CaBa | beautiful. i think. | 12:30 |
@HeikoS | slayerjain: did you manage to compile? | 12:31 |
CaBa | whatsup with the buildbot btw? | 12:31 |
slayerjain | Manually - No | 12:31 |
slayerjain | Using cmake - yes | 12:32 |
@HeikoS | slayerjain: ok well that should be good enough, it is the whole point of cmake to not having to mess with compiler setups locally | 12:33 |
-!- abhinav_ [~abhinav@103.211.18.27] has joined #shogun | 12:34 | |
slayerjain | HeikoS: cool. Then that's how I'll work then:) | 12:35 |
@HeikoS | wiking: so about the revert | 12:38 |
@HeikoS | wiking: shall I just revert all the commits (conflicts since was modified since merge), or shall I just assert the num threads to 1? | 12:38 |
@HeikoS | wiking: I mean the code itself is OK, it is more like it needs fixes in the rest of the framework .... | 12:38 |
-!- travis-ci [~travis-ci@ec2-54-158-43-162.compute-1.amazonaws.com] has joined #shogun | 12:38 | |
travis-ci | it's Heiko Strathmann'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/214193265 | 12:38 |
-!- travis-ci [~travis-ci@ec2-54-158-43-162.compute-1.amazonaws.com] has left #shogun [] | 12:38 | |
@HeikoS | CaBa: you remember what the problem with the latest xvalidation was? | 12:48 |
@sukey | Pull Request #3742 "xvalidation back to one thread" opened by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3742 | 12:48 |
@sukey | Issue #3238 "cmake not turning off USE_HDF5" closed by karlnapf - https://github.com/shogun-toolbox/shogun/issues/3238 | 12:49 |
@wiking | HeikoS, since it's openmp | 12:57 |
@wiking | just comment out the pragma | 12:57 |
@wiking | and put there a todo | 12:57 |
@HeikoS | wiking: see the pr | 12:57 |
@wiking | yeah | 12:58 |
@wiking | that's why i'm saying | 12:58 |
@wiking | reverint | 12:58 |
@wiking | reversion | 12:58 |
@wiking | of reversion | 12:58 |
@wiking | of reversion | 12:58 |
@wiking | seems a bit extreme | 12:58 |
@HeikoS | thats what is is | 12:58 |
@wiking | ok then pl | 12:58 |
@wiking | plz | 12:58 |
@wiking | dont merge it | 12:58 |
@wiking | and i'll do it | 12:58 |
@wiking | thnx | 12:58 |
CaBa | well | 12:58 |
@HeikoS | sure I can change as well | 12:58 |
@HeikoS | let me do | 12:58 |
CaBa | one layer or probs was related to the cloning being broken in many classes. this we have cleaned up quite a bit now, but who knows... | 12:59 |
CaBa | the other layer are 3rd party libs that can't be used in parallel | 13:00 |
@wiking | and when bullshit hits the sealing | 13:00 |
@wiking | http://blog.seraum.com/compiled-javascript-with-nectarjs-can-be-faster-than-nodejs-and-c?hn | 13:00 |
@wiking | gcc -O7 :D | 13:00 |
@HeikoS | CaBa: wiking I think we need tests for all machines before we can merge that stuff | 13:00 |
CaBa | one of the 3rd party libs that can't be parallel is GLPK | 13:00 |
@HeikoS | since otherwise stuff will pop up | 13:01 |
@wiking | ? | 13:01 |
CaBa | so set_solver_type() stuff should be aware or parallel / not parallel somehow | 13:01 |
CaBa | so that the machine can pick another solver if parallel or stuff like that | 13:01 |
@wiking | i'm not so sure if i get it | 13:01 |
@wiking | what's the problem | 13:01 |
@wiking | with single threaded xval? | 13:01 |
CaBa | none that i know of | 13:02 |
@sukey | Issue #3743 "parallelize xvalidation " opened by karlnapf - https://github.com/shogun-toolbox/shogun/issues/3743 | 13:03 |
@sukey | Issue #3743 "parallelize xvalidation " karlnapf added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/3743 | 13:03 |
@sukey | Issue #3743 "parallelize xvalidation " karlnapf added label: "BUG" - https://github.com/shogun-toolbox/shogun/issues/3743 | 13:03 |
@sukey | Issue #3743 "parallelize xvalidation " karlnapf added label: "development tasks" - https://github.com/shogun-toolbox/shogun/issues/3743 | 13:03 |
@sukey | Pull Request #3742 "xvalidation back to one thread" synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3742 | 13:04 |
