--- Log opened Sun Mar 16 00:00:40 2014 | ||
-!- grecocd [~Cassio@100.43.25.51] has joined #shogun | 00:13 | |
grecocd | Hi, I need help on something. On issue #1944 (https://github.com/shogun-toolbox/shogun/issues/1944), iglesias links an algorithm already implemented from another project (github.com/scikit-learn/scikit-learn/blob/master/sklearn/neighbors/kde.py) to be used as inspiration. Reading this code I see that it uses two different trees that can be used for the nearest neighbor algorithm, k-d tree and ball-tree. How can I implement the KDE with nearest neighbor withou | 00:20 |
---|---|---|
grecocd | these trees being implemented in shogun? | 00:20 |
-!- khalednasr [~k.nasr92@41.69.133.1] has quit [Quit: Leaving] | 00:35 | |
-!- grecocd [~Cassio@100.43.25.51] has left #shogun [] | 00:46 | |
-!- thoralf [~thoralf@91-66-32-99-dynip.superkabel.de] has quit [Quit: Konversation terminated!] | 02:02 | |
-!- kislay_ [~kislay@117.239.94.112] has joined #shogun | 02:54 | |
-!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has quit [Quit: PirosB3] | 03:28 | |
-!- kislay_ [~kislay@117.239.94.112] has quit [Remote host closed the connection] | 04:48 | |
-!- fs_ [~floriansc@f052185147.adsl.alicedsl.de] has quit [Ping timeout: 252 seconds] | 05:17 | |
-!- wavelander [~udit@122.177.168.86] has joined #shogun | 05:48 | |
-!- wavelander is now known as [paladin] | 05:48 | |
-!- kislay_ [~kislay@210.212.61.185] has joined #shogun | 05:56 | |
-!- vperic [~vperic@194.228.13.108] has quit [Ping timeout: 240 seconds] | 06:28 | |
-!- _flatline_ [~flatline@2607:f470:22:8:beae:c5ff:fe5a:903f] has quit [Ping timeout: 240 seconds] | 07:25 | |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has joined #shogun | 07:39 | |
-!- kislay_ [~kislay@210.212.61.185] has quit [Read error: Connection reset by peer] | 08:55 | |
-!- newlido [~quassel@105.200.8.132] has joined #shogun | 09:24 | |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has joined #shogun | 09:42 | |
-!- Rho [~rho@217.55.207.227] has joined #shogun | 09:59 | |
-!- kislay [~kislay@14.139.41.150] has joined #shogun | 10:00 | |
-!- Rho [~rho@217.55.207.227] has quit [Remote host closed the connection] | 10:21 | |
-!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has joined #shogun | 10:28 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 10:28 | |
-!- lambday [67157d54@gateway/web/freenode/ip.103.21.125.84] has joined #shogun | 10:34 | |
-!- Saurabh7 [~Saurabh7@59.88.33.201] has joined #shogun | 10:36 | |
-!- newlido [~quassel@105.200.8.132] has quit [Ping timeout: 252 seconds] | 10:39 | |
-!- newlido [~quassel@105.90.207.185] has joined #shogun | 10:41 | |
-!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has joined #shogun | 10:49 | |
PirosB3 | wiking: Hi! did you have a look at the SaaS mailing list? | 10:55 |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has quit [Ping timeout: 245 seconds] | 11:05 | |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has joined #shogun | 11:16 | |
-!- cameo54321_ [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has joined #shogun | 11:19 | |
-!- cameo54321_ [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has quit [Client Quit] | 11:20 | |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has quit [Ping timeout: 245 seconds] | 11:20 | |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has joined #shogun | 11:22 | |
-!- dhruv13 [~dhruv13@84.64.13.134] has joined #shogun | 11:36 | |
-!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has quit [Quit: Leaving] | 11:38 | |
-!- cameo54321 [2729a8ff@gateway/web/freenode/ip.39.41.168.255] has quit [Ping timeout: 245 seconds] | 11:39 | |
-!- cameo54321 [272975de@gateway/web/freenode/ip.39.41.117.222] has joined #shogun | 11:44 | |
-!- cameo54321 [272975de@gateway/web/freenode/ip.39.41.117.222] has quit [Ping timeout: 245 seconds] | 11:50 | |
-!- cameo54321 [27298acb@gateway/web/freenode/ip.39.41.138.203] has joined #shogun | 11:55 | |
@wiking | PirosB3: saas mailing list? | 11:58 |
@wiking | no i have not subscribed to a saas mailing list | 11:59 |
PirosB3 | mailing list, where I have posted a thread regarding the SaaS | 11:59 |
PirosB3 | Mark has asked you a couple of questions | 11:59 |
-!- Rho [~rho@193.227.20.2] has joined #shogun | 12:02 | |
Rho | hi | 12:02 |
Rho | what is the initial contribution that i can do to work in this project "Dictionary Learning" in GSoC 2014 | 12:03 |
@wiking | PirosB3: mmm i'll try to answer them asap... but you must bare in mind that we all do this for free, i.e. we do have job that we have to do.... so be patient about reviews/answers, we all try our best but again our resources are not dedicated at this very moment to shogun | 12:03 |
@wiking | Rho: any | 12:03 |
@wiking | Rho: https://github.com/shogun-toolbox/shogun/issues?labels=entrance&page=1&state=open | 12:04 |
@wiking | here's like 94 tasks | 12:04 |
@wiking | choose any of them and try to solve it | 12:04 |
@wiking | it does not matter that it's not connected with your desired task | 12:04 |
Rho | thanks ;) | 12:04 |
@wiking | we just want you to get familiar with shogun's codebase and the developing cycle | 12:05 |
Rho | that sounds great thanks so much | 12:05 |
PirosB3 | wiking: I perfectly understand this, as my idea is not really part of the ideas list, it is my responsibility to make sure mentors agree with it and it is in line with everyone's general expectations. For this reason I am trying to get everyone to review my proposal. Anyways I'm skyping with Mark soon, and we will come back with a roadmap and deliverables | 12:10 |
@wiking | PirosB3: i cannot stress it enough | 12:13 |
@wiking | that my opinion is that you should try to solve more entrance issues | 12:13 |
@wiking | as it's all good that you have something sketched up with Mark | 12:14 |
@wiking | but Mark can be only an external mentor in this story and you really should first convince shogun developer team that you do have the right skills for this task. as in fact shogun developer team will decide whether this project will go or not... | 12:15 |
@wiking | and the only way to do this, i.e. convince the developer team is actually solve more entrance issues | 12:16 |
@wiking | i dont know if i made my self clear here | 12:16 |
PirosB3 | clear enough. I should focus more on entrance tasks | 12:19 |
@wiking | indeed | 12:20 |
@wiking | you should have ... | 12:20 |
@wiking | and you should | 12:20 |
@wiking | although i really like the whole way of how this task has been built up from the beginning | 12:21 |
@wiking | and it's great that we get an external mentor for this task and he is very entusiastic about the project as well | 12:21 |
@wiking | but still, we need to be able to see your skills | 12:21 |
@wiking | and the only way we can evaluate that is via entrance tasks | 12:21 |
-!- cameo54321 [27298acb@gateway/web/freenode/ip.39.41.138.203] has quit [Ping timeout: 245 seconds] | 12:22 | |
PirosB3 | I have done 1 task up till now, will be doing more | 12:25 |
-!- khmarehman [~khmarehma@39.47.64.216] has joined #shogun | 12:42 | |
-!- cameo54321 [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has joined #shogun | 12:45 | |
-!- vperic [~vperic@194.228.13.108] has joined #shogun | 13:19 | |
-!- Saurabh7 [~Saurabh7@59.88.33.201] has quit [Quit: Leaving] | 13:24 | |
-!- khmarehman_ [~khmarehma@39.47.182.70] has joined #shogun | 13:27 | |
-!- khmarehman [~khmarehma@39.47.64.216] has quit [Ping timeout: 240 seconds] | 13:29 | |
dhruv13 | lisitsyn: Hey! Are you free to answer a question regarding CKNN? | 13:41 |
lisitsyn | dhruv13: hey, yeah | 13:41 |
dhruv13 | lisitsyn: the train_machine(data) function sets the data and query set as the same (ie. lhs = rhs), why is that? | 13:43 |
dhruv13 | lisitsyn: shouldn't the user be free to choose the query set? | 13:43 |
lisitsyn | dhruv13: it is perfect ok to do that once you need to really do anything with the query set | 13:43 |
lisitsyn | but as training would need distances between training set and training set | 13:44 |
lisitsyn | lhs gets equal to rhs | 13:44 |
dhruv13 | lisitsyn: okay, so it's basically for convenience with the distance functions right? | 13:45 |
-!- HeikoS [~heiko@nat-184-43.internal.eduroam.ucl.ac.uk] has joined #shogun | 13:45 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 13:45 | |
lisitsyn | not really about convenience | 13:45 |
lisitsyn | we'd have to have lhs==rhs when we train | 13:45 |
