--- Log opened Fri Jun 07 00:00:18 2013 | ||
--- Day changed Fri Jun 07 2013 | ||
wiking | i mean i dont want us to start creating a M/R arch within shogun | 00:00 |
---|---|---|
wiking | but still having multicore support would be like | 00:00 |
wiking | or *should* be like default | 00:00 |
wiking | in shogun | 00:00 |
@sonney2k | wiking, I fully agree | 00:01 |
lisitsyn | wiking: I believe we should go for performance in shogun | 00:01 |
lisitsyn | like the fastest SVM (in means of classify) | 00:01 |
@sonney2k | HeikoS, what is your vision? | 00:01 |
@sonney2k | lisitsyn, we have that already | 00:02 |
lisitsyn | have you measured? | 00:02 |
wiking | lisitsyn: i second your idead | 00:02 |
wiking | *idea | 00:02 |
@sonney2k | lisitsyn, yes | 00:02 |
lisitsyn | sonney2k: with what and when? | 00:02 |
wiking | sonney2k: for me liblinear was fucking slow :) | 00:02 |
@HeikoS | sonney2k: I dont know, to be honest I think that multicore is not really useful since max speedup is 4x at the moment, later maybe 8x, maybe 16x buts thats it | 00:02 |
wiking | sonney2k: compared to a random forest classifier :) | 00:02 |
lisitsyn | HeikoS: what speedup you want then :D | 00:02 |
@HeikoS | lisitsyn: well 100x | 00:03 |
@HeikoS | 1000x | 00:03 |
wiking | HeikoS: then distributed | 00:03 |
wiking | easy | 00:03 |
@sonney2k | lisitsyn, HeikoS my vision is to get all the 'reasonable baseline' ML methods into shogun | 00:03 |
@HeikoS | sonney2k: ah I see that is what you meant | 00:03 |
@HeikoS | yeah same | 00:03 |
wiking | HeikoS: can we do distributed now... not | 00:03 |
lisitsyn | I'd move all the things we do not support to unsupported | 00:03 |
@HeikoS | I would like to see more people appreciating shogun | 00:03 |
@sonney2k | improve usability and have things that need to be large scale too | 00:03 |
@HeikoS | since its quite good | 00:03 |
lisitsyn | lets just create a module and move the shit out | 00:04 |
@HeikoS | so get away from this old-school installation procedure and offer packages | 00:04 |
@HeikoS | tutorial | 00:04 |
@HeikoS | etc etc | 00:04 |
@HeikoS | large scale is done by others anyway | 00:04 |
@sonney2k | wiking, how can we do distributed if we don't all have clusters? | 00:04 |
@sonney2k | whatever large scale means | 00:04 |
lisitsyn | HeikoS: if largescale is done by others | 00:04 |
wiking | sonney2k: we can haz one :) | 00:04 |
lisitsyn | what do we do | 00:04 |
wiking | lisitsyn: mahout | 00:04 |
@HeikoS | shogun's focus is not on distributed anyway | 00:04 |
wiking | lisitsyn: they do that | 00:05 |
@HeikoS | we can maybe do PBS stuff (independent) | 00:05 |
@HeikoS | but thats optional | 00:05 |
lisitsyn | wiking: what do we do then? | 00:05 |
@HeikoS | multicore would be nice | 00:05 |
wiking | lisitsyn: desktop in 2000 | 00:05 |
wiking | :> | 00:05 |
lisitsyn | I keep asking that question and can't find an answer for myself | 00:05 |
@HeikoS | but if the code looks as in CKernel, I would rather not do it | 00:05 |
@sonney2k | I am perfectly fine seeing that others need a cluster for what I am doing on a single core with shogun | 00:05 |
@sonney2k | HeikoS, yeah CKernel is a nightmare | 00:05 |
@HeikoS | we are also far away from being able to do stuff on clusters | 00:05 |
@HeikoS | since we have other problems | 00:05 |
lisitsyn | wiking: I'd like to see microtuned crazy fast shit then | 00:05 |
wiking | sonney2k: well there are some datasets that you just cannot handle with one machine ;) | 00:05 |
@HeikoS | lisitsyn: I would rather go for stability | 00:06 |
@HeikoS | computers get faster on their own | 00:06 |
@HeikoS | stable and reliable | 00:06 |
wiking | sonney2k: have u tried doing PCA on 10^10 instances each having 512 features :P | 00:06 |
lisitsyn | HeikoS: it is average | 00:06 |
lisitsyn | no stable is not the way | 00:06 |
@HeikoS | wiking: no needs to do that | 00:06 |
lisitsyn | software doesn't evolves trying to stabilize | 00:06 |
@sonney2k | wiking, sure there are. but I never had where having more than a few GB of examples did help | 00:06 |
@HeikoS | using 10^3 will give almost same results | 00:07 |
wiking | HeikoS: really? :) | 00:07 |
lisitsyn | software dies when it stabilizes | 00:07 |
@HeikoS | lisitsyn: no | 00:07 |
@sonney2k | HeikoS, wiking well I had applicatiosn where I used 50 million examples in 10 million dims | 00:07 |
@sonney2k | and whenever I doubled the number of examples | 00:07 |
@sonney2k | I could halve the error | 00:08 |
@HeikoS | lisitsyn: whats the point of having software that is not reliable | 00:08 |
wiking | sonney2k: on thing for sure: you r not the average user of shogun | 00:08 |
@HeikoS | sonney2k: that might be true, but not for PCA | 00:08 |
@sonney2k | man we should organize a fight at the workshop | 00:08 |
lisitsyn | HeikoS: reliability is not stabilization it is good design usually | 00:08 |
wiking | sonney2k: i might even say you are not even the target user i guess ;) | 00:08 |
@HeikoS | lisitsyn: yes | 00:08 |
@HeikoS | agreed | 00:08 |
@sonney2k | wiking, hehe | 00:08 |
@HeikoS | I dont care how to achieve this | 00:08 |
@HeikoS | so btw all my office mates so far failed to even install shogun | 00:08 |
@sonney2k | wiking, so far I have been writing what I need | 00:08 |
wiking | sonney2k: yeah i know | 00:09 |
lisitsyn | to become better we would have to break everything | 00:09 |
wiking | sonney2k: i'm the same | 00:09 |
@sonney2k | wiking, it is hard to write code I cannot imagine to use | 00:09 |
@HeikoS | and they are serious ML scientists | 00:09 |
@sonney2k | wiking, so it should stay that way | 00:09 |
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has joined #shogun | 00:09 | |