@sukey | Pull Request #3742 "xvalidation back to one thread" - https://github.com/shogun-toolbox/shogun/pull/3742 | 13:05 |
CaBa | why do you want to undo it all? just remove the pragma or even better lets add a manual switch for now with a doc that says that you have to take care that everything you use works in parallel :P | 13:05 |
@HeikoS | CaBa: no manual switch, we either offer it or not | 13:05 |
@HeikoS | would be good to have a test so we at least have a "whitelist" of algos that work with it | 13:05 |
CaBa | well races not the best friends of tests | 13:06 |
@HeikoS | CaBa: I know but thats how we detected stuff in the first place right? | 13:07 |
CaBa | what's with all those solvers btw? i mean we already know that CPLEX hasn't been used for ages. how about the others? maybe they should go on the review board first before one starts to test them all | 13:07 |
@HeikoS | CaBa: send a patch that removes it :D | 13:07 |
@HeikoS | or make an entrance task :D | 13:07 |
CaBa | not any time soon :P | 13:08 |
@HeikoS | CaBa: joking aside | 13:08 |
@HeikoS | you are right | 13:08 |
@HeikoS | old unmaintained code needs to go | 13:08 |
@HeikoS | especially the corner case research code from 10 years ago | 13:08 |
@HeikoS | we should only keep useful things that are well tested | 13:08 |
@HeikoS | so I agree | 13:08 |
@HeikoS | "whats with all those solvers?!?!?!?!" | 13:09 |
@HeikoS | wiking: updated, merge if you agree | 13:11 |
@wiking | bien | 13:11 |
@wiking | danke | 13:12 |
@wiking | mergen :) | 13:12 |
@wiking | tschuss pass | 13:12 |
@wiking | :D | 13:12 |
@sukey | Pull Request #3701 "use get and set in SVM class" merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3701 | 13:13 |
@sukey | New Commit "Merge pull request #3701 from MikeLing/use_getter_and_setter_in_SVM | 13:13 |
@sukey | use get and set in SVM class" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/c0e565237bb4f36131973cb78e8307bad7a5459b | 13:13 |
@HeikoS | wiking: ok mach es! !!! | 13:13 |
CaBa | o.O | 13:19 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 13:21 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 13:25 | |
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Quit: Leaving.] | 13:28 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 13:36 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 13:41 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 13:55 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 14:03 | |
CaBa | wiking: #3689 another button for you ;) | 14:10 |
@wiking | sukey, flip | 14:10 |
@sukey | (╯°Д°)╯︵ /(.□ . ) | 14:10 |
@wiking | dadada | 14:10 |
@wiking | no | 14:10 |
@wiking | why no | 14:10 |
@wiking | sukey, merge | 14:10 |
CaBa | nice try | 14:11 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 14:14 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 14:19 | |
@sukey | Pull Request #3734 "Use get and set in regression classes" synchronized by abhinavrai44 - https://github.com/shogun-toolbox/shogun/pull/3734 | 14:24 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 14:30 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 14:35 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 14:46 | |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has joined #shogun | 14:46 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 256 seconds] | 14:51 | |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has joined #shogun | 15:01 | |
travis-ci | it's Viktor Gal'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/214206692 | 15:01 |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has left #shogun [] | 15:01 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 15:02 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 15:06 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 15:23 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds] | 15:28 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 15:39 | |
CaBa | broadcast ping | 15:46 |