dhruv13 | lisitsyn: but there is no (apparent) way to change the query set, since the distance member is protected | 13:45 |
lisitsyn | you can get distance with getter though :) | 13:46 |
lisitsyn | dhruv13: when you do apply | 13:46 |
lisitsyn | it changes rhs | 13:46 |
lisitsyn | to the query set | 13:46 |
dhruv13 | lisitsyn: oh, sorry for the confusion! | 13:47 |
lisitsyn | dhruv13: no problem | 13:47 |
lisitsyn | I remember I had 'what the heck' thoughts on that before | 13:48 |
lisitsyn | :) | 13:48 |
dhruv13 | lisitsyn: totally missed that part!! :D | 13:48 |
dhruv13 | lisitsyn: ha ha | 13:49 |
dhruv13 | lisitsyn: a lot of the code makes sense now, especially the ASSERT(m_k<=num_labels) | 13:49 |
dhruv13 | lisitsyn: whew! | 13:50 |
lisitsyn | heh | 13:50 |
dhruv13 | lisitsyn: wait, line 167 here :https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/multiclass/KNN.cpp | 13:55 |
dhruv13 | lisitsyn: still doesn't make sense, shouldn't it be rhs? | 13:56 |
lisitsyn | dhruv13: that's correct as you find neighbors among the training set | 13:56 |
dhruv13 | lisitsyn: right! sorry again, i'm just panicked because i'm getting weird errors... | 13:57 |
lisitsyn | heh | 13:57 |
lisitsyn | no worries | 13:57 |
dhruv13 | lisitsyn: last clarification, line 340, m_k should be <= distance->get_num_vec_lhs() | 14:06 |
lisitsyn | but that's the same assert :) | 14:06 |
dhruv13 | it is right now <= num_rhs | 14:07 |
lisitsyn | ahh | 14:07 |
lisitsyn | looks like a bug | 14:07 |
dhruv13 | lisitsyn: yeah | 14:07 |
dhruv13 | lisitsyn: that was causing problems | 14:08 |
dhruv13 | lisitsyn: thanks a lot! | 14:08 |
lisitsyn | you're welcome | 14:08 |
-!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has quit [Quit: PirosB3] | 14:08 | |
dhruv13 | lisitsyn: oh, and classify_for_multiple_k() makes the assumption that init_data has already been called | 14:09 |
dhruv13 | lisitsyn: I should add that too right? | 14:09 |
lisitsyn | yeah probably | 14:09 |
lisitsyn | dhruv13: actually the best way is | 14:09 |
lisitsyn | to add a unit test | 14:10 |
lisitsyn | with possible failure and then fix the code to make it pass | 14:10 |
dhruv13 | lisitsyn: yep, on it... I'm currently working on nanoflann integration for kdtree support | 14:10 |
@HeikoS | cameo54321: hi | 14:12 |
@HeikoS | cameo54321: just reading your issue, sorry for the delay, was quite busy the last two days | 14:12 |
cameo54321 | HeikoS: Hi | 14:12 |
cameo54321 | HeikoS: ya didn't see you here | 14:13 |
cameo54321 | cameo54321: Thought you're busy | 14:13 |
@HeikoS | now commenting all my 50 emails :) | 14:13 |
cameo54321 | HeikoS: haha Good luck :-) | 14:14 |
@HeikoS | cameo54321: I will put something in your issue now, lets discuss then | 14:14 |
cameo54321 | HeikoS: Ya Sure.. Thanks.. Meanwhile I have corrected the GP Notebook.. | 14:14 |
lisitsyn | HeikoS: hey | 14:14 |
@HeikoS | cameo54321: nice! I will have alook soon | 14:15 |
@HeikoS | lisitsyn: hey! | 14:15 |
lisitsyn | I am a bit closer to the proof of concept for plugin stuff | 14:15 |
lisitsyn | let me show you | 14:15 |
cameo54321 | HeikoS: How to share? Github Gist? | 14:15 |
lisitsyn | HeikoS: https://github.com/lisitsyn/aer/blob/master/main.cpp | 14:16 |
-!- dhruv13_ [~dhruv13@84.64.15.29] has joined #shogun | 14:16 | |
@HeikoS | cameo54321: yeah, send a PR (with output removed from the notebook) and put a list to the nbviewer of your gist (with output) | 14:16 |
@HeikoS | lisitsyn: I will have a look in a bit, currently fightign through emails :) | 14:17 |
lisitsyn | sure | 14:17 |
-!- dhruv13 [~dhruv13@84.64.13.134] has quit [Ping timeout: 240 seconds] | 14:19 | |
@HeikoS | cameo54321: I updated the issue | 14:28 |
-!- Rho [~rho@193.227.20.2] has quit [Remote host closed the connection] | 14:28 | |
-!- PirosB3 [~pirosb3@162.219.6.191] has joined #shogun | 14:35 | |
jiaolong | Hi, I have written a new example, how to make shogun automatically compile my new cpp file? I placed it under shogun/examples/undocumented/linshogun/ | 14:42 |
@HeikoS | jiaolong: you have to add it to the makefile andor cmakefile | 14:49 |
jiaolong | OK, thank you, could you tell me which make/cmake file? | 14:52 |
@HeikoS | Ok use grep on an existing example name to find out, I dont know by heart | 14:52 |
jiaolong | OK, thanks, I will find it. :) | 14:52 |
-!- cameo54321 [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has quit [Ping timeout: 245 seconds] | 14:55 | |
-!- cameo54321 [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has joined #shogun | 14:56 | |
cameo54321 | HeikoS: Thanks, Seen.. | 14:57 |
@HeikoS | cameo54321: does it all make sense? | 14:57 |
@HeikoS | cameo54321: I am here now for a while so we could also discuss | 14:57 |
cameo54321 | HeikoS: Umm Yes.. | 15:00 |
cameo54321 | HeikoS: I need to think this a bit.. | 15:01 |
cameo54321 | HeikoS: You here for about half hour? | 15:01 |
@HeikoS | yes | 15:01 |
jiaolong | Hi, I have updated my code for the last PR, should I make a new PR? | 15:03 |
@HeikoS | jiaolong: no you can always update the current one via pushing to the branch that you sent the PR against | 15:03 |
jiaolong | OK, perfect. BTW, the cmake file to add example is just under the examples/undocumented/libshogun/ folder | 15:04 |
@HeikoS | yes | 15:04 |
jiaolong | Thanks a lot! | 15:05 |
@HeikoS | jiaolong: looking forward to your patch | 15:06 |
jiaolong | :) I have pushed my update | 15:07 |
jiaolong | How do I know Travis's result? It doesn't appear in my PR this time. | 15:08 |
@HeikoS | jiaolong: travis takes a while to trigger, depening on the number of PRs in the queue | 15:09 |
jiaolong | OK, thanks, hope it can pass the test! | 15:09 |
@HeikoS | jiaolong: always make sure to run all tests localy before sending a PR | 15:10 |
@HeikoS | jiaolong: that saves a lot of computation time | 15:10 |
jiaolong | Yes, all the local test are passed including memory leak check | 15:10 |
@HeikoS | jiaolong: nice, so lets see what travis sais | 15:11 |
-!- thoralf [~thoralf@91-66-32-99-dynip.superkabel.de] has joined #shogun | 15:14 | |
thoralf | Hey. | 15:14 |
@HeikoS | thoralf: hey! | 15:14 |
thoralf | Heiko is awake... seen that because I'm getting hundrets of mails/hour. :) | 15:14 |
@HeikoS | thoralf: hehe :) | 15:14 |
thoralf | hundreds | 15:14 |
@HeikoS | just working through my 2 days of not responding | 15:14 |
@HeikoS | 100+ mails in here | 15:14 |
@HeikoS | thoralf: just added a filter to put all of them into a sperate folder, could not see my own mails anymore ;) | 15:15 |
-!- dhruv13_ [~dhruv13@84.64.15.29] has quit [Ping timeout: 240 seconds] | 15:16 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 15:22 | |
shogun-notifier- | shogun: Sunil Mahendrakar :develop * 936bfc2 / src/shogun/mathematics/linalg/ratapprox/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/936bfc25ca0c36c2630bf61438c5ce54bd45840a | 15:22 |
shogun-notifier- | shogun: fixed compile error when eigen3 not installed | 15:22 |
shogun-notifier- | shogun: Heiko Strathmann :develop * 06e76ac / src/shogun/mathematics/linalg/ratapprox/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/06e76acd81880f985d5701c556c3a359c5192705 | 15:22 |
shogun-notifier- | shogun: Merge pull request #2008 from sunil1337/compile_error | 15:22 |
shogun-notifier- | shogun: | 15:22 |
shogun-notifier- | shogun: fixed compile error when eigen3 not installed | 15:22 |
thoralf | HeikoS: That was the first thing I did after beeing added to shogun-staff. ;) | 15:23 |
thoralf | Meanwhile I'm only watching threads I'm participating. | 15:24 |
@HeikoS | hehe :) | 15:24 |
thoralf | So add @tklein23 if you want me to see it. | 15:24 |
@HeikoS | thoralf: I do :) | 15:25 |
jiaolong | Sorry, I am learning ICR commands. :p | 15:33 |
cameo54321 | HeikoS: Ok so there are two base files in shogun at the moment | 15:33 |
cameo54321 | Legacy CDistribution | 15:34 |
cameo54321 | and Newer ProbabilityDistribution | 15:34 |
cameo54321 | What I understand is that I have to Write abstract base classes | 15:35 |
cameo54321 | which should provide interface points | 15:35 |
cameo54321 | for the distributions to expand on.. | 15:36 |
cameo54321 | Most of the CDistribution functions will be used, | 15:37 |
cameo54321 | But the structure will be inspired from stan | 15:37 |
cameo54321 | univariate, multivariate, both discreet and continuous. | 15:37 |
cameo54321 | and boost should/will be used for pdf/cdf functions | 15:38 |
cameo54321 | Sorry pdf/cdf functions will depend upon boost for computation.. | 15:39 |