@HeikoS | just no hackers | 00:09 |
lisitsyn | because we fucked with designing it | 00:09 |
@HeikoS | normal people who usually use matlab | 00:09 |
@HeikoS | ah, too many things going on at once | 00:09 |
@sonney2k | lisitsyn, no because they cannot type | 00:09 |
lisitsyn | hah | 00:09 |
@HeikoS | haha, I like the fight idea ;) | 00:09 |
@sonney2k | ./configure; make; make install | 00:09 |
lisitsyn | sonney2k: who cannot type? | 00:09 |
lisitsyn | ah | 00:09 |
wiking | HeikoS: hehehe normal people & using matlab? :DDDDDDDDDDDDDDDDDD | 00:09 |
lisitsyn | wrong person | 00:09 |
@sonney2k | lisitsyn, real scientists :) | 00:10 |
wiking | HeikoS: best sarcasm ever!!! | 00:10 |
lisitsyn | sonney2k: it was HeikoS ;) | 00:10 |
@HeikoS | sonney2k: no | 00:10 |
@HeikoS | that is not true | 00:10 |
@sonney2k | lisitsyn, but HeikoS learns quickly :D | 00:10 |
@HeikoS | dependencies blabla | 00:10 |
@sonney2k | lisitsyn, from trees to matlab to shogun in just a few years | 00:10 |
wiking | there's no normal person who uses matlab in 2013 | 00:10 |
lisitsyn | then lets get rid of dependencies | 00:10 |
@HeikoS | most scientists I know dont know how to install stuff from the source | 00:10 |
wiking | there's just those guys who write in cobol | 00:10 |
wiking | today | 00:10 |
wiking | :> | 00:10 |
@HeikoS | if we make fun of them, they wont use our software | 00:11 |
@sonney2k | wiking, wrong unfortunately | 00:11 |
@HeikoS | have packages | 00:11 |
@sonney2k | wiking, ML'ers still use matlab a lot | 00:11 |
wiking | sonney2k: well of course wrong but it's a fucking hell | 00:11 |
@HeikoS | sonney2k: what do you mean with "still" this is not going to change | 00:11 |
@sonney2k | wiking, but if people switch away from matlab | 00:11 |
@sonney2k | the switch to R or python! | 00:11 |
wiking | sonney2k: hahahaha | 00:11 |
wiking | R | 00:11 |
wiking | :D | 00:11 |
wiking | i will kill R one day | 00:11 |
wiking | i wanted to love that software | 00:11 |
wiking | i honesst really tried hard | 00:11 |
wiking | but it's just a fucking nightmare | 00:12 |
lisitsyn | it is unlearnable for me | 00:12 |
@sonney2k | wiking, yeah... writing an extension in are is like chopping your head of and attaching it to your butt or sth | 00:12 |
wiking | lisitsyn: heheh yeah :) | 00:12 |
@sonney2k | it is CRAZY | 00:12 |
wiking | yep | 00:12 |
@sonney2k | the language itself is OKish IMHO | 00:12 |
lisitsyn | I must be stupid or what | 00:12 |
wiking | python on the otherhand is great | 00:12 |
lisitsyn | but I can't use it too | 00:12 |
@sonney2k | python is written by CS people | 00:12 |
@sonney2k | so it has style | 00:12 |
wiking | and it's great that shogun interfaces to it | 00:12 |
wiking | since then you can just mix all the great libs for python | 00:13 |
@sonney2k | alright so what I learn from that | 00:13 |
wiking | but still | 00:13 |
@sonney2k | we focus on nice python support | 00:13 |
wiking | yes | 00:13 |
lisitsyn | we should just rewrite everything to python that's all | 00:13 |
@sonney2k | and openmp | 00:13 |
wiking | and openmp | 00:13 |
wiking | lisitsyn: \o/ | 00:13 |
@sonney2k | and wiking does openmp tomorrow to configure | 00:13 |
lisitsyn | anyway what matters is design | 00:13 |
wiking | sonney2k: like how about python3 support? | 00:13 |
@sonney2k | wiking, whenever we have *long* running algorithms though we have to go pthread | 00:14 |
wiking | sonney2k: i mean i have no fucking clue how's that going now with shogun | 00:14 |
lisitsyn | if we just reproduce ideas of shogun in python we'd get a great library :D | 00:14 |
@sonney2k | wiking, long running as in days | 00:14 |
@sonney2k | wiking, we have that since last year | 00:14 |
@sonney2k | python3 I mean | 00:14 |
@HeikoS | lisitsyn: rewrite in python | 00:14 |
@HeikoS | nice | 00:14 |
lisitsyn | haha | 00:14 |
@HeikoS | we can just integrate ourselves into scikit | 00:15 |
lisitsyn | HeikoS: jk but there is a truth beneath | 00:15 |
@HeikoS | lisitsyn: I know | 00:15 |
wiking | sonney2k: cool that's great! | 00:15 |
lisitsyn | HeikoS: I mean what is the most valuable | 00:15 |
@HeikoS | python is so much easier while not being slow if you know how to use it | 00:15 |
wiking | ah and one more tinny little thing | 00:15 |
lisitsyn | HeikoS: is the design | 00:15 |
wiking | SG_REF + SG_UNREF | 00:15 |
wiking | :D | 00:15 |
@HeikoS | yep | 00:15 |
@sonney2k | lisitsyn, won't work for me. I needed shogun from java | 00:15 |
lisitsyn | sonney2k: no that's just a joke, it is too hard to do that | 00:15 |
wiking | let's j2ee shogun | 00:16 |
wiking | :> | 00:16 |
lisitsyn | wiking: oh come on | 00:16 |
wiking | put it in a container | 00:16 |
lisitsyn | and clearcase | 00:16 |
lisitsyn | as CVS | 00:16 |
lisitsyn | right? | 00:16 |
* sonney2k vomits | 00:16 | |
wiking | and then just deploy it all the fucking time | 00:16 |
wiking | ;D | 00:16 |
wiking | anyhow. so what about SG_REF+UNREF | 00:16 |
lisitsyn | wiking: ahh so that's how you solve HeikoS's colleagues' problem | 00:16 |
lisitsyn | we put a container into container | 00:16 |
wiking | can we pretty please | 00:16 |
lisitsyn | wiking: kill it? | 00:17 |
@sonney2k | wiking, I would have wished pluskid stayed with us to do smart pointers after gsoc | 00:17 |
wiking | start at least thinking about | 00:17 |
wiking | having smart pointers | 00:17 |
lisitsyn | sonney2k: why pluskid? | 00:17 |
@sonney2k | lisitsyn, he knows *exactly* how to do this | 00:17 |
lisitsyn | I know too | 00:17 |
@sonney2k | lisitsyn, recall that he helped doing the SGReferenced* datatypes? | 00:17 |
wiking | i mean in 2 years i'm for sure that 4 out of 5 compilers will just natively support c++11 | 00:17 |