-!- HeikoS [~heiko@82-132-238-154.dab.02.net] has joined #shogun | 15:47 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 15:47 | |
-!- slayerjain [uid117577@gateway/web/irccloud.com/x-rkafcakpjapscpps] has quit [Quit: Connection closed for inactivity] | 16:13 | |
geektoni | ping HeikoS | 16:16 |
@HeikoS | geektoni: hi! | 16:17 |
@HeikoS | I am only half present | 16:17 |
@HeikoS | geektoni: whats up? | 16:17 |
geektoni | No problem :) I have a small question | 16:17 |
@HeikoS | sure shoot | 16:17 |
geektoni | I am currently working on a prototype that should replace SG_PROGRESS | 16:18 |
geektoni | can I open a PR to show you what I have in mind? even if it is not a complete solution? (I want to avoid spam mails) | 16:19 |
@HeikoS | geektoni: sure just say it is WIP | 16:20 |
@HeikoS | no worries, you can update it, just ask the question in the text | 16:20 |
CaBa | HeikoS: https://github.com/shogun-toolbox/shogun/pull/3689 push the button! :) | 16:20 |
* CaBa hat heute seinen persönlichen PR list aufräumtag... | 16:21 | |
@HeikoS | CaBa: kommt gleich | 16:21 |
geektoni | HeikoS: Ok! Got it. Thanks :) | 16:22 |
@HeikoS | cool! | 16:22 |
@sukey | Pull Request #3689 "Add CLabels::get_subset_stack()" merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3689 | 16:26 |
@sukey | New Commit "Merge pull request #3689 from lkuchenb/addLabelsGetSubsetStack | 16:26 |
@sukey | Add CLabels::get_subset_stack()" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/e57211ae3799f55498a7dbba4bc1889f1e8e6ca0 | 16:26 |
@sukey | Pull Request #3742 "xvalidation back to one thread" merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/3742 | 16:27 |
@sukey | New Commit "Merge pull request #3742 from karlnapf/feature/kill_that_guy | 16:27 |
@sukey | xvalidation back to one thread" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/604c29a30a29a73bdc115c3b16dd0d66970343f8 | 16:27 |
-!- slayerjain [uid117577@gateway/web/irccloud.com/x-lrozulivabznoigm] has joined #shogun | 16:29 | |
CaBa | so sad. | 16:31 |
-!- HeikoS [~heiko@82-132-238-154.dab.02.net] has quit [Remote host closed the connection] | 16:31 | |
-!- lambday [6a33104e@gateway/web/freenode/ip.106.51.16.78] has joined #shogun | 16:37 | |
-!- mode/#shogun [+o lambday] by ChanServ | 16:37 | |
@sukey | Issue #3048 "Improve Shogun PCA" karlnapf added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/3048 | 16:39 |
@sukey | Issue #3048 "Improve Shogun PCA" karlnapf added label: "performance improvement" - https://github.com/shogun-toolbox/shogun/issues/3048 | 16:39 |
@sukey | Wiki page: GSoC_2017_applications edited on shogun-toolbox/shogun by lgoetz | 16:39 |
@sukey | Issue #3744 "Classify core developers irc output" karlnapf added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/3744 | 16:43 |
@sukey | Issue #3744 "Classify core developers irc output" opened by karlnapf - https://github.com/shogun-toolbox/shogun/issues/3744 | 16:43 |
@sukey | Issue #3744 "Classify core developers irc output"- https://github.com/shogun-toolbox/shogun/issues/3744 | 16:44 |
@sukey | Wiki page: Google-Summer-of-Code-2017-Projects edited on shogun-toolbox/shogun by lgoetz | 16:48 |
-!- HeikoS [~heiko@eduroam-int-pat-8-31.ucl.ac.uk] has joined #shogun | 16:53 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 16:53 | |
-!- nadia [cffb672e@gateway/web/freenode/ip.207.251.103.46] has joined #shogun | 16:54 | |
-!- abhinav_ [~abhinav@103.211.18.27] has quit [Ping timeout: 240 seconds] | 16:57 | |
-!- abhinav_ [~abhinav@103.211.18.27] has joined #shogun | 17:00 | |
@lambday | HeikoS: yo | 17:08 |
@HeikoS | jojo | 17:08 |
@lambday | HeikoS: so... | 17:08 |
@HeikoS | so you mean operator always creates a new feature object? | 17:08 |
@lambday | yeah.. with the same data... with mask on | 17:08 |
@lambday | CFeaturesBuilder.withExistingFeatures(old_one).withMask(mask_indices).build(); | 17:10 |
@lambday | you never change the indices | 17:10 |
@lambday | once you're done constructing it | 17:10 |