thoralf | jiaolong: As far as I know, nobody is working on reading multilabels in LibSVM. | 15:40 |
thoralf | jiaolong: If you like to do it, place a comment on the issue. | 15:40 |
cameo54321 | for now, lets say I focus on univariate only, | 15:40 |
cameo54321 | for starters | 15:40 |
jiaolong | OK, perfect | 15:41 |
jiaolong | which ICR command do you use to reply to specific person in this channel? :p | 15:41 |
thoralf | jiaolong: Regarding your current PR - I'm waiting for hushell to do the review, since he's used to FactorGraph stuff. | 15:43 |
jiaolong | OK, no problem. | 15:43 |
thoralf | jiaolong: ICR? Internet Chat Relay? ;) | 15:43 |
jiaolong | thoralf: yes, freenode | 15:43 |
jiaolong | OK | 15:43 |
thoralf | Slightly different meaning. ;) | 15:43 |
cameo54321 | Then I'll be writing interface points considering stan's structure, but Cdistribution as guidance on what the interface points can be.. | 15:43 |
cameo54321 | HeikoS: Am I right? | 15:44 |
thoralf | IRC <- Internet Relay Chat, but ICR is a funny typo. | 15:44 |
jiaolong | Sorry. I got it. :) | 15:44 |
cameo54321 | HeikoS: I didn't mention your name cz I was collecting my own thoughts.. :p | 15:44 |
@HeikoS | cameo54321: yes | 15:44 |
@HeikoS | exactöy | 15:44 |
@HeikoS | ah wait you wrote more :) | 15:45 |
cameo54321 | HeikoS: I wrote a lot .. :pp | 15:45 |
@HeikoS | cameo54321: so the interface should be able to deal with all types | 15:45 |
@HeikoS | uni/multi discrete/continu | 15:45 |
@HeikoS | yeah and all those things that we cannot do on our own should come from external libs | 15:46 |
@HeikoS | cameo54321: I wonder whether we even need to distinguish between univariate and multivariate | 15:46 |
@HeikoS | cameo54321: there is also the kameleon-mcmc framework which doesnt do that | 15:46 |
@HeikoS | cameo54321: which has the advantage to have a common base class | 15:46 |
cameo54321 | HeikoS: That should be more generic | 15:47 |
@HeikoS | cameo54321: Also, the distributions base class should be defined on general CFeatures | 15:47 |
@HeikoS | not SGMatrix Vector | 15:47 |
@HeikoS | but the discrete continuous should maybe be subclasses | 15:47 |
@HeikoS | since continuous will all work on 64 bit real | 15:47 |
@HeikoS | and discrete might work on strings too | 15:48 |
@HeikoS | cameo54321: also, we dont need too many distributions for now, this is more about setting up the base classes | 15:48 |
@HeikoS | univariate: Gamma, beta | 15:48 |
@HeikoS | multivariate: normal only for now | 15:49 |
thoralf | HeikoS: Ready to merge? https://github.com/shogun-toolbox/shogun/pull/2011 | 15:49 |
-!- _flatline_ [~flatline@2607:f470:22:8:beae:c5ff:fe5a:903f] has joined #shogun | 15:49 | |
thoralf | HeikoS: Travis is :) | 15:49 |
@HeikoS | cameo54321: then we need a discrete with tables, and a mixture distribution to re-combine existing other distributions | 15:49 |
thoralf | HeikoS: Automatic fixing of imports is messing with everything that uses eigen3.h - probably eigen3.h itself is bogus. | 15:50 |
@HeikoS | cameo54321: then the Gaussian should have different options fors its covariance, like CGaussian (which should be merged) into CGaussianDistribution | 15:50 |
@HeikoS | thoralf: go ahead and merge | 15:50 |
@HeikoS | thoralf: yeah ok, thanks a lot for your effort on that! | 15:50 |
cameo54321 | HeikoS: We will not use SGMatrix? That's an importantpoint.. Can you please elaborate a little? | 15:50 |
shogun-notifier- | shogun: Thoralf Klein :develop * cf0db85 / tests/unit/ (67 files): https://github.com/shogun-toolbox/shogun/commit/cf0db854de23b8a071f4be3c0f9a2b8c8c32a9d9 | 15:51 |
shogun-notifier- | shogun: Cleaning unit tests imports: Removed unneeded #include, i.e. <init.h> | 15:51 |
shogun-notifier- | shogun: Thoralf Klein :develop * 7f7f4f2 / tests/unit/ensemble/ (2 files): https://github.com/shogun-toolbox/shogun/commit/7f7f4f2bfb1afe2c3c1f6eb1d17fd62ac1828464 | 15:51 |
shogun-notifier- | shogun: Cleaning unit tests imports: Don't include Labels, when only Math was wanted. | 15:51 |
shogun-notifier- | shogun: Thoralf Klein :develop * faa7b5c / tests/unit/ (2 files): https://github.com/shogun-toolbox/shogun/commit/faa7b5cc8e96cd7b1576701dbd0b06e944f2ca74 | 15:51 |
shogun-notifier- | shogun: Cleaning unit tests imports: Don't just include what's working. Include what's needed. | 15:51 |
shogun-notifier- | shogun: tklein23 :develop * f971d4f / tests/unit/ (69 files): https://github.com/shogun-toolbox/shogun/commit/f971d4f5d57d8bc5efc9d318846b67b227b2524c | 15:51 |
shogun-notifier- | shogun: Merge pull request #2011 from tklein23/clean_unittest_imports | 15:51 |
shogun-notifier- | shogun: | 15:51 |
shogun-notifier- | shogun: Clean unittest imports (removed obsolete imports as a first step) | 15:51 |
@HeikoS | cameo54321: we should rather do things on feature objects | 15:51 |
jiaolong | Hi, How to check the Travis error? e.g. https://travis-ci.org/shogun-toolbox/shogun/builds/20878522 | 15:51 |
@HeikoS | cameo54321: and the continous distributions can then have some additional methods which take matrices | 15:51 |
@HeikoS | cameo54321: but the log_pdf should be define on CFeatures | 15:51 |
@HeikoS | jiaolong: press on one of the builds | 15:52 |
@HeikoS | say the gcc one | 15:52 |
@HeikoS | https://travis-ci.org/shogun-toolbox/shogun/jobs/20878523 | 15:52 |
jiaolong | OK | 15:52 |
@HeikoS | you will see that the data version is not present | 15:52 |
-!- PirosB3 [~pirosb3@162.219.6.191] has quit [Quit: PirosB3] | 15:53 | |
jiaolong | Thanks, it seems I have to push the submodule | 15:54 |
@HeikoS | jiaolong: yeah you have to merge your PR against shogun-data first | 15:55 |
jiaolong | OK, I will try it | 15:55 |
-!- cameo54321_ [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has joined #shogun | 16:07 | |
-!- cameo54321 [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has quit [Ping timeout: 245 seconds] | 16:07 | |
-!- cameo54321_ [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has quit [Client Quit] | 16:10 | |
-!- besser82 [quassel@fedora/besser82] has quit [Ping timeout: 240 seconds] | 16:10 | |
-!- cameo54321 [2729c2c5@gateway/web/freenode/ip.39.41.194.197] has joined #shogun | 16:10 | |
-!- kislay [~kislay@14.139.41.150] has quit [Read error: Connection reset by peer] | 16:11 | |
-!- grzes [~grzes@host-81-190-236-173.wroclaw.mm.pl] has joined #shogun | 16:12 | |
@HeikoS | going for lunch now, be back in a little while | 16:13 |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has quit [Ping timeout: 252 seconds] | 16:14 | |
cameo54321 | HeikoS: Ok.. Me for dinner.. But seems a little discussion will be needed | 16:15 |
cameo54321 | afterwards | 16:16 |
-!- Saurabh7 [~Saurabh7@117.200.181.180] has joined #shogun | 16:25 | |
-!- besser82 [quassel@fedora/besser82] has joined #shogun | 16:26 | |
-!- mode/#shogun [+o besser82] by ChanServ | 16:26 | |
@HeikoS | cameo54321: let me know! | 16:42 |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has joined #shogun | 16:44 | |
lambday | HeikoS: hey.. | 16:46 |
@HeikoS | lambday: hey! | 16:50 |
@HeikoS | how are things? | 16:50 |
lambday | HeikoS: umm having doubts | 16:51 |
lambday | :( | 16:51 |
lambday | HeikoS: thought I could finish different number of samples issue... | 16:51 |
@HeikoS | lambday: but? | 16:51 |
@HeikoS | lambday: btw sorry for being not so responsive recently | 16:52 |
lambday | HeikoS: but what should I use as an estimator of the eigenvalues? | 16:52 |
@HeikoS | lambday: it turns out you have found a bug in the paper code, we hav ebeen discussing it :) | 16:52 |
@HeikoS | lambday: ah for the spectral test? | 16:52 |
lambday | HeikoS: yeah please check this [img]http://www.sciweavers.org/tex2img.php?eq=2%5Csum_l%5Cleft%28%5Cfrac%7B1%7D%7B2m%7D%5Cnu_l%5Cright%29%20z_l%5E2&bc=White&fc=Black&im=png&fs=12&ff=arev&edit=0[/img] | 16:53 |
lambday | oops | 16:53 |
lambday | not that | 16:53 |
lambday | HeikoS: https://github.com/shogun-toolbox/shogun/issues/1918 | 16:53 |
lambday | HeikoS: no worries - you took care of your things this weekend? | 16:53 |
lambday | I myself am going slow.. reading papers mostly | 16:53 |
shogun-buildbot_ | build #1886 of cyg1 - libshogun is complete: Failure [failed compile test] Build details are at http://buildbot.shogun-toolbox.org/builders/cyg1%20-%20libshogun/builds/1886 blamelist: tklein23 <tklein23@users.noreply.github.com>, Thoralf Klein <thoralf@fischlustig.de> | 16:58 |