lisitsyn | sonney2k: what is difficult in shared_ptr? | 00:17 |
@sonney2k | lisitsyn, to convert the whole shogun codebase and swig stuff? | 00:18 |
lisitsyn | sonney2k: it is a matter of time and work | 00:18 |
@sonney2k | sure but lisitsyn are you sure you can do that? | 00:18 |
wiking | sonney2k: we do it on the workshop | 00:18 |
lisitsyn | no because I don't have that amount of time | 00:18 |
wiking | :) | 00:18 |
@sonney2k | yeah... | 00:19 |
@sonney2k | this clearly is a non GSoC project | 00:19 |
lisitsyn | speaking of concept - I am sure it can be done | 00:19 |
@sonney2k | we would have to do together | 00:19 |
@sonney2k | lisitsyn, and timing is important too | 00:19 |
lisitsyn | sonney2k: then it is not the time | 00:20 |
wiking | well | 00:20 |
@sonney2k | we would have to do this right *after* the release we do after gsoc | 00:20 |
wiking | workshop? :) | 00:20 |
lisitsyn | sonney2k: branching it is not a solution at all | 00:20 |
@sonney2k | exactly | 00:20 |
lisitsyn | because merging it afterwards will be as hard | 00:20 |
wiking | lisitsyn: noup | 00:20 |
lisitsyn | as doing it from the very beginning | 00:20 |
lisitsyn | :D | 00:20 |
wiking | lisitsyn: i bet u 100 euros that it can be done with having to much troubles | 00:20 |
@HeikoS | maybe just rewrite from scratch | 00:20 |
@HeikoS | have a new shogun branch | 00:20 |
@HeikoS | with similar api | 00:20 |
@HeikoS | freeze the old one | 00:21 |
@HeikoS | in the new one, only do things properly | 00:21 |
wiking | HeikoS: there's no proper way :) | 00:21 |
lisitsyn | HeikoS: very very good idea | 00:21 |
@HeikoS | we know more than before wiking | 00:21 |
@HeikoS | support the old one, like fix bugs etc | 00:21 |
@HeikoS | but continue writing the ne wone | 00:21 |
@sonney2k | no no no | 00:21 |
@HeikoS | decide on core methods before doing that | 00:21 |
@HeikoS | haha :) | 00:21 |
@sonney2k | that is very dangerous | 00:21 |
@HeikoS | sorry sonney2k | 00:21 |
@sonney2k | you might not finish the new one | 00:21 |
@sonney2k | and then be stuck forever | 00:22 |
wiking | lisitsyn: ok what do u need for openmp | 00:22 |
lisitsyn | wiking: -fopenmp | 00:22 |
lisitsyn | :D | 00:22 |
wiking | lisitsyn: what are your macros in tapkee? | 00:22 |
@HeikoS | sonney2k: well has to be designed well | 00:22 |
@sonney2k | it is like what happened with the mplayer -> mplayer2 design | 00:22 |
@HeikoS | I am sure that one can bring something to run quite quickly | 00:22 |
lisitsyn | wiking: no macros it is just ignored | 00:22 |
@HeikoS | but mplayer is way more complex | 00:22 |
wiking | lisitsyn: or it's just sitting there as #pragma? | 00:22 |
@HeikoS | we just need CSGObject | 00:22 |
lisitsyn | wiking: yes just pragma | 00:22 |
@HeikoS | features (not templated) | 00:22 |
@sonney2k | much better design does not mean people will use it | 00:22 |
wiking | lisitsyn: ok cool... here we go | 00:22 |
@HeikoS | sonney2k: people dont really use shogun currently | 00:22 |
@HeikoS | compared to what it can do | 00:22 |
@sonney2k | wiking, we have one openmp show stopper which is clang not supporting it IIRC? | 00:22 |
@HeikoS | its so nice, so many cool things | 00:23 |
wiking | btw | 00:23 |
wiking | sonney2k: wroooong | 00:23 |
@HeikoS | but not really recognized as this | 00:23 |
wiking | :) | 00:23 |
@sonney2k | HeikoS, yeah only very few features of it true | 00:23 |
@sonney2k | wiking, since when? | 00:23 |
wiking | llvm33 | 00:23 |
@HeikoS | so these can be brought to life quite quickly | 00:23 |
@HeikoS | dont know | 00:23 |
@sonney2k | wiking, I have 3.0 | 00:23 |
@HeikoS | just a thought | 00:23 |
wiking | sonney2k: well that's too bad :P | 00:23 |
lisitsyn | HeikoS: I like your idea | 00:23 |
wiking | btw | 00:24 |
lisitsyn | HeikoS: I have an approach for that | 00:24 |
wiking | preprocessor | 00:24 |
@HeikoS | on the other hand, lisitsyn, there are so many things unsolved | 00:24 |
wiking | we need a factory for that :) | 00:24 |
lisitsyn | HeikoS: yes | 00:24 |
@sonney2k | HeikoS, IMHO if we switch to smart pointers we should discuss how to do it at the workshop and then we all do it together from say october to november | 00:25 |
wiking | i mean it's just plain stupid that almost all the f*ing preprocessor is hardcodedly explicit casted to DenseFeatures<float64_t> | 00:25 |
@HeikoS | sonney yep sounds good! | 00:25 |
wiking | sonney2k: we just fucking do it | 00:25 |
lisitsyn | wiking: I find it stupid to have templated features now | 00:25 |
wiking | i'm adding check_boost() {} into ./configure ;) | 00:25 |
lisitsyn | uh boost why? | 00:26 |
@sonney2k | lisitsyn, the exist only for one reason: to have type safety... | 00:26 |
lisitsyn | wiking: MPL? | 00:26 |
wiking | lisitsyn: smart pointer... in case c++11 not supoprted | 00:26 |
lisitsyn | :D | 00:26 |
wiking | :) | 00:26 |
lisitsyn | sonney2k: compile-time right? | 00:26 |
@sonney2k | lisitsyn, yes | 00:26 |
lisitsyn | sonney2k: but no compile time in python? | 00:26 |
lisitsyn | :) | 00:26 |
lisitsyn | see what I mean? | 00:26 |
@sonney2k | lisitsyn, sure | 00:27 |
@sonney2k | anyway, I think we should focus on writing shogun algorithms for the machines we have access to | 00:27 |
wiking | ah and talking about mplayer.... this ./configure script stolen from mplayer starts to be bloated ;) | 00:27 |
@sonney2k | and include the algorithms we intend / want to use | 00:27 |
lisitsyn | wiking: yeah one million (or how many lines does it have) | 00:27 |
wiking | sonney2k: ok i have a nvidia tesla | 00:27 |
@sonney2k | wiking, really? | 00:28 |
wiking | sonney2k: i cannot run fucking shogun on it | 00:28 |
@sonney2k | richy rich? | 00:28 |
wiking | :( | 00:28 |