@lambday | would cause an alloc of 3-4 ptrs | 17:11 |
@HeikoS | lambday: that seems fine no? | 17:11 |
-!- nadia [cffb672e@gateway/web/freenode/ip.207.251.103.46] has quit [Ping timeout: 260 seconds] | 17:11 | |
@lambday | yeah | 17:11 |
@HeikoS | lambday: some masks might be performance killers | 17:11 |
@lambday | HeikoS: but what about the data itself? | 17:11 |
@HeikoS | like subsetting dimensions | 17:11 |
@HeikoS | because then we need a loop for dot product inside | 17:12 |
@HeikoS | no more SIMD | 17:12 |
@HeikoS | so that operator e.g. should be able to take a "copy" argument | 17:12 |
@lambday | HeikoS: for that, one might want to copy the data to a new matrix.. | 17:12 |
@lambday | cause even now, we don't do SIMD for dimension masking | 17:12 |
@HeikoS | I like the masking otherwise | 17:12 |
@HeikoS | no more "add_subset" etc | 17:12 |
@HeikoS | lambday: what about thread safety? | 17:12 |
@lambday | HeikoS: yeah that's first step towards making it immutable | 17:12 |
@HeikoS | since that will be important for xvalidation | 17:13 |
@lambday | HeikoS: yeah.. so here's the main point | 17:13 |
@lambday | HeikoS: would you want a const ptr, or const ptr to const data? | 17:13 |
@HeikoS | not sure I follow | 17:13 |
@lambday | cause in the second case, you get thread safety for free | 17:13 |
@lambday | HeikoS: meaning, no under the hood changing the matrix itself | 17:13 |
@HeikoS | lambday: oh | 17:14 |
@HeikoS | yeah I mean the data itself should be fixed | 17:14 |
@lambday | HeikoS: yeah.. | 17:14 |
@HeikoS | it might be copied at some point and then that is masked again | 17:14 |
@lambday | we need to optimize these things properly | 17:14 |
@HeikoS | but in general, can be const data | 17:14 |
@lambday | HeikoS: yeah | 17:14 |
@HeikoS | so then every thread would have their own feature instances | 17:14 |
@HeikoS | that all point to the same data | 17:14 |
@HeikoS | but never can change it | 17:14 |
@HeikoS | so all good | 17:14 |
@lambday | HeikoS: yeah that's the idea | 17:15 |
-!- abhinav_ [~abhinav@103.211.18.27] has quit [Ping timeout: 240 seconds] | 17:15 | |
@HeikoS | and then we can replace the whole "subset" stuff | 17:15 |
@HeikoS | and also remove subsampling dimensions code | 17:15 |
@lambday | HeikoS: yeah - anything that modifies.. becomes an operator of some sort | 17:15 |
@lambday | cfeatures lean and clean | 17:15 |
@HeikoS | yep | 17:16 |
@HeikoS | lambday: I mean there are cases where you want to modify in place | 17:16 |
@HeikoS | lambday: what do we do about that? | 17:16 |
@HeikoS | zero mean features | 17:17 |
@lambday | HeikoS: we need to make sure that it is std::unique_ptr :) | 17:17 |
@HeikoS | can be done either: | 17:17 |
@HeikoS | change existing matrix | 17:17 |
@lambday | should not do when it is shared_ptr | 17:17 |
@HeikoS | compute mean and remove on the fly in dot (more slow) | 17:17 |
@HeikoS | or allocate new matrix and put it there | 17:17 |
@HeikoS | how are in place modifications handled in the framework? | 17:17 |
@lambday | umm thinking | 17:17 |
@HeikoS | like in xvalidation, in place would not be allowed | 17:18 |
@HeikoS | that can be done with "const" | 17:18 |
@HeikoS | but maybe in other applications it is allowed | 17:18 |
@HeikoS | but then thread safety is an issue | 17:18 |
@lambday | HeikoS: if the object is being referred from only one place (i.e. the owner of that instance wants to modify), it should be able to | 17:18 |
@lambday | but if the instance is shared, we simply shouldn't allow any modification | 17:19 |
@HeikoS | lambday: yeah, how do the others know that it was changed? | 17:19 |
@HeikoS | so if one wants the zero mean preprocessor in xvalidation | 17:19 |
@HeikoS | it would have to work on the fly | 17:19 |
@HeikoS | so we can offer an API for that | 17:19 |
@lambday | HeikoS: yeah, unless we implement an observer or something, it cannot know.. | 17:19 |
@lambday | so we shouldn't do that | 17:20 |