@HeikoS | lambday: yeah lots of things to do, currently working on a poster | 16:59 |
@HeikoS | lambday: checking the link | 16:59 |
@HeikoS | lambday: it will be some weighted sum of things | 17:00 |
@HeikoS | lambday: I think dino once wroting something on that | 17:00 |
@HeikoS | but let me ask him :) | 17:00 |
@HeikoS | since I always forget this stuff | 17:00 |
lambday | hehe :D | 17:00 |
@HeikoS | lambday: I mean dont worry, that will be quite straight forward | 17:00 |
lambday | HeikoS: alright :) | 17:01 |
@HeikoS | lambday: just put in no weights for now, and wait until we get an answer | 17:01 |
@HeikoS | lambday: how are you feeling about all this stuff? | 17:01 |
lambday | HeikoS: great stuffs man! I am hoping I could gather enough understanding and use some of these in my thesis as well :) | 17:02 |
@HeikoS | lambday: that would be amazinh | 17:02 |
@HeikoS | lambday: you will see, the feature selection stuff works very nice! | 17:02 |
lambday | HeikoS: ah just waiting for that part... | 17:02 |
@HeikoS | lambday: yeah, it will take some time/work before that though :) | 17:03 |
lambday | HeikoS: I'm up for it :) | 17:03 |
@HeikoS | lambday: btw would you also like to guide/push the linear algebra API a bit? For moving dot-products, matrix-matrix products and factorizations within a shogun API, and then having different backends for those? | 17:03 |
@HeikoS | since you know that stuff very well | 17:03 |
@HeikoS | for example the eigen-solver class | 17:04 |
lambday | HeikoS: yeah that sounds very interesting - although I don't know any gpu stuffs.. | 17:04 |
@HeikoS | lambday: no need | 17:04 |
@HeikoS | lambday: first backend should be eigen3 | 17:05 |
@HeikoS | so nothing really changes | 17:05 |
lambday | but let me think of something - generalized front end | 17:05 |
@HeikoS | just having all those base classes | 17:05 |
@HeikoS | and then have different modes | 17:05 |
@HeikoS | the gpu stuff can be done by guys who know that, but we need the interface for that | 17:05 |
@HeikoS | lambday: and it would be so nice to have that before gsoc since then all new codes code be against that | 17:05 |
@HeikoS | lambday: so that switching to GPU once its time is just changing a class | 17:06 |
lambday | HeikoS: ah cool! I'll see if I cam come up with something useful | 17:06 |
lambday | checking the related issues | 17:06 |
@HeikoS | lambday: you dont have to code it up, just tell the people what to do :) | 17:06 |
lambday | hehe yeah :) | 17:07 |
lisitsyn | HeikoS: ! | 17:08 |
@HeikoS | lisitsyn: ? :D | 17:08 |
lambday | lisitsyn: happy birthday :P | 17:08 |
lisitsyn | HeikoS: why avoid opencl? ;) | 17:08 |
lisitsyn | lambday: thanks | 17:08 |
@HeikoS | lisitsyn: interface should be priority | 17:08 |
@HeikoS | lisitsyn: and eigen3 is easier to handle for backend for now | 17:08 |
@HeikoS | lisitsyn: once things work, have another backend | 17:08 |
lisitsyn | HeikoS: I myself need some explanation what do you mean by interface | 17:08 |
@HeikoS | lisitsyn: just a way of prioritising the work | 17:09 |
@HeikoS | lisitsyn: the base classes in math/linalg | 17:09 |
lisitsyn | dot(Any x, Any y) | 17:09 |
lambday | HeikoS: so, ideally everything related to linalg stuffs should be moved out from SGMatrix, right? like | 17:09 |
lisitsyn | :D | 17:09 |
lisitsyn | I mean what is the interface here | 17:09 |
@HeikoS | lisitsyn: no, please do proper representation here, lambday already did such great work woth those lienar operators etc | 17:09 |
@HeikoS | lisitsyn: way more generic and flexible, cleaner | 17:10 |
@HeikoS | lambday: yeah all math stuff should go into linalg | 17:10 |
@HeikoS | lisitsyn: and then offer wrapper methods maybe that are very simple | 17:10 |
lisitsyn | I don't really get what could be more generic for dot product than a function ;) | 17:10 |
@HeikoS | lisitsyn: for example matrices are linear operators that are applied to vectors | 17:10 |
@HeikoS | lisitsyn: should be represented like that | 17:10 |
@HeikoS | so that everything also covers cases where we dont want to do matrix dot products but say convolutions | 17:11 |
lisitsyn | in grothendieck group of abstract operators | 17:11 |
lisitsyn | :D | 17:11 |
@HeikoS | which is also a linear operation | 17:11 |
@HeikoS | lisitsyn: just internally, to the outside we can have whatever | 17:11 |
-!- newlido [~quassel@105.90.207.185] has quit [Ping timeout: 240 seconds] | 17:11 | |
lisitsyn | okay maybe I get it | 17:11 |
@HeikoS | lisitsyn: for take for example the eigenvector method in SGMatrix | 17:11 |
@HeikoS | this is not good, it just works for symmetric things | 17:12 |
@HeikoS | so this should be the self-adjoint linear operator | 17:12 |
@HeikoS | see what I mean? | 17:12 |
@HeikoS | just to structure things | 17:12 |
lisitsyn | probably | 17:12 |
@HeikoS | to avoid the usual mess we have | 17:12 |
-!- newlido [~quassel@105.90.207.185] has joined #shogun | 17:12 | |
@HeikoS | lisitsyn: you can also compute eigenstuff via svd | 17:12 |
@HeikoS | or EVD | 17:12 |
@HeikoS | for other matrice | 17:12 |
@HeikoS | s | 17:12 |
@HeikoS | or not, | 17:12 |
lisitsyn | HeikoS: have you got a chance to take a look at my proof of concept of that modularization? | 17:12 |
@HeikoS | if condition number is large, the usual stuff doesnt work | 17:12 |
@HeikoS | lisitsyn: not yet, sorry, but I will soon | 17:13 |
@HeikoS | lisitsyn: later today | 17:13 |
lisitsyn | alright | 17:13 |
lisitsyn | HeikoS: as you wanted it is crazy advanced C++ | 17:13 |
lisitsyn | :D | 17:13 |
@HeikoS | lisitsyn: I love that stuff, giveittomebaby | 17:14 |
lisitsyn | HeikoS: I have my own Any type :D | 17:14 |
lisitsyn | ah btw C++14 is here | 17:16 |
lisitsyn | so finally we got std::optional | 17:16 |
@HeikoS | lisitsyn: what does it bring? | 17:16 |
lisitsyn | it is kind of stabilization thing | 17:16 |
lisitsyn | not really full of new features | 17:16 |
lisitsyn | but some things like | 17:16 |
lisitsyn | auto type deduction for functions | 17:17 |
@HeikoS | cool | 17:17 |
@HeikoS | well, in steps we go | 17:17 |
lisitsyn | I am afraid some day C++ would be impossible to learn | 17:17 |
lisitsyn | :D | 17:17 |
@HeikoS | some day = 10 years ago :D | 17:18 |
thoralf | :D | 17:18 |
lambday | lisitsyn: you should start a tutorial class then :P | 17:18 |
lisitsyn | to be honest even with my love to C++ | 17:18 |
lisitsyn | I don't know quite a bunch of details | 17:18 |
@HeikoS | lisitsyn: same here | 17:18 |
lisitsyn | template <typename T, class... Args> | 17:19 |
lisitsyn | As<T> create(Args&&... args) | 17:19 |
lisitsyn | { | 17:19 |
lisitsyn | return As<T>(std::make_shared<T, Args...>(std::forward<Args>(args)...)); | 17:19 |
lisitsyn | } | 17:19 |
lisitsyn | like that | 17:19 |
lambday | holy | 17:19 |
lisitsyn | it took like 10-20 minutes to understand that for me | 17:19 |
@HeikoS | ah I wont try for now ;) | 17:19 |
lisitsyn | Args&&... args | 17:20 |
lisitsyn | :D | 17:20 |
lisitsyn | looks like somebody is cursing | 17:20 |
lambday | lisitsyn: HeikoS: btw how do you plan the plugin stuff? I mean, is it like - download basic shogun modules - then download (like firefox plugins) other modules that you need and not the whole thing | 17:20 |
lisitsyn | lambday: well I don't have real vision on that | 17:21 |
lambday | as in, most of the users that use string kernels won't require the log-det specific modules per se | 17:21 |
lisitsyn | what I am technically interested | 17:21 |
lisitsyn | is how to do that :D | 17:21 |
lisitsyn | what I have now is more or less good concept | 17:21 |
lisitsyn | how to define that | 17:21 |
lambday | sounds really cool! | 17:22 |
lisitsyn | due to C++ nature we'd have to keep base classes in core shogun | 17:22 |
@HeikoS | lisitsyn: reduce compile time | 17:22 |
lisitsyn | e.g. multiclass machine | 17:22 |
lisitsyn | etc | 17:22 |
@HeikoS | lisitsyn: I just want to compile a certain example | 17:22 |
@HeikoS | lisitsyn: it tells me which modules I need | 17:22 |
@HeikoS | that would be great | 17:23 |
lisitsyn | okay that core I am working on won't be fast to compile | 17:23 |
lisitsyn | due to heavy usage of templates | 17:23 |
@HeikoS | lisitsyn: still I dont need to compile all shogun right? | 17:23 |
lisitsyn | HeikoS: base classes | 17:23 |
lambday | lisitsyn: from all modules? | 17:23 |
lisitsyn | well | 17:24 |
lisitsyn | okay let me explain | 17:24 |
lisitsyn | say we are using python | 17:24 |
lisitsyn | and we want to call method 'apply' | 17:24 |