wiking | sonney2k: ok i was expressing myself in a wrong way | 00:28 |
wiking | sonney2k: i have access to nvidia tesla | 00:28 |
lisitsyn | haha | 00:28 |
wiking | i cannot run shogun on it | 00:28 |
wiking | :) | 00:28 |
lisitsyn | how much it costs? | 00:28 |
@sonney2k | lisitsyn, 1k | 00:28 |
wiking | i guess it's about 2k eur | 00:28 |
wiking | or something like that | 00:29 |
wiking | ah ok | 00:29 |
wiking | even less u see :) | 00:29 |
lisitsyn | what is the slot? pci-e? | 00:29 |
wiking | a gsocer can buy it | 00:29 |
@sonney2k | but it varies of course | 00:29 |
@sonney2k | high end is more $$$ | 00:29 |
wiking | this one has like kazillion cores and memory | 00:29 |
lisitsyn | want one | 00:29 |
@sonney2k | ok sound smore like >1k | 00:29 |
wiking | and i would loooove to ML the hell out of it :P | 00:29 |
wiking | anyhow ./configure is like | 00:30 |
lisitsyn | well I can buy it | 00:30 |
@sonney2k | wiking, well neural networks are the only algorithms I know that *highly* benefit from GPUs | 00:30 |
wiking | edited lately only by sonney2k and me :) | 00:30 |
@HeikoS | please no gpus | 00:30 |
lisitsyn | so not richy rich | 00:30 |
@HeikoS | we cannot handle this | 00:30 |
@HeikoS | its hard to code on them | 00:30 |
wiking | i think nobody else fucking has the guts to edit add stuff to ./configure | 00:30 |
wiking | am i right HeikoS ? :) | 00:30 |
@sonney2k | wiking, no also lambday IIRC | 00:30 |
lisitsyn | HeikoS: why? | 00:30 |
lisitsyn | it is not that hard to code in opencl | 00:30 |
@HeikoS | wiking: I dont touch configure! :D | 00:31 |
wiking | sonney2k: was lambday with the gtest patch :) | 00:31 |
wiking | sonney2k: ^ see i told u | 00:31 |
lisitsyn | wiking: I edit configure sometimes | 00:31 |
@HeikoS | lisitsyn: the syntax is not hard, but to write a good algorithm is hard | 00:31 |
wiking | ./configure is like magic script to most of the shogun core developers ;) | 00:31 |
lisitsyn | when I feel too happy and want to get depressed wiking | 00:31 |
wiking | lisitsyn: heheh i know i know | 00:31 |
wiking | :) | 00:31 |
lisitsyn | it works all the time | 00:31 |
lisitsyn | I just open it | 00:32 |
lisitsyn | and bam | 00:32 |
wiking | i just fucking hate it that when i want to add a new test | 00:32 |
lisitsyn | I am depressed | 00:32 |
wiking | i have to edit the file like 3-4 different places | 00:32 |
lisitsyn | guys | 00:32 |
@sonney2k | wiking, but you extremely rarely touch it | 00:32 |
wiking | sonney2k: heheheh | 00:32 |
wiking | sonney2k: actually no | 00:32 |
lisitsyn | I had some ideas let me put them here | 00:33 |
wiking | sonney2k: i touched that shit like almost all the fucking time i did somethign with shogun | 00:33 |
@sonney2k | wiking, ok ou touched it 18 times | 00:33 |
@sonney2k | you | 00:33 |
wiking | sonney2k: see it's a lot | 00:33 |
wiking | :) | 00:33 |
wiking | for me it was already too much | 00:33 |
@sonney2k | sergay 15 times :D | 00:33 |
lisitsyn | 1) storing *all* parameters as some class parameter<type> with checkers and other stuff | 00:33 |
wiking | or enough to HATE it | 00:33 |
wiking | lisitsyn: sergay ;) | 00:33 |
wiking | ^ | 00:33 |
lisitsyn | 2) submitting tasks (like svm.train().submit()) | 00:34 |
lisitsyn | thank you sonney2k | 00:34 |
lisitsyn | 3) thread/memory pools | 00:34 |
@sonney2k | oops | 00:34 |
@sonney2k | lisitsyn, sry | 00:34 |
lisitsyn | 4) convenient modelselection | 00:34 |
@sonney2k | no idea how I unintentionally mistyped sergey | 00:34 |
lisitsyn | 5) generic optimization out of box | 00:34 |
lisitsyn | 6) extensive caching | 00:35 |
lisitsyn | that's what I have collected | 00:35 |
wiking | lisitsyn: for 3) memory pool: just support pluging in std malloc replacements.... | 00:35 |
wiking | lisitsyn: like jmalloc and such | 00:35 |
lisitsyn | wiking: no I know it is all easy and etc - just my thoughts on say good approaches | 00:35 |
wiking | lisitsyn: yeah yeah sure | 00:35 |
wiking | lisitsyn: and we take the threadpool from boost ;) | 00:36 |
wiking | so | 00:36 |
wiking | somebody start fucking write the CMakefiles.txt | 00:36 |
wiking | thankuloveu | 00:36 |
wiking | :> | 00:36 |
lisitsyn | haha | 00:36 |
lisitsyn | well I have got some experience on that | 00:36 |
lisitsyn | can do | 00:36 |
@sonney2k | CMakeLists.txt is in shogun/src | 00:37 |
lisitsyn | would still take time you know | 00:37 |
lisitsyn | sonney2k: no the detection thing | 00:37 |
@sonney2k | lots of time I think | 00:37 |
lisitsyn | not moduling | 00:37 |
wiking | yeah | 00:37 |
lisitsyn | brb | 00:37 |
wiking | that is great that we have the src | 00:37 |
wiking | but with what lib and flags to compile | 00:37 |
wiking | that should be detected by cmake | 00:38 |
@sonney2k | lisitsyn, I don't get your list | 00:38 |
@sonney2k | is that a priority list of features you want to have or what? | 00:38 |
wiking | no | 00:38 |
wiking | just a random list | 00:38 |
@sonney2k | or general ideas? | 00:38 |
@sonney2k | ok | 00:38 |
wiking | afa i understand | 00:38 |
wiking | but ok | 00:39 |
wiking | what is that hoooorible hard thing to | 00:39 |
@sonney2k | also I don't understand 5) 6) | 00:39 |
wiking | detect with CMake | 00:39 |
@sonney2k | and 3) | 00:39 |
wiking | that is done now in ./configure | 00:39 |
@sonney2k | wiking, could you create some ticket 'shogun future' with these brainstormed ideas? | 00:40 |
@sonney2k | wiking, I think adding .equals() method and .clone() | 00:40 |
@sonney2k | would be necessary | 00:40 |
@sonney2k | then baseline algorithms ( random forests for classification / regression) | 00:41 |
@sonney2k | basic adaboost etc | 00:41 |
wiking | giiiiiiiiiiiiiiiiive meeeeeeeeeeeeeeeeeee clone() | 00:41 |