@HeikoS | doc will say: this might be slower than in place, but you *can* use it in xvalidation | 17:20 |
@HeikoS | mmh | 17:20 |
@HeikoS | but then | 17:20 |
@HeikoS | what if someone wants zero mean preprocessor (or kenrel normaliser) in xvalidation | 17:20 |
@HeikoS | these would have to copy? | 17:20 |
@HeikoS | I guess either copy or on the fly | 17:21 |
@HeikoS | but whats the API to choose that? | 17:21 |
@HeikoS | either I have enough memory to copy things, or not. How can I specify that? | 17:21 |
@lambday | umm.. wait.. if we ensure that we use const refs to refer these instances where thread safety is required, and use non const refs at other places, we can do that | 17:22 |
@HeikoS | sure | 17:22 |
@HeikoS | Actually in the case above | 17:23 |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has joined #shogun | 17:23 | |
travis-ci | it's Heiko Strathmann'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/214218807 | 17:23 |
-!- travis-ci [~travis-ci@ec2-54-196-243-107.compute-1.amazonaws.com] has left #shogun [] | 17:23 | |
@HeikoS | the user will just choose the operator to be either "copy" or "on-the-fly" before he passes it to xvalidation | 17:23 |
@HeikoS | mmh wait | 17:23 |
@HeikoS | not sure I really understand | 17:23 |
@HeikoS | so use case: | 17:23 |
@HeikoS | I have zero mean preprocessor | 17:23 |
@HeikoS | and I want to do xvalidation | 17:23 |
@HeikoS | how do I choose between on-the-fly and copy ? | 17:24 |
@lambday | you apply your preprocessor beforehand ... then start your xval | 17:24 |
@lambday | oh no | 17:24 |
@lambday | soryr | 17:24 |
@lambday | with different indices the zero mean will change | 17:24 |
@lambday | umm | 17:24 |
@lambday | okay so let's think about how are we handling this thing ATM | 17:26 |
@lambday | we can't really make this parallel as of now, right? | 17:26 |
@lambday | we can make it parallel, only if we choose to stay with on-the-fly | 17:27 |
@lambday | if we keep modifying the underlying data with every fold, it cannot be parallel | 17:27 |
@lambday | unless we copy the entire thng | 17:27 |
@lambday | so there is a trade-off | 17:27 |
@lambday | and there should be a config that specifies that | 17:28 |
@lambday | depending on user's choice, the algorithm then decides to do whatever | 17:28 |
@lambday | one idea regarding SIMD in dimension subset : we compute dot with normal vector then compute dot with the mask vector.. not sure whether that will be faster | 17:35 |
@lambday | but worth trying out | 17:35 |
@HeikoS | sorry was afk for some min | 17:35 |
@HeikoS | reading :) | 17:35 |
@HeikoS | lambday: sure there is a trade off | 17:36 |
@HeikoS | but user should decide | 17:36 |
@HeikoS | so how is that handled | 17:36 |
@HeikoS | since the subsetting operators are applied by xvalidation | 17:36 |
@HeikoS | but the preprocessor (which also is an operator btw!!!!) will be applied by the user | 17:36 |
@HeikoS | that makes me think | 17:36 |
@HeikoS | preprocessors should just be operators as well | 17:36 |
@HeikoS | aaha | 17:37 |
@HeikoS | problems more problems | 17:37 |
@HeikoS | preprocessor thing needs to be re-applied for every fold | 17:37 |
@lambday | exactly | 17:37 |
@HeikoS | so kind of separate API I guess? | 17:37 |
@HeikoS | it is more like a stack of operations that is applied to data before training | 17:38 |
@HeikoS | but it kind of also is part of what we are talking about | 17:38 |
@sukey | Pull Request #3745 "Add PRange class to replace old SG_PROGRESS (WIP)" opened by geektoni - https://github.com/shogun-toolbox/shogun/pull/3745 | 17:38 |
@lambday | yeah.. stack of operation is the bigger picture | 17:38 |
@lambday | would the user every be specifying those stacks? | 17:39 |
@lambday | or it is what our machines have already encoded | 17:39 |
@HeikoS | user wants to say | 17:39 |
@HeikoS | I want zero mean | 17:39 |
@HeikoS | and feature extraction | 17:39 |
@HeikoS | the other way around | 17:39 |