lisitsyn | this 'apply' method *have to* be defined in some class of core shogun | 17:24 |
lisitsyn | if you don't add new methods - you are ok to exclude it from base shogun | 17:25 |
lisitsyn | by new methods I mean real actions but not just getters/setters | 17:25 |
lisitsyn | if you add new parameters that's ok | 17:25 |
lisitsyn | I am 95% sure it is impossible otherwise :D | 17:27 |
lisitsyn | we are using wrong language to have something dynamic :D | 17:28 |
-!- cs_shadow [~cs_shadow@14.139.82.6] has joined #shogun | 17:28 | |
lambday | lisitsyn: how does mozilla do it for plugins, per se? I mean its written in c++, right? | 17:28 |
lambday | oh but not modular bindings | 17:29 |
-!- cs_shadow [~cs_shadow@14.139.82.6] has left #shogun [] | 17:29 | |
-!- cs_shadow [~cs_shadow@14.139.82.6] has joined #shogun | 17:29 | |
-!- cs_shadow [~cs_shadow@14.139.82.6] has left #shogun [] | 17:29 | |
lambday | I don't know how technically it can be done - just talking from a user's viewpt | 17:30 |
lisitsyn | lambday: well they have api | 17:30 |
lisitsyn | you just load dynamic library and use it | 17:30 |
lisitsyn | but you can't add class to any scope dynamically | 17:31 |
lambday | oh I see what you mean | 17:31 |
lisitsyn | so basically here in shogun we'd have to choose what is the API we expose | 17:31 |
lisitsyn | no plugin can alter this API | 17:32 |
lisitsyn | it seems I've found some solution for dynamic parameters | 17:33 |
lisitsyn | but not for methods | 17:34 |
-!- newlido [~quassel@105.90.207.185] has quit [Quit: No Ping reply in 180 seconds.] | 17:34 | |
lambday | lisitsyn: you need a c++ guru like you for this project :P | 17:36 |
lisitsyn | lambday: well nobody applied so far :D | 17:37 |
-!- newlido [~quassel@105.90.207.185] has joined #shogun | 17:37 | |
lambday | lisitsyn: it grows exponentially towards the deadline | 17:38 |
lisitsyn | lambday: you know we have a function like inverse of that | 17:38 |
lisitsyn | which tells you about chances :D | 17:39 |
lambday | lisitsyn: lol yeah but nobody started working on any issue related to that too? | 17:39 |
lisitsyn | there is no issue actually | 17:39 |
lambday | lisitsyn: you scared people off :P | 17:40 |
lisitsyn | exactly that's totally my fail | 17:40 |
lambday | lisitsyn: I was just kiddin' | 17:40 |
lisitsyn | lambday: I am not ;) | 17:40 |
lambday | lisitsyn: I hope to learn a bit regarding these cool stuffs at my job - hope I'll be of some use after a year | 17:41 |
lisitsyn | lambday: where will you work? | 17:41 |
lambday | lisitsyn: oracle | 17:42 |
lambday | any idea? | 17:42 |
lisitsyn | oracle is evil you know :D | 17:42 |
lisitsyn | no I don't have any insight on how do they work | 17:42 |
lambday | in server tech profile - so mostly c/c++ code | 17:42 |
lambday | lisitsyn: oracle is evil... so is MD | 17:43 |
lambday | MS*** | 17:43 |
lisitsyn | well you will learn some really industrial code then :) | 17:43 |
lambday | lisitsyn: I heard that they contribute stuffs to linux kernel as well.. have this open source team as well... | 17:44 |
lisitsyn | lambday: sure | 17:44 |
lambday | well, that must be in the US office... | 17:44 |
@wiking | oracle | 17:44 |
@wiking | evil! | 17:45 |
lisitsyn | haha | 17:45 |
lambday | poor me :P | 17:45 |
@wiking | lambday: why would u work for oracle?! | 17:45 |
@wiking | they are like worse than m$ :D | 17:45 |
lambday | wiking: umm.. for starters, money :( | 17:45 |
@wiking | lambday: yeah ... :< | 17:45 |
@wiking | run while u can :D | 17:45 |
lambday | wiking: no bond.... I can run any time I wanna | 17:46 |
@wiking | good | 17:46 |
@wiking | :) | 17:46 |
lambday | wiking: may be besser82 could get me into a RHEL job later this year :P (with relocation) | 17:46 |
lambday | wiking: lol you spelled m$ lol | 17:47 |
lisitsyn | haha so you are like in search of relocation positions | 17:47 |
lisitsyn | (like me) | 17:47 |
lambday | lisitsyn: I wanna travel :D | 17:47 |
lambday | lisitsyn: until I get married its like run forrest run | 17:48 |
lisitsyn | lambday: oh tell me | 17:48 |
@wiking | lambday: so the only option to travel is with oracle? :) | 17:48 |
lisitsyn | is that true that in india | 17:48 |
lisitsyn | father of any daughter have to get some huge amount of money | 17:48 |
lisitsyn | to afford crazy wedding | 17:48 |
lambday | wiking: no I got posted in oracle's india office | 17:48 |
lisitsyn | like for few hundred of people | 17:49 |
lambday | lisitsyn: unfortunately, that crazy shit (called dowry) still happens in places in India | 17:49 |
lisitsyn | haha | 17:49 |
lambday | lisitsyn: well, weddings here are almost like huge occassions | 17:50 |
lambday | crazy ceremony.. parents save for that day since the mother is pregnant | 17:50 |
lisitsyn | lambday: what's typical age when people get married? | 17:50 |
lambday | lisitsyn: mean 27, std-dev 3 | 17:51 |
lisitsyn | hmm I see | 17:51 |
lambday | I totally made that up btw | 17:51 |
lambday | but the numbers sounds pretty close as per my exp | 17:52 |
lisitsyn | cool | 17:52 |
lisitsyn | that's interesting actually | 17:52 |
lambday | lisitsyn: and painful... (noticing that I turned 27) | 17:53 |
lisitsyn | haha | 17:53 |
lambday | lisitsyn: why don't you apply for Google :) | 17:55 |
lisitsyn | lambday: why do you think I didn't ;) | 17:55 |
lambday | lisitsyn: ah no! | 17:55 |
lisitsyn | but well you see I am not in google :D | 17:55 |
Saurabh7 | Hi guys, | 17:58 |
lambday | hi Saurabh7 | 17:58 |
Saurabh7 | lambday, hi | 17:58 |
@wiking | lisitsyn: :DDDDDDDDDDDDDDDDDDDD | 17:58 |
Saurabh7 | so i was planning to apply for Shogun Missionary & Shogun in Education idea | 17:59 |
Saurabh7 | but i cant figure how to organize my work week by week | 17:59 |
@wiking | Saurabh7: try to make it | 18:02 |
@wiking | and then post it | 18:02 |
@wiking | and then we can give u feedback | 18:02 |
Saurabh7 | wiking, yes, but all i can think of is write notebooks,write demos...how would i structure it? | 18:05 |
Saurabh7 | maybe i should plan it based on the src directory | 18:05 |
Saurabh7 | like classifier, regressions,etc...does that sound okay? | 18:05 |
@wiking | Saurabh7: well first of all think about what sorts of notebooks would u like to write | 18:06 |
@wiking | and then start trying to categorize them | 18:06 |
Saurabh7 | wiking, hmm... okay i am thinking about looking at some fundamental algorithms which require some easing into | 18:07 |
@wiking | thoralf: please try to fix your git identity: Author: tklein23 <tklein23@users.noreply.github.com> | 18:07 |
Saurabh7 | just as the idea page suggests | 18:08 |
cameo54321 | Ok I can understand CFeatures... What are Efeatures? | 18:09 |
cameo54321 | Anybody please? | 18:10 |
cameo54321 | Feature.cpp | 18:10 |
cameo54321 | src/features | 18:10 |
thoralf | wiking: Locally it's fine. I think this happened because I said github to keep it private. | 18:12 |
@wiking | cameo54321: there is not Efeauters :) | 18:13 |
cameo54321 | wiking: Sorry I didn't compete... EFeatureProperty... | 18:14 |
cameo54321 | wiking: seems it is a feature type | 18:15 |
cameo54321 | enum EFeatureProperty { FP_NONE = 0, FP_DOT = 1, FP_STREAMING_DOT = 2 }; | 18:15 |
@wiking | yes as it's name says | 18:15 |
lisitsyn | it is some kind of enum to indicate capabilities of various features | 18:15 |
@wiking | it indicates the type of feature | 18:16 |
cameo54321 | wiking: for example what might FP_NONE mean? | 18:17 |
@wiking | that it has no special capability like FP_DOT ;) | 18:17 |
lisitsyn | FP_USELESS_SH*T | 18:17 |
lisitsyn | sorry :D | 18:18 |
@wiking | cameo54321: r u implementing a new feature type? | 18:18 |
@wiking | or what? | 18:18 |
cameo54321 | wiking: No, | 18:18 |
cameo54321 | I was trying to understand Features.cpp | 18:18 |
cameo54321 | Cfeatures | 18:18 |
@wiking | just out of curiousity | 18:19 |
@wiking | ? | 18:19 |
cameo54321 | As it seems we'll be using Cfeatures in probability distributions instead of SGVectors/SCMatrices | 18:19 |
cameo54321 | SGMatrix | 18:19 |
@wiking | mmm interesting | 18:20 |
@wiking | any reason for this bald move? | 18:20 |
cameo54321 | wiking: BTW did you see the Issue 1998? | 18:20 |
cameo54321 | https://github.com/shogun-toolbox/shogun/issues/1998 | 18:21 |
@wiking | yes what's with it? | 18:21 |
cameo54321 | Well the move resulted in my latest discussion with Heiko | 18:21 |
@wiking | cameo54321: a) i don't think it's a wise idea to create distributions that are only working if you have c++11 | 18:22 |