@sonney2k | then SGString -> SGReferencedData | 00:42 |
@HeikoS | wiking: I made some progress on it while we were talking | 00:42 |
@HeikoS | its coming .... | 00:42 |
@sonney2k | same for SGSparse* | 00:42 |
wiking | HeikoS: can'twait | 00:42 |
wiking | i waited a month man | 00:42 |
wiking | i mean seriously :D | 00:42 |
@sonney2k | wiking, please do the list! | 00:42 |
@sonney2k | wiking, seriously | 00:42 |
@HeikoS | wiking: sorry there was NIPS :) | 00:42 |
wiking | HeikoS: noexcuses :) | 00:42 |
@sonney2k | someone has to protocol that otherwise our brainstorming session is lost | 00:42 |
@sonney2k | wiking, yeah no excuses | 00:42 |
@sonney2k | wiking, then smart pointers | 00:43 |
@sonney2k | wiking, openmp too | 00:43 |
@sonney2k | wiking, but *NO* GPU, cluster stuff | 00:43 |
@sonney2k | so we focus on commodity machines | 00:43 |
@HeikoS | lets have a priority list for short /mid /long(ideas) | 00:44 |
@sonney2k | multicore | 00:44 |
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has quit [Ping timeout: 250 seconds] | 00:44 | |
@HeikoS | what do you think? | 00:44 |
@sonney2k | HeikoS, first collect the ideas | 00:44 |
@HeikoS | well yes | 00:44 |
@sonney2k | then we decide about priorities | 00:44 |
@sonney2k | wiking, also please add documention/howto/examples for each and every algorithm we have in shogun | 00:44 |
@sonney2k | and I mean reasonable | 00:45 |
lisitsyn | re | 00:45 |
@sonney2k | not just take data X not suited for a task | 00:45 |
wiking | sonney2k: i have 6 points now | 00:45 |
wiking | mmm ah yeah | 00:45 |
wiking | that's the 7th | 00:45 |
@sonney2k | HeikoS, btw have you seen how hard it is to keep documentation up-to-date? already our README* is outdated | 00:46 |
@HeikoS | sonney2k: yes | 00:46 |
@sonney2k | no one is keeping documentation up-to-date | 00:46 |
lisitsyn | documentation is impossible to keep up-to-date | 00:46 |
@HeikoS | I have a suggestion for this | 00:46 |
@HeikoS | *one* central place for documentation | 00:46 |
@HeikoS | preferbly web-based | 00:46 |
@sonney2k | my experience is that the only documentation (if at all) that will survive is in code | 00:46 |
@HeikoS | I loose track on where things are, too many places | 00:46 |
@sonney2k | but even there it is hard | 00:46 |
@HeikoS | mnuch easier if central | 00:46 |
@HeikoS | no readme files | 00:47 |
@HeikoS | this concept is so outdated | 00:47 |
wiking | sonney2k: https://gist.github.com/vigsterkr/5725617 | 00:47 |
@sonney2k | HeikoS, no we need readmes | 00:47 |
@HeikoS | rather have a webpage with installation/explanation/tutorial | 00:47 |
@HeikoS | class references, etc | 00:47 |
@HeikoS | can be generated from READMEs maybe? | 00:47 |
@sonney2k | wiking, you left out like half of the ideas | 00:47 |
wiking | sonney2k: that's why it's a gist...? | 00:48 |
@HeikoS | also the tutorial should be converted to weg based | 00:48 |
@HeikoS | too much effort this book stuff | 00:48 |
@HeikoS | but there is valuable stuff in there | 00:48 |
@HeikoS | (at least kernel MMD) | 00:48 |
lisitsyn | weg HeikoS? | 00:48 |
@HeikoS | with a wiki-like website, its also easier to update/maintain | 00:48 |
@sonney2k | wiking, then rather use a wiki | 00:48 |
@HeikoS | web | 00:48 |
wiking | sonney2k: just edit it... wont bite u | 00:49 |
lisitsyn | HeikoS: I'd use markdown | 00:49 |
@HeikoS | lisitsyn: whatever, just has to be easy | 00:49 |
lisitsyn | HeikoS: nothing easier than md | 00:49 |
@HeikoS | and we make it mandatory for all GSoC to write stuff there | 00:49 |
@HeikoS | *properly* | 00:50 |
@sonney2k | https://github.com/shogun-toolbox/shogun/wiki/Future-of-Shogun-Brainstorming | 00:50 |
@sonney2k | lets write stuff there | 00:50 |
wiking | sonney2k: https://github.com/shogun-toolbox/shogun/wiki/Feature-WishList | 00:51 |
wiking | ahhaha | 00:51 |
wiking | ok i'll delete it | 00:51 |
wiking | i mean mine | 00:51 |
wiking | no yours | 00:51 |
wiking | because mine is at least idented | 00:51 |
@sonney2k | wiking, no mine is :P | 00:51 |
wiking | hehehe | 00:52 |
wiking | ok start editing whichever | 00:52 |
wiking | the other will be deleted then | 00:52 |
wiking | ;) | 00:52 |
lisitsyn | wiking: sonney2k: HeikoS: what I do want to push is that we have to find a market slot | 00:53 |
lisitsyn | I don't want to be an average so we have to be the greatest in something | 00:53 |
wiking | :D | 00:54 |
@HeikoS | lisitsyn: good point! | 00:54 |
@HeikoS | maybe discussions? :D | 00:54 |
wiking | lisitsyn: well we are already the greatest in fucking around on irc channel ;) | 00:54 |
wiking | what else u want? :D | 00:54 |
lisitsyn | haha | 00:54 |
lisitsyn | the greatest discussions okay | 00:54 |
lisitsyn | I'd be interested in | 00:55 |
lisitsyn | fastest + lowest memory footprint | 00:55 |
lisitsyn | it is the only excuse to use C++ nowadays | 00:55 |
@sonney2k | lisitsyn, we are that for SVMs | 00:56 |
@sonney2k | lisitsyn, but you cannot have that for each and every algorithm | 00:56 |
lisitsyn | sonney2k: then we should sell it on the every page | 00:56 |
lisitsyn | sonney2k: shogun: the fastest svm in the west | 00:56 |
wiking | lisitsyn: u r in the east :) | 00:57 |
@sonney2k | lisitsyn, ok then olvier chapelle did back in his yahoo time an experiment where he trained distributed on > 1000 nodes | 00:57 |
lisitsyn | sonney2k: otherwise we are just another machine learning framework | 00:57 |
@sonney2k | the experiment I did in the coffin paper | 00:57 |
lisitsyn | jamlf | 00:57 |
@sonney2k | on a single core | 00:57 |
lisitsyn | wiking: west enough! | 00:57 |
lisitsyn | sonney2k: you lack PR then | 00:58 |
@sonney2k | it took him like 1 hour to finish and my single core variant (I don't recall exaclty) 1-2 days | 00:58 |