@HeikoS | and then xvalidation on top of that does the subsetting | 17:40 |
@HeikoS | in every fold the preprocessor needs to be applied | 17:40 |
@HeikoS | but it is actually part of training | 17:40 |
@HeikoS | before machine is trained, all preprocessors are applied | 17:40 |
@HeikoS | they might be operators as well | 17:40 |
@HeikoS | but they are applied at different point | 17:40 |
@lambday | HeikoS: how the user specifies this setup now? | 17:41 |
@lambday | do they apply the preprocessors on the features beforehand before feeding it to the machines? | 17:42 |
@lambday | I mean, how would that work in xval setup? | 17:42 |
@HeikoS | lambday: that already works | 17:43 |
@HeikoS | check the code | 17:43 |
@HeikoS | preprocessors are applied to features | 17:43 |
@HeikoS | and the machine train method applied them | 17:43 |
@HeikoS | it is a bit dirty and manual, but it works | 17:43 |
@HeikoS | lambday: I think we can discuss this later | 17:44 |
@HeikoS | it is kind of independent | 17:44 |
@HeikoS | let us just talk about masking data | 17:44 |
@HeikoS | the preprocessors are applied after that ... | 17:44 |
@HeikoS | the only thing is | 17:44 |
@HeikoS | some of them will modify the data | 17:45 |
@HeikoS | so they wont be possible to be used when we switch to read only data | 17:45 |
@HeikoS | but we can figure that out then | 17:45 |
@HeikoS | (and clean up as well) | 17:45 |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has quit [Ping timeout: 240 seconds] | 17:51 | |
@lambday | those non const methods should be private and only accessible by friends | 17:51 |
@lambday | or maybe there are better ways | 17:51 |
@lambday | but none of the other classes should directly just get the private data and do whatever they want | 17:52 |
@lambday | I mean, private is not the right accessor then | 17:52 |
@lambday | maybe some other nicer design pattern | 17:52 |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has joined #shogun | 17:53 | |
@HeikoS | geektoni: saw your patch | 17:53 |
@HeikoS | geektoni: so whats the question? | 17:54 |
@HeikoS | lambday: the preprocessors will have to deal with constness | 17:54 |
geektoni | HeikoS: which patch? The new one? | 17:54 |
@HeikoS | lambday: and either copy, or do on-the fly (as specified by the user when instantiating the thing) | 17:54 |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has quit [Client Quit] | 17:57 | |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has joined #shogun | 17:57 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection] | 18:08 | |
geektoni | HeikoS: My question was related to the fact that I didn't know if I could open a PR with "unfinished" code. | 18:10 |
@lambday | HeikoS: hey.. let's chalk some actual xval python code that the user might want to use | 18:13 |
@lambday | rest of the design will follow the lead | 18:13 |
@lambday | let's put in all the complexities so that we cover all the different use-cases and keep those in mind while designing | 18:14 |
@lambday | do we have any meta example for that | 18:15 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 18:18 | |
@HeikoS | lambday: nope | 18:30 |
@HeikoS | but no need to have executable | 18:30 |
-!- ptizoom [~user@79-70-50-227.dynamic.dsl.as9105.com] has quit [Ping timeout: 260 seconds] | 18:33 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-ifvusntgxvzlnccg] has quit [Quit: Connection closed for inactivity] | 18:40 | |
@HeikoS | geektoni: pretty cool syntax ! :) | 18:41 |
geektoni | HeikoS: I was inspired mostly by tqmd. They have a very cool progress bar ;) | 18:42 |
@HeikoS | geektoni: you think we can bind those two together somehow? | 18:43 |
@HeikoS | so that if you use shogun from python, and have the progress activated, it prints it? | 18:43 |
@HeikoS | or is that too much magic? :D | 18:43 |
@HeikoS | or in other words | 18:43 |
@HeikoS | if you activate, how does it look when shogun is called from python? | 18:43 |