Saurabh7 | hey lisitsyn , what do you think about my pr for shogun-demo, is it mergeable? | 18:22 |
lisitsyn | Saurabh7: yeah I think so | 18:22 |
lisitsyn | wiking: elaborate this ;) | 18:22 |
@wiking | cameo54321: b) removing SFMT dependency is eeeeeven more balder move | 18:22 |
cameo54321 | HeikoS Sugestion is that we shaould implement the base classes using CFeatures | 18:22 |
@wiking | as we just added that lately | 18:22 |
-!- dhruv13 [~dhruv13@84.64.13.72] has joined #shogun | 18:22 | |
lisitsyn | wiking: come on lets forget about anything before C++11 | 18:22 |
@wiking | lisitsyn: :DDDDDD | 18:23 |
lisitsyn | wiking: really | 18:23 |
@wiking | lisitsyn: so s/auto/[a-zA-Z]*\*/g | 18:23 |
@wiking | right? | 18:23 |
@wiking | :) | 18:23 |
lisitsyn | ain't we have it mandatory already? | 18:23 |
lisitsyn | wiking: sure thing | 18:23 |
Saurabh7 | lisitsyn, ok I was pondering coz noone commented for changes | 18:24 |
cameo54321 | HeikoS: I just created some more trouble here :D Can you please look at it? :D | 18:24 |
@wiking | lisitsyn: well if that's the standpoint | 18:25 |
@wiking | then why the hell do we still have | 18:25 |
@wiking | SGReferencedData | 18:25 |
@wiking | and all that shitty wrapper stuff | 18:25 |
@wiking | instead of directly using std::shared_ptr | 18:25 |
@wiking | i mean dont get me wrong | 18:25 |
@wiking | i'm fine by it... just let's then make it right | 18:26 |
lisitsyn | wiking: yeah sure | 18:26 |
@wiking | cameo54321: i've just commented one more thing on your issue... | 18:26 |
@wiking | cameo54321: please before suggesting something like that... do some benchmarking | 18:26 |
@wiking | cameo54321: i reckon the c++11 generated random will be much slower than the dSFMT/SFMT | 18:26 |
@wiking | but then again i might be wrong | 18:26 |
@wiking | so before doing anything it would be worth to benchmark it | 18:27 |
@wiking | lisitsyn: so we should do it, but actually we cannot | 18:27 |
@wiking | because we still have a lot of scenarios | 18:27 |
@wiking | where we just cannot use c++11 | 18:27 |
@wiking | not because the compiler does not support it | 18:27 |
@wiking | it just because the combination of various libraries and stuff will crash | 18:28 |
cameo54321 | wiking: No intention of removing SFMT (or for that matter anything) at the moment... | 18:28 |
@wiking | cameo54321: "What if the classes for the Issue #1929 are written utilizing C++11s distributions in mind instead of (or alongwith) SFMT?" | 18:28 |
@wiking | cameo54321: so what is the meaning of this sentence then? | 18:28 |
-!- PirosB3 [~pirosb3@162.219.6.191] has joined #shogun | 18:28 | |
@wiking | lisitsyn: so i reckon for another year we still have to use #ifdef HAVE_CXX11_something... :S | 18:29 |
cameo54321 | HeikoS suggestion is to write new base classes, using CFeatures.. Iwent looking into that direction | 18:30 |
cameo54321 | wiking: as for the sentence, I asked this if this can be done or not.. No other thing | 18:30 |
-!- dhruv13_ [~dhruv13@84.64.14.115] has joined #shogun | 18:31 | |
@wiking | 18:27 < cameo54321> wiking: No intention of removing SFMT (or for that matter anything) at the moment... | 18:31 |
@wiking | 18:27 <@wiking> cameo54321: "What if the classes for the Issue #1929 are written utilizing C++11s distributions in mind instead of (or alongwith) SFMT?" | 18:31 |
cameo54321 | wiking: I do not know if SFMT will be good or bda | 18:31 |
cameo54321 | bad* | 18:31 |
@wiking | ? | 18:31 |
@wiking | what? | 18:31 |
@wiking | SFMT is currently the PRNG in whole shogun | 18:31 |
@wiking | you are at the moment talking about try to use c++11's prng | 18:32 |
@wiking | it's all good | 18:32 |
@wiking | but please do some benchmarking | 18:32 |
@wiking | before continue on that road | 18:32 |
@wiking | so that we see what's the difference between the current PRNG and c++11's prng | 18:32 |
cameo54321 | wiking: Ok.. By the way I am still looking for advice.. | 18:33 |
@wiking | advice on? | 18:33 |
@wiking | i've just given u one | 18:33 |
-!- PirosB3 [~pirosb3@162.219.6.191] has quit [Quit: PirosB3] | 18:33 | |
cameo54321 | wiking: as for how base classes should be written for unifying prob distributions | 18:33 |
@wiking | benchmark c++11 vs sfmt PRNG | 18:34 |
-!- dhruv13 [~dhruv13@84.64.13.72] has quit [Ping timeout: 240 seconds] | 18:34 | |
@wiking | cameo54321: what do u mean by how it should be written? | 18:34 |
@wiking | i guess it should be written in c++ | 18:34 |
@wiking | as well as with a text editor | 18:34 |
-!- tomy18 [~alex@p6.eregie.pub.ro] has joined #shogun | 18:34 | |
cameo54321 | Using SGMatrices or CFeatures? | 18:34 |
@wiking | or sgvector | 18:35 |
@wiking | ? | 18:35 |
cameo54321 | or sgvector | 18:35 |
@wiking | or sgsparsevector | 18:35 |
-!- _flatline_ [~flatline@2607:f470:22:8:beae:c5ff:fe5a:903f] has quit [Quit: Leaving] | 18:35 | |
@wiking | or sgsparsematrix? | 18:35 |
-!- khmarehman_ [~khmarehma@39.47.182.70] has quit [Read error: No route to host] | 18:38 | |
@wiking | wooohooo HeikoS is being brave again: "I would much rather depend on http://www.boost.org/doc/libs/1_55_0/libs/math/doc/html/dist.html which is mature, tested, etc. " | 18:38 |
@wiking | :D | 18:39 |
thoralf | wiking: I don't get if you like it or not. | 18:39 |
cameo54321 | wiking: Please do not take me wrong, but in which direction is it recommended should i move? | 18:40 |
@wiking | thoralf: which? :) | 18:40 |
@wiking | cameo54321: well i reckon you should study other libraries that actually do the same functionality | 18:40 |
thoralf | wiking: boost | 18:40 |
@wiking | thoralf: no-go | 18:40 |
-!- kislay [~kislay@14.139.41.150] has joined #shogun | 18:40 | |
thoralf | wiking: Really? | 18:40 |
@wiking | thoralf: we went through this like 10 times | 18:41 |
@wiking | thoralf: yes really :) | 18:41 |
@wiking | thoralf: but ask sonney2k ;))))) | 18:41 |
@wiking | cameo54321: first of all it would be a good idea that the whole thing is templated at least by the PRNG | 18:41 |
lisitsyn | haha boost | 18:41 |
jiaolong | Hi guys, if I want to PR to shogun-data, do I need to fork shogun-data first? I am using update submodule only currently. | 18:41 |
@wiking | thoralf: ^ see lisitsyn's reaction :D) | 18:41 |
@wiking | jiaolong: u need to do the same thing as in case of shogun | 18:42 |
lisitsyn | we have anti stl commitee | 18:42 |
@wiking | fork etc.. | 18:42 |
lisitsyn | what the boost are you talking about :D | 18:42 |
thoralf | wiking: For me it sounds like a good idea, because we're re-inventing data structures like refcounting, vectors, sparsevectors, etc. | 18:42 |
@wiking | cameo54321: then maybe even to template the type... | 18:42 |
@wiking | thoralf: :DDDDDDDDDDDDDDDDDDDDDDDDDDD | 18:42 |
thoralf | lisitsyn: :) | 18:42 |
@wiking | thoralf: aren't we already reinventing STL? :) | 18:42 |
thoralf | wiking: No, we're still using C ;) | 18:43 |
@wiking | thoralf: thing is think about the current situation with swig | 18:43 |
lisitsyn | I have a library simplifying stl algorithms btw | 18:43 |
lisitsyn | :D | 18:43 |
lisitsyn | self promotion is good | 18:43 |
@wiking | lisitsyn: :DDD | 18:43 |
lisitsyn | a few years later I will have a library for any boosting case | 18:43 |
@wiking | thoralf: yeah we never use #include <algorithm> anywhere. and most of the things are being implemented by ourselves instead of using plain ol' STL | 18:43 |
lisitsyn | I will replace the f word with the boost word | 18:43 |
@wiking | but it's all because that we don't have d-pointers | 18:44 |
@wiking | as soon as we would have that | 18:44 |
@wiking | noobody would care what bloated library u want to depend on | 18:44 |
lisitsyn | wiking: can you guess what do I have? | 18:44 |
@wiking | boost, fooost, stl, motherfuckingSTL++ | 18:44 |
@wiking | :) | 18:44 |
thoralf | wiking: ../src/shogun/lib/SGVector.cpp:#include <algorithm> | 18:44 |
@wiking | thoralf: yeah that's .cpp ;) | 18:44 |
@wiking | that doesn't count :D | 18:44 |
@wiking | lisitsyn: u have moost++ ? :) | 18:45 |
lisitsyn | wiking: yeah | 18:45 |
lisitsyn | wiking: well actually I have a class for dptrs :D | 18:45 |
@wiking | \o/ | 18:45 |
@wiking | yeah so now apply * | 18:45 |
lisitsyn | default constructor creates object | 18:45 |
lisitsyn | quite obvious right? | 18:45 |
lisitsyn | :D | 18:45 |
@wiking | mmm woah :D | 18:45 |
@wiking | a ctor that creates an object | 18:45 |
lisitsyn | hahah | 18:46 |
lisitsyn | no no | 18:46 |