@sonney2k | another example. | 00:58 |
lisitsyn | sonney2k: I haven't seen people talking about coffin - that's a problem | 00:58 |
wiking | what... coffein?: ) | 00:59 |
wiking | yes pleeeease! | 00:59 |
lisitsyn | well I am not into research | 00:59 |
lisitsyn | but still | 00:59 |
@sonney2k | I trained on histogram based features computed on-the-fly on some mini desktop machine again in like several hundred million dimensional spaces | 00:59 |
@sonney2k | lisitsyn, shogun is known for large scale in SVMs and MKL | 00:59 |
@sonney2k | and structured output algorithms | 01:00 |
lisitsyn | sonney2k: then we have to make exactly that convenient for an user | 01:00 |
@sonney2k | yes we suck at advertising | 01:00 |
lisitsyn | sonney2k: then we could avoid basic algorithms at all | 01:00 |
@sonney2k | but that can only be fixed by writing cool applications /documentation / examples | 01:00 |
wiking | sonney2k: no we suck at fancy examples ;) | 01:00 |
@sonney2k | wiking, exactly | 01:00 |
lisitsyn | we have a lot of examples | 01:00 |
lisitsyn | and they are out of my control | 01:00 |
wiking | but we have a gsoc for that this year | 01:00 |
lisitsyn | I bet of yours too | 01:00 |
@sonney2k | lisitsyn, but nothing that is tuned to the algorithm underneath | 01:01 |
lisitsyn | sonney2k: we are getting big and useless like russia | 01:01 |
@sonney2k | lisitsyn, but van51 will do some cool ones this year for the dot features | 01:01 |
@HeikoS | hey guys, I am going home now | 01:01 |
@sonney2k | lisitsyn, not as long as we are our own customers | 01:01 |
@HeikoS | still 20 mins cycle ride ahead :) | 01:02 |
@sonney2k | HeikoS, wow still at work! | 01:02 |
@sonney2k | ?! | 01:02 |
lisitsyn | sonney2k: we should get customers first then | 01:02 |
lisitsyn | or make ourselves real customers | 01:02 |
wiking | HeikoS: havefun | 01:02 |
@sonney2k | lisitsyn, chicken and egg | 01:02 |
wiking | sonney2k: noup | 01:02 |
@HeikoS | well, was at the pub and returned to get my stuff, then this dicussion was going on ;) | 01:02 |
wiking | it's easy | 01:02 |
@sonney2k | lisitsyn, I am but not as often as I used to be | 01:02 |
wiking | there's ML everywhere u look around | 01:02 |
@sonney2k | lisitsyn, in academia I was doing all experiments with shogun | 01:02 |
@HeikoS | and wrote tome clone stuff on the fly, getting there | 01:02 |
wiking | the question is what is the thing one wants to try out first | 01:02 |
wiking | scikits | 01:02 |
wiking | or shogun | 01:02 |
@HeikoS | bye all! | 01:02 |
wiking | ;) | 01:02 |
@sonney2k | but maybe hushell and the others will pick up... | 01:03 |
@sonney2k | HeikoS, bye! | 01:03 |
lisitsyn | sonney2k: I haven't used shogun for real for a loooooong | 01:03 |
lisitsyn | how can I know what is good for ML | 01:03 |
@sonney2k | lisitsyn, well that is a problem | 01:03 |
lisitsyn | yes | 01:03 |
wiking | lisitsyn: dont be a bitch | 01:03 |
@sonney2k | lisitsyn, it is useless then for you | 01:03 |
wiking | quit the job | 01:03 |
wiking | and code shogun for free | 01:03 |
wiking | :D | 01:03 |
@sonney2k | no | 01:03 |
lisitsyn | wiking: haha! nice suggestion | 01:03 |
@sonney2k | get a job that needs ML | 01:03 |
wiking | lisitsyn: code 4 food | 01:04 |
-!- HeikoS [~heiko@nat-189-255.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 01:04 | |
lisitsyn | I enjoy my job now too | 01:04 |
@sonney2k | houston we have a problem | 01:04 |
wiking | lisitsyn: i bet u dont use ibm software :D | 01:04 |
@sonney2k | wiking, lisitsyn - I updated the wiki | 01:05 |
lisitsyn | sonney2k: I am not that tied to ML - I am interested in software engineering in general | 01:05 |
lisitsyn | doing great software is interesting | 01:05 |
wiking | sonney2k: what, lisitsyn enjoys his job....?! fuuuck! :D | 01:05 |
@sonney2k | any further ideas? | 01:05 |
@sonney2k | https://github.com/shogun-toolbox/shogun/wiki/Future-of-Shogun-Brainstorming | 01:05 |
lisitsyn | sonney2k: we will see | 01:06 |
@sonney2k | wiking, no not that but not needing ML for anything | 01:06 |
wiking | sonney2k: well he cannot drop shogun in a j2ee container :D | 01:06 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 01:06 | |
wiking | sonney2k: so what do u expect ? :) | 01:06 |
@sonney2k | hehe | 01:06 |
lisitsyn | wiking: I am not doing j2ee anymore do you know? ;) | 01:06 |
wiking | lisitsyn: well still | 01:07 |
lisitsyn | not EE at all | 01:07 |
wiking | lisitsyn: u cannot do it, or? :) | 01:07 |
wiking | even if u would like to | 01:07 |
wiking | there's no support for that | 01:07 |
@sonney2k | my experience is that this is all about motivation | 01:07 |
lisitsyn | wiking: I do distributed discrete optimization or so :) | 01:07 |
@sonney2k | motivation for me is usally - I want to do X | 01:07 |
wiking | sonney2k: yeah i'm motivated to have a begging machine for a month | 01:07 |
@sonney2k | for that I need Y | 01:07 |
lisitsyn | wiking: begging machine you got it | 01:07 |
@sonney2k | so I do Y | 01:07 |
wiking | lisitsyn: i'm begging now for a month | 01:07 |
@sonney2k | wiking, I can be your begging machine :D | 01:08 |
wiking | yeah i'm the begging machine itself | 01:08 |
wiking | begging for an api | 01:08 |
* sonney2k begs | 01:08 | |
@sonney2k | ohhh dear wiking give us the real shit | 01:08 |
wiking | i would love to | 01:08 |
wiking | but heiko is not giving any | 01:08 |
wiking | :))) | 01:08 |
@sonney2k | wiking, now it is heiko I know :D | 01:08 |
@sonney2k | lazy you! | 01:09 |
wiking | without him there's no real shit | 01:09 |
lisitsyn | okay I drop all branches now | 01:09 |
lisitsyn | and we start from the very beginning | 01:09 |