geektoni | HeikoS: Mmh I need to investigate a bit | 18:44 |
geektoni | because we have SWING that do the conversion from C++ to Python right? | 18:45 |
@HeikoS | yeah | 18:45 |
@HeikoS | but shogun stuff is printed in python anyways | 18:45 |
@HeikoS | so good idea to check that stuff | 18:45 |
@HeikoS | just to see what happens | 18:45 |
@HeikoS | dont think swig gets involved | 18:45 |
@HeikoS | as shogun prints, not python | 18:45 |
@HeikoS | _modshogun.so does it | 18:46 |
geektoni | I will do some testing later and will comment the PR with the results. | 18:48 |
geektoni | By the way, what do you mean by use cases? Since the PR contains only the "skeleton" of the progress bar there is no working code. | 18:49 |
geektoni | I mean, I do have the progress bar code, but it is not integrated with the PR's code. | 18:49 |
@HeikoS | geektoni: sure, I mean more your vision how it is used and what it produces | 18:49 |
@HeikoS | just like "this is what I am aiminmg for" | 18:50 |
geektoni | got it. | 18:50 |
geektoni | A bunch of gif is coming | 18:50 |
geektoni | ;) | 18:50 |
@HeikoS | hehe ;D | 18:51 |
-!- yashna [4cf7bd28@gateway/web/freenode/ip.76.247.189.40] has joined #shogun | 18:52 | |
-!- abhinav_ [~abhinav@103.211.18.27] has joined #shogun | 18:55 | |
geektoni | HeikoS: I've added a basic use case to the PR. | 18:57 |
geektoni | That's the idea. | 18:57 |
-!- travis-ci [~travis-ci@ec2-54-198-124-107.compute-1.amazonaws.com] has joined #shogun | 19:00 | |
travis-ci | it's Heiko Strathmann'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/214288829 | 19:00 |
-!- travis-ci [~travis-ci@ec2-54-198-124-107.compute-1.amazonaws.com] has left #shogun [] | 19:00 | |
@sukey | Pull Request #3745 "Add PRange class to replace old SG_PROGRESS (WIP)" synchronized by geektoni - https://github.com/shogun-toolbox/shogun/pull/3745 | 19:04 |
-!- geektoni [~geektoni@93-34-234-212.ip52.fastwebnet.it] has left #shogun [] | 19:08 | |
@HeikoS | looks great !!! | 19:09 |
-!- slayerjain [uid117577@gateway/web/irccloud.com/x-lrozulivabznoigm] has quit [Quit: Connection closed for inactivity] | 19:13 | |
@sukey | Pull Request #3719 "Implement QR solver in linalg library" synchronized by micmn - https://github.com/shogun-toolbox/shogun/pull/3719 | 19:35 |
-!- travis-ci [~travis-ci@ec2-54-234-171-120.compute-1.amazonaws.com] has joined #shogun | 19:49 | |
travis-ci | it's Heiko Strathmann'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/214288943 | 19:49 |
-!- travis-ci [~travis-ci@ec2-54-234-171-120.compute-1.amazonaws.com] has left #shogun [] | 19:49 | |
OXPHOS | HeikoS: that irc ouput issue is gorgeous | 20:14 |
@HeikoS | OXPHOS: it is :) | 20:14 |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection] | 20:15 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun | 20:24 | |
-!- HeikoS [~heiko@eduroam-int-pat-8-31.ucl.ac.uk] has quit [Ping timeout: 240 seconds] | 20:25 | |
-!- micmn [uid216702@gateway/web/irccloud.com/x-ejkkdlrwrsyfsjpe] has quit [Quit: Connection closed for inactivity] | 21:48 | |
-!- abhinav_ [~abhinav@103.211.18.27] has quit [Quit: Leaving] | 21:54 | |
-!- magicwund [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection] | 22:07 | |
@sukey | New Commit "add header numeric in PermutationMMD unit test" to shogun-toolbox/shogun by OXPHOS: https://github.com/shogun-toolbox/shogun/commit/75cde4fd1ca87ccf9c8c5a1152f76a38424ca9b8 | 22:18 |
@sukey | Pull Request #3705 "Add SGIO::progress unit tests and enhance SGIO::progress" merged by vigsterkr - https://github.com/shogun-toolbox/shogun/pull/3705 | 22:32 |
@sukey | New Commit "Merge pull request #3705 from geektoni/patch-13 | 22:32 |
@sukey | Add SGIO::progress unit tests and enhance SGIO::progress" to shogun-toolbox/shogun by vigsterkr: https://github.com/shogun-toolbox/shogun/commit/17d36a63376981cad3adc1baec652242ff8aebc6 | 22:32 |
--- Log closed Fri Mar 24 00:00:17 2017 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!