@wiking | maaaan we are like 1990!!! | 18:46 |
@wiking | operator new! | 18:46 |
lisitsyn | I mean it is unique_ptr which default ctor creates the object | 18:46 |
@wiking | ah | 18:46 |
@wiking | ++ | 18:46 |
lisitsyn | so what you have to do is just | 18:46 |
lisitsyn | struct Private; | 18:46 |
lisitsyn | d_ptr<Private> private; | 18:46 |
@wiking | noooooooo, why struct | 18:47 |
@wiking | ? | 18:47 |
@wiking | why no class? :D | 18:47 |
lisitsyn | are you visual studio or what? | 18:47 |
lisitsyn | :D | 18:47 |
@wiking | fucking struct :) '80s | 18:47 |
lisitsyn | it is visual studio that feels the difference between struct and class | 18:47 |
@wiking | lisitsyn: and me! | 18:47 |
lisitsyn | haha | 18:47 |
cameo54321 | wiking: Any recommended library other than the f library? :D | 18:47 |
thoralf | wiking: Ah, now I got it. | 18:47 |
thoralf | Interfering jokes. | 18:47 |
lisitsyn | you can declare thing like struct | 18:47 |
@wiking | cameo54321: boost? | 18:48 |
lisitsyn | and then define as class | 18:48 |
@wiking | cameo54321: stan | 18:48 |
cameo54321 | wiking: haha yes | 18:48 |
@wiking | lisitsyn: reeeeealllllllllllllly? :D | 18:48 |
@wiking | lisitsyn: woah man i didn't know!!!! | 18:48 |
lisitsyn | wiking: yeah that sounds obvious right? | 18:48 |
@wiking | brand new INFORMATION! | 18:48 |
cameo54321 | wiking: we have already discarded stan sa for now.. | 18:48 |
lisitsyn | wiking: go try that in VS :D | 18:48 |
@wiking | cameo54321: not because of it's API | 18:48 |
@wiking | lisitsyn: noooooooooooooooooooooooooooooooo i dont want to open VS! | 18:49 |
@wiking | lisitsyn: actually the whole macro story of QT should be fine | 18:49 |
lisitsyn | yeah sure | 18:49 |
@wiking | or do we want something more 21st century? :D | 18:50 |
lisitsyn | though that's like 3 lines | 18:50 |
@wiking | no | 18:50 |
cameo54321 | wiking: stan's structure is not so good, we have gone through that | 18:50 |
@wiking | 5 | 18:50 |
@wiking | ! | 18:50 |
lisitsyn | I don't think it is really needed | 18:50 |
lisitsyn | :D | 18:50 |
@wiking | cameo54321: did u? sorry i missed that... i thought it's because it's insane coupling with boost, and we dont want boost + that they include dependent library in their repo... i do not remember a discussion about how bad stan's API is.... | 18:51 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 18:51 | |
@wiking | lisitsyn: :D | 18:51 |
cameo54321 | wiking: boost, put it aside for now lets say.. | 18:52 |
@wiking | cameo54321: so where have u discussed the API problems of stan? | 18:52 |
cameo54321 | wiking: we have already discussed and discarded stan (Because I couldn't understand its make structure.).. | 18:53 |
lisitsyn | the reasoning sounds cool :) | 18:53 |
@wiking | cameo54321: mmmm that's interesting | 18:53 |
cameo54321 | wiking: As for its API, I do not remember we had a discussion specifically for that | 18:53 |
lisitsyn | I haven't been following that discussion | 18:54 |
lisitsyn | what's with that stan at all | 18:54 |
cameo54321 | But point is, we moved on | 18:54 |
@wiking | cameo54321: so you are saying that stan's API is not good because you couldn't undestand it's make structure? | 18:54 |
@wiking | cameo54321: i think you are just mixing things up | 18:54 |
@wiking | there's one thing | 18:54 |
@wiking | that how aweful one's repository or make process is | 18:54 |
@wiking | and another thing how it API looks like | 18:54 |
@wiking | that's like saying: i don't like you house beause i did not like the construction worker | 18:55 |
@wiking | i mean seriously...? | 18:55 |
cameo54321 | wiking: Well I am saying that we had decided not to consider stan | 18:55 |
@wiking | cameo54321: yes we decided that we dont want to integrate stan | 18:55 |
@wiking | but that doesn't mean that we couldn't mimic stan's API | 18:55 |
@wiking | especially if it's good | 18:55 |
@wiking | but that should be investiaged | 18:56 |
@wiking | *investigated | 18:56 |
cameo54321 | wiking: well right now what I am looking for is how to move onwards.. | 18:56 |
@wiking | cameo54321: yeah and one option would be to check stan's api | 18:57 |
@wiking | i've been saying this like 20 minutes before | 18:57 |
@wiking | and it goes for boost as well | 18:57 |
@wiking | that's one thing that we dont want to integrate or depend on either of those two libraries... | 18:57 |
@wiking | that doesn't mean that we dont want to have anything do with their API | 18:58 |
@wiking | those two things are completely unrelated stuff | 18:58 |
@wiking | maybe they have some very nice design choices in their api (i've just suggested 2 from boost) that we should follow as well in our implementation | 18:59 |
cameo54321 | wiking: And the implementation has to be with using SGVector,SGMatrix etc...? | 19:01 |
@wiking | cameo54321: mmm i believe you dont understand me... | 19:07 |
shogun-buildbot_ | build #223 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/223 blamelist: Sunil Mahendrakar <sunil.mahendrakar19@gmail.com>, Heiko Strathmann <heiko.strathmann@gmail.com> | 19:07 |
lisitsyn | return recall_type<std::function<As<Metaclass<T>>>>(metaclass_factory)(); | 19:10 |
lisitsyn | I've gone too far | 19:10 |
lisitsyn | :D | 19:10 |
lisitsyn | wiking: you gonna like it ^ | 19:12 |
@wiking | lisitsyn: :D | 19:12 |
thoralf | >>>>> | 19:12 |
thoralf | Nice. | 19:12 |
thoralf | And a factory. | 19:13 |
thoralf | Why not a FactoryBuilder? | 19:13 |
lisitsyn | thoralf: metaclass factory! | 19:13 |
lisitsyn | metaclass abstract factory builder of singletons | 19:13 |
lisitsyn | so you can put a value to lvalue and rvalue and back and forth | 19:14 |
thoralf | lisitsyn: Touché. | 19:14 |
lisitsyn | I want to rename that as.hpp | 19:14 |
lisitsyn | to ass.hpp | 19:14 |
-!- tomy18 [~alex@p6.eregie.pub.ro] has quit [Ping timeout: 260 seconds] | 19:16 | |
lisitsyn | Internal compiler error: Error reporting routines re-entered. | 19:18 |
lisitsyn | Please submit a full bug report, | 19:18 |
lisitsyn | hahaha | 19:18 |
lisitsyn | wiking: advanced C++ is when you break the compiler | 19:18 |
thoralf | :) | 19:18 |
lisitsyn | oh okay our tv just announced we can turn the USA into 'radioactive ashes' :D | 19:26 |
-!- jiaolong [9e6d09e4@gateway/web/freenode/ip.158.109.9.228] has quit [] | 19:40 | |
@wiking | lisitsyn: :DDDDDDDDDDDDDDDDDDDDDDDDDDD | 19:41 |
thoralf | lisitsyn: Hope the won't try. ;) | 19:49 |
-!- parijat [75c7d31f@gateway/web/freenode/ip.117.199.211.31] has joined #shogun | 19:53 | |
parijat | HeikoS: Hello! | 19:54 |
parijat | HeikoS: about softmax function, in case of derivatives of softmax, I have to return the exponential.. which is unlike log_sum_exp... anyway to prevent underflow here as well? | 19:58 |
-!- khalednasr [~k.nasr92@41.69.143.152] has joined #shogun | 20:00 | |
-!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has joined #shogun | 20:00 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 20:00 | |
@HeikoS | wiking, cameo54321 hey! | 20:00 |
@HeikoS | parijat: hi! | 20:00 |
@HeikoS | wiking, cameo54321 time for some flame wars ;D | 20:01 |
cameo54321 | cameo54321: Present | 20:01 |
@HeikoS | wiking: so whats wrong with boost? | 20:01 |
parijat | HeikoS: hi! :) | 20:01 |
@HeikoS | wiking: What are the alternatives for one dimensional pdf functions? | 20:01 |
@HeikoS | parijat: let me see, could you point me to a line again? | 20:01 |
parijat | HeikoS: yup just a min | 20:02 |
parijat | HeikoS: 81 in .cpp | 20:02 |
parijat | Hei | 20:02 |
parijat | HeikoS: i just return the eigen_ret matrix a few lines later. | 20:03 |
@HeikoS | parijat: get_log_probability_derivative1_f ? | 20:03 |
parijat | yup | 20:03 |
@HeikoS | parijat: ah I see | 20:04 |
@HeikoS | parijat: I think that should be ok shouldnt it? | 20:04 |
@HeikoS | parijat: point was more in the log-sum-exp | 20:04 |
parijat | HeikoS: i should be. I think. I am getting accurate results uptil 4th decimal point in the unittest | 20:05 |
@HeikoS | parijat: the problem with that is always small numbers which pop up when we have a lot of data in the GP. then individual probabilities are like 1e-10 | 20:06 |
@HeikoS | exponentiating that quickly breaks float64_t | 20:06 |
parijat | HeikoS: hmm.. i see | 20:06 |
parijat | HeikoS: do you have any idea how matlab etc handles that? | 20:07 |
@HeikoS | parijat: well there is no way around it if one stores probabilities in the normal domain | 20:07 |
@HeikoS | parijat: so maybe try some extreme examples and see what happens | 20:08 |