wiking | lisitsyn: wooooooooooooooooohooooooooooooooooo! | 01:09 |
wiking | git push upstream :* | 01:09 |
wiking | :> | 01:09 |
wiking | say good bye \. | 01:09 |
lisitsyn | I feel hair of sonney2k is getting gray | 01:10 |
@sonney2k | that is not going to happen :D | 01:10 |
@sonney2k | wiking, lisitsyn did you update the wiki? | 01:11 |
@sonney2k | what did I miss from this discussion? | 01:11 |
lisitsyn | sonney2k: I don't see anything missed | 01:11 |
@sonney2k | wiking, ? | 01:11 |
lisitsyn | I will put any crrazy idea I have | 01:12 |
@sonney2k | please do | 01:12 |
@sonney2k | we then have to prioritize | 01:12 |
lisitsyn | sonney2k: I will be working on one of them next days | 01:12 |
lisitsyn | you know which | 01:12 |
wiking | sonney2k: i've put those 7 ... i could't recall the others so i'm good with this list | 01:13 |
lisitsyn | currently I see that like: | 01:13 |
lisitsyn | metric.parameter("k").is_linear_range().with_bounds(1.0, 2.0) | 01:13 |
lisitsyn | no more tree please | 01:13 |
@sonney2k | lisitsyn, wait what is wrong with the tree? | 01:14 |
lisitsyn | sonney2k: user should not see the tree | 01:14 |
@sonney2k | lisitsyn, I thought you wanted to create the tree based on this syntax? | 01:14 |
lisitsyn | sonney2k: yes but no tree for the user | 01:15 |
wiking | lisitsyn: http://iridia.ulb.ac.be/irace/ | 01:15 |
lisitsyn | wiking: I don't give a heck what is it for :D | 01:15 |
lisitsyn | what race | 01:15 |
@sonney2k | racy! | 01:15 |
@sonney2k | as cryptic as shogun doc | 01:15 |
@sonney2k | oh well but we have formulas from time to time ;) | 01:16 |
lisitsyn | TSP | 01:16 |
lisitsyn | travelling salesman problem! | 01:16 |
@sonney2k | yeah well I got that | 01:16 |
lisitsyn | it must be something with NP | 01:16 |
lisitsyn | :D | 01:16 |
@sonney2k | but what is the iterated f-race procedure | 01:16 |
@sonney2k | and i-race? | 01:16 |
@sonney2k | and extens | 01:16 |
@sonney2k | excellent!~ | 01:16 |
lisitsyn | sonney2k: something very important for racing problem | 01:16 |
@sonney2k | maybe tomtom should use this :D | 01:17 |
wiking | sonney2k: it's actually a good sw for doing model selection | 01:17 |
wiking | good (apart from being implemented in R) | 01:17 |
@sonney2k | lisitsyn, back to the syntax | 01:17 |
lisitsyn | wiking: oh so you understand what happens here? | 01:17 |
lisitsyn | explain to us!:) | 01:17 |
wiking | lisitsyn: yeah | 01:17 |
@sonney2k | lisitsyn, so you would do obj.parameter* etc? | 01:17 |
wiking | lisitsyn: i'm using this shit for finding the optimal parameters | 01:17 |
lisitsyn | sonney2k: yes | 01:17 |
lisitsyn | wiking: how? | 01:18 |
@sonney2k | so you add to sgobject a parameter method right? | 01:18 |
lisitsyn | sonney2k: yes | 01:18 |
@sonney2k | that then can look up in the MS parameter list the parameter | 01:18 |
lisitsyn | sonney2k: and if one sgobject sets other object as some value | 01:18 |
wiking | lisitsyn: it's a pretty simple to use | 01:18 |
lisitsyn | sonney2k: it merges the tree | 01:18 |
@sonney2k | and then for the parameter you can get ranges etc | 01:18 |
wiking | lisitsyn: but the theoretical background behind it is strong actually :P | 01:18 |
lisitsyn | sonney2k: yes you declare what is the values | 01:19 |
lisitsyn | range or set of values | 01:19 |
@sonney2k | lisitsyn, wait where is the tree stored? | 01:19 |
lisitsyn | sonney2k: each object has node | 01:19 |
lisitsyn | so the tree starts from the object you start the tree | 01:19 |
@sonney2k | each sgobject is a node? | 01:19 |
lisitsyn | yes | 01:19 |
wiking | lisitsyn: and yes this runs on a cluster ;) | 01:19 |
@sonney2k | lisitsyn, so you kill the current tree class? | 01:19 |
lisitsyn | sonney2k: no re-use it | 01:20 |
lisitsyn | sonney2k: I wouldn't like to touch anything there | 01:20 |
lisitsyn | just make it work with smallest possible changes | 01:20 |
@sonney2k | lisitsyn, do you already have a feature branch to look at? | 01:20 |
lisitsyn | sonney2k: yes but I didn't start with that thing | 01:20 |
@sonney2k | I mean it sounds reasonable so just do it | 01:20 |
lisitsyn | sonney2k: https://github.com/shogun-toolbox/shogun/tree/feature/model_selection_syntax | 01:20 |
lisitsyn | I will push things here | 01:20 |
lisitsyn | 47 ····classifier=LibSVM() | 01:21 |
lisitsyn | + 48 ····classifier.parameter("C1").is_exp_range().with_bounds(-1.0, 1.0) | 01:21 |
lisitsyn | + 49 ····classifier.parameter("C2").is_exp_range().with_bounds(-1.0, 1.0) | 01:21 |
lisitsyn | + 50 | 01:21 |
lisitsyn | + 51 ····gaussian_kernel = GaussianKernel() | 01:21 |
lisitsyn | + 52 ····gaussian_kernel.parameter("width").is_exp_range().with_bounds(0.0, 10.0) | 01:21 |
lisitsyn | + 53 | 01:21 |
lisitsyn | + 54 ····power_kernel = PowerKernel() | 01:21 |
lisitsyn | + 55 ····power_kernel.parameter("degree").is_linear_range().with_bounds(1.0, 2.0) | 01:21 |
lisitsyn | + 56 | 01:21 |
lisitsyn | + 57 ····metric = MinkowskiMetric(10) | 01:21 |
lisitsyn | + 58 ····metric.parameter("k").is_linear_range().with_bounds(1.0, 2.0) | 01:21 |
@sonney2k | I don't get yet though how the nodes and the tree comes into play | 01:21 |
@sonney2k | lisitsyn, we totally agree on the syntax | 01:21 |
lisitsyn | sonney2k: classifier.parameter("kernel").is_set([gaussian_kernel, power_kernel]) | 01:21 |
@sonney2k | lisitsyn, btw can you do setters/getters via that too? | 01:21 |
lisitsyn | sonney2k: here the nodes should get connected | 01:21 |
@sonney2k | I mean plain setters/getters? | 01:21 |
lisitsyn | sonney2k: plain setters for what? | 01:22 |
@sonney2k | lisitsyn, for the parameters | 01:22 |
lisitsyn | what should it set/get? | 01:22 |
lisitsyn | modelselection vlaues? | 01:22 |