@HeikoS | whether its all fine even with extremely hard cases | 20:08 |
-!- PirosB3 [~pirosb3@cpc4-newt31-2-0-cust631.newt.cable.virginm.net] has joined #shogun | 20:08 | |
parijat | HeikoS: these small errors might accrue in multiclass gp | 20:09 |
parijat | HeikoS: yes, let me test some hard cases | 20:10 |
@HeikoS | parijat: cool! | 20:15 |
@HeikoS | I gotta cycle home now, but I will be back a bit later | 20:15 |
@HeikoS | wiking: lets also discuss some things then | 20:15 |
-!- HeikoS [~heiko@nat-184-43.internal.eduroam.ucl.ac.uk] has quit [Ping timeout: 240 seconds] | 20:20 | |
-!- cod123 [514481ef@gateway/web/freenode/ip.81.68.129.239] has joined #shogun | 20:20 | |
-!- dhruv13__ [~dhruv13@84.64.15.73] has joined #shogun | 20:32 | |
-!- dhruv13_ [~dhruv13@84.64.14.115] has quit [Ping timeout: 240 seconds] | 20:34 | |
-!- iglesiasg [~iglesiasg@524AE0A7.cm-4-3d.dynamic.ziggo.nl] has quit [Read error: Connection reset by peer] | 20:35 | |
-!- parijat [75c7d31f@gateway/web/freenode/ip.117.199.211.31] has quit [Quit: Page closed] | 20:35 | |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has quit [Ping timeout: 252 seconds] | 20:37 | |
-!- grzes [~grzes@host-81-190-236-173.wroclaw.mm.pl] has quit [Quit: Konversation terminated!] | 20:47 | |
-!- Rho [~rho@217.55.207.227] has joined #shogun | 20:52 | |
thoralf | wiking: 32 of the SerializationHDF5 tests are failing on my machine. | 20:59 |
thoralf | wiking: Do you know if are working somewhere else? | 21:00 |
-!- tomy18 [~alex@p6.eregie.pub.ro] has joined #shogun | 21:01 | |
thoralf | In file src/shogun/io/SerializableFile.cpp line 120: Could not read `current_aggregate' from `/tmp/StoreScalarAggregator_PT_FLOATMAX.Gsq4vu | 21:01 |
thoralf | or lisitsyn? | 21:03 |
lisitsyn | wwwwat | 21:03 |
-!- sonne|osx [~sonne@e178048019.adsl.alicedsl.de] has joined #shogun | 21:07 | |
thoralf | Hey Sören. | 21:13 |
-!- Saurabh7 [~Saurabh7@117.200.181.180] has quit [Ping timeout: 240 seconds] | 21:14 | |
-!- kislay [~kislay@14.139.41.150] has quit [Read error: Connection reset by peer] | 21:18 | |
-!- kislay [~kislay@14.139.41.150] has joined #shogun | 21:19 | |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has joined #shogun | 21:22 | |
dhruv13__ | hey! does anyone know why valgrind would report crc errors? | 21:24 |
thoralf | dhruv13__: Can post the complete output to a gist? | 21:25 |
-!- Saurabh7 [~Saurabh7@117.200.179.88] has joined #shogun | 21:26 | |
dhruv13__ | https://gist.github.com/dhruv13J/9589396 | 21:26 |
dhruv13__ | that's without -v, with is too huge | 21:27 |
dhruv13__ | but it really should not be reporting problems with a lot of those files | 21:28 |
dhruv13__ | could it happen that the program terminates, and all the memory is dumped? | 21:28 |
dhruv13__ | nope, that's not the case | 21:30 |
@wiking | thoralf: mmmmmm all the buildbots? | 21:31 |
-!- Rho [~rho@217.55.207.227] has quit [Remote host closed the connection] | 21:31 | |
@wiking | thoralf: see here: http://buildbot.shogun-toolbox.org/waterfall | 21:32 |
@wiking | thoralf: and travis | 21:32 |
thoralf | wiking: Strange. | 21:34 |
thoralf | Fails here. | 21:34 |
dhruv13__ | hey! I ran ./shogun-unit-test with valgrind, and i am getting huge data losses... is that normal? | 21:37 |
dhruv13__ | I did not mess around with any important stuff, and none of the unit tests access my code changes | 21:38 |
thoralf | dhruv13__: What's "data losses"? | 21:40 |
dhruv13__ | sorry, memory leaks | 21:41 |
-!- dhruv13__ is now known as dhruv13 | 21:41 | |
dhruv13 | am i missing something important | 21:42 |
thoralf | dhruv13: Well, first checkout a fresh develop and confirm that's leaking there. If so, make a bug request (with shogun version/git hash) and an attached valgrind log. | 21:42 |
dhruv13 | ? | 21:42 |
dhruv13 | okay | 21:43 |
thoralf | If you're sure that it's not your fault, then the above is the best way. | 21:43 |
thoralf | If you are not sure, you can try with individual tests | 21:43 |
thoralf | For example: ./shogun-unit-test --gtest_filter="*suspectTest*" | 21:44 |
thoralf | valgrind ./shogun-unit-test --gtest_filter="*suspectTest*" | 21:44 |
shogun-buildbot_ | build #224 of debian wheezy - memcheck is complete: Failure [failed memory check] Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/224 blamelist: tklein23 <tklein23@users.noreply.github.com>, Thoralf Klein <thoralf@fischlustig.de> | 21:44 |
dhruv13 | thoralf: yep, tried those | 21:46 |
dhruv13 | thoralf: getting leaks with nearly all matrix tests | 21:46 |
@wiking | dhruv13: before u report all the huge data losses check if they are the same as http://buildbot.shogun-toolbox.org/memcheck/20140316-1857.html | 21:47 |
@wiking | since then we know about them.. | 21:47 |
@wiking | dhruv13: and if u use valgrind bare foot then dont forget to use the suppress file as well... | 21:47 |
dhruv13 | wiking: yep, some of those... will do | 21:48 |
@wiking | btw thoralf there's a leak u've introduced: http://buildbot.shogun-toolbox.org/memcheck/20140316-1857.html#idp259632 | 21:48 |
dhruv13 | wiking: there's a memory leak in SGObject?! | 21:49 |
@wiking | thoralf: or at least revealed :D | 21:49 |
@wiking | dhruv13: yes ... 5 of them why? | 21:49 |
thoralf | wiking: Which one? | 21:49 |
@wiking | thoralf: there's the link | 21:50 |
dhruv13 | wiking: I'm testing leaks on my changes to KNN.cpp, and was losing ~1800 bytes, could that be because of this leak? | 21:50 |
@wiking | dhruv13: no | 21:51 |
@wiking | dhruv13: that's your problem.. | 21:51 |
dhruv13 | wiking: the -v option on valgrind reveals CRC mismatch errors, and i have no clue what those are | 21:51 |
@wiking | dhruv13: rtfm | 21:52 |
@wiking | that's your best way to go | 21:52 |
dhruv13 | wiking: okay | 21:52 |
thoralf | wiking: Oh, yes. Thanks. | 21:58 |
thoralf | wiking: I see the leak. Should be easy to fix. | 21:59 |
-!- newlido [~quassel@105.90.207.185] has quit [Ping timeout: 264 seconds] | 22:07 | |
-!- Saurabh7 [~Saurabh7@117.200.179.88] has quit [Excess Flood] | 22:19 | |
-!- Saurabh7 [~Saurabh7@117.200.179.88] has joined #shogun | 22:20 | |
thoralf | wiking: https://github.com/shogun-toolbox/shogun/pull/2014 | 23:02 |
lisitsyn | finally PR 2014 | 23:05 |
lisitsyn | thoralf: I cna't believe we are doing that in 2014 :D | 23:06 |
thoralf | lisitsyn: :D | 23:06 |
lisitsyn | thoralf: your patch is totally ok but the codestyle in general | 23:06 |
lisitsyn | that's C :( | 23:06 |
thoralf | libncbm is C ;) | 23:06 |
lisitsyn | yeah | 23:06 |
thoralf | And I copy-pasted it from libbmrm | 23:07 |
thoralf | But while(cp_ptr!=NULL) is really great. Reminds me when I started with C :D | 23:08 |
lisitsyn | yes yes | 23:08 |
lisitsyn | why oh why | 23:08 |
-!- [paladin] [~udit@122.177.168.86] has quit [Quit: leaving] | 23:11 | |
lisitsyn | thoralf: this code also has inconsistent variable naming | 23:11 |
lisitsyn | BufSize | 23:11 |
lisitsyn | maxCPs | 23:11 |
lisitsyn | ACPs | 23:11 |
lisitsyn | :D | 23:11 |
lisitsyn | totally crazy | 23:11 |
lisitsyn | or even CPList_head | 23:11 |
-!- newlido [~quassel@41.237.225.99] has joined #shogun | 23:14 | |
-!- tomy18 [~alex@p6.eregie.pub.ro] has quit [Ping timeout: 260 seconds] | 23:16 | |
-!- sonne|osx [~sonne@e178048019.adsl.alicedsl.de] has quit [Quit: sonne|osx] | 23:26 | |
-!- lambday [67157d54@gateway/web/freenode/ip.103.21.125.84] has quit [Ping timeout: 245 seconds] | 23:27 | |
-!- fs_ [~floriansc@e179222254.adsl.alicedsl.de] has quit [Remote host closed the connection] | 23:33 | |
-!- newlido_ [~quassel@41.237.225.99] has joined #shogun | 23:38 | |
-!- Netsplit *.net <-> *.split quits: vperic, kislay, thoralf, newlido | 23:44 | |
-!- Netsplit over, joins: kislay, thoralf, vperic | 23:45 | |
@wiking | lisitsyn: u dont like my code? :D | 23:49 |
@wiking | lisitsyn: wrote it in 12 hours :> | 23:49 |
thoralf | wiking, lisitsyn: What do you think about changing a few headers to forward declarations instead of #includes and moving them to the implementations? | 23:55 |
thoralf | I just ran an analysis over a few classes and I moved the #include's to .cpp files and only used forward declarations. | 23:55 |
thoralf | According to what I know, this is the way to got, but I'm not sure how we handle it in shogun. | 23:56 |
-!- HeikoS [~heiko@2.219.252.217] has joined #shogun | 23:59 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 23:59 | |
--- Log closed Mon Mar 17 00:00:41 2014 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!