@sonney2k | say I want to do gaussiankernel.parameter('width')=13 | 01:22 |
lisitsyn | ahh I didn't expect it to use for that | 01:23 |
lisitsyn | just for model selection so far | 01:23 |
@sonney2k | it would make sense though | 01:23 |
@sonney2k | I don't mind a .get | 01:23 |
@sonney2k | .set | 01:23 |
@sonney2k | gaussiankernel.parameter('width').set(13) | 01:23 |
lisitsyn | I hate getters and setters they should go | 01:23 |
@sonney2k | it would just be very convenient to iterate over parameters e.g. from python and be able to set things based on some dictionary | 01:24 |
@sonney2k | lisitsyn, for example if you have some elaborate model selection to do | 01:24 |
@sonney2k | you could have some python dict with parameter name as a key | 01:24 |
@sonney2k | and then a parameter(key).set(d[key][i]) would be *very* handy | 01:25 |
lisitsyn | sonney2k: may be we should just drop setters and getters | 01:25 |
lisitsyn | sonney2k: in python we could use [] | 01:25 |
lisitsyn | svm["C"] = 3 | 01:26 |
@sonney2k | lisitsyn, some things might need speedy parameter access. string based access certainly is not. but it is a great idea :) | 01:26 |
@sonney2k | svm[k].set(3) | 01:26 |
@sonney2k | is cool | 01:26 |
lisitsyn | sonney2k: we could use enums here | 01:27 |
@sonney2k | anyway I understand the node connection thing now | 01:27 |
@sonney2k | lisitsyn, but where does the tree come into play | 01:27 |
lisitsyn | sonney2k: kernel has ModelSelectionParameters() instance | 01:27 |
@sonney2k | I mean you have to generate it at some point | 01:27 |
lisitsyn | so once you attach it to classifier | 01:27 |
lisitsyn | it just adds that instance to the root | 01:27 |
lisitsyn | classifier.parameter("kernel").is_set([gaussian_kernel, power_kernel])) | 01:28 |
lisitsyn | it is the same to | 01:28 |
@sonney2k | and then it can do the expansion and do MS | 01:28 |
lisitsyn | gkmsp = ModelSelectionParameters("kernel", gaussian_kernel) | 01:28 |
lisitsyn | pkmsp = ModelSelectionParameters("kernel", power_kernel) | 01:28 |
lisitsyn | root.add_child(gkmsp) | 01:28 |
lisitsyn | root.add_child(pkmsp) | 01:29 |
@sonney2k | I seebreb | 01:29 |
@sonney2k | brb | 01:29 |
lisitsyn | hah | 01:29 |
lisitsyn | looks like a working idea (I hope so) | 01:29 |
lisitsyn | wiking: but how does it work? | 01:29 |
lisitsyn | is that grid search in worst case | 01:30 |
lisitsyn | or what? | 01:30 |
lisitsyn | really cryptic | 01:31 |
lisitsyn | :) | 01:31 |
lisitsyn | uh it is getting really late | 01:32 |
lisitsyn | see yoou later | 01:33 |
-!- lisitsyn [~blackburn@188.168.14.83] has left #shogun [] | 01:33 | |
@sonney2k | RE | 01:35 |
@sonney2k | wiking, still there? | 01:35 |
@sonney2k | wiking, why are you not OP? | 01:36 |
wiking | ? | 01:36 |
-!- mode/#shogun [+o wiking] by sonney2k | 01:36 | |
wiking | mmm | 01:36 |
-!- mode/#shogun [-o wiking] by ChanServ | 01:36 | |
@sonney2k | hah | 01:36 |
wiking | i guess i was splitted | 01:36 |
@sonney2k | yeah | 01:36 |
-!- wiking [~wiking@info2k1.hu] has quit [Changing host] | 01:36 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 01:36 | |
-!- mode/#shogun [+o wiking] by ChanServ | 01:36 | |
@wiking | now | 01:36 |
@wiking | :) | 01:36 |
@sonney2k | wiking, I think we should have this kind of discussion once a month or so | 01:36 |
@wiking | lisityn i'll give u a paper to read about irace | 01:36 |
@sonney2k | wiking, he is asleep already | 01:37 |
@wiking | yeah | 01:37 |
@wiking | but he reads logs | 01:37 |
@wiking | ;) | 01:37 |
@wiking | *always | 01:37 |
@wiking | * | 01:37 |
@wiking | :> | 01:37 |
@sonney2k | I like the MS thing very much | 01:37 |
@sonney2k | lets hope it is as easy as it sounds | 01:37 |
@wiking | well i'm using this as it really proven to be faster than the brute force method | 01:37 |
@wiking | although it still takes a lot of iterations | 01:38 |
@sonney2k | wiking, btw will you bring back the ubu1 buildslave? | 01:38 |
@sonney2k | ohh you did | 01:38 |
@wiking | sonney2k: ;) | 01:38 |
@sonney2k | let me do the rpm | 01:38 |
@wiking | sonney2k: the server got online today finally | 01:39 |
@sonney2k | btw for the record. libshogun compiletime (from scratch) here is like 53 sec | 01:40 |
@wiking | woah | 01:40 |
@wiking | that's fast | 01:40 |
@sonney2k | with optimizations 1:53 | 01:40 |
@wiking | -j16 ? :) | 01:40 |
@sonney2k | core i7 desktop | 01:41 |
@sonney2k | wiking, back when I was at max planck I used -j 32 :D | 01:41 |
@sonney2k | compile time was never an issue there | 01:41 |
@wiking | :> | 01:42 |
@wiking | ok time to crash | 01:43 |
@wiking | gotta wake up early tomorrow | 01:43 |
@sonney2k | rpm1 is back online | 01:44 |
@sonney2k | wiking, true true I won't get even 4 hours sleep again | 01:45 |
@sonney2k | iglesiasg, still awake? | 01:45 |
@iglesiasg | sonney2k: yeah | 01:46 |
@sonney2k | iglesiasg, I have a job for you :D | 01:46 |
@iglesiasg | always awake :S | 01:46 |
@sonney2k | iglesiasg, excellent | 01:46 |
@sonney2k | you are my man then :D | 01:46 |
@iglesiasg | haha not so excellent actually! | 01:47 |
@sonney2k | I need some help | 01:47 |
@iglesiasg | I hope you don't need it now | 01:47 |
@sonney2k | you know we have mailinglists right? | 01:47 |
@sonney2k | so I need someone to manage them | 01:47 |
@iglesiasg | I know we have a mailing list | 01:47 |
@iglesiasg | what does it involve? | 01:47 |
@sonney2k | as in whenever there are bounces etc there is an admin console where you can get mails pass through or discard them | 01:48 |
@sonney2k | not a lot of work but I am administrating the servers and buildbot etc already | 01:48 |
@sonney2k | so I was hoping that you could do that | 01:48 |
@sonney2k | you basically get an email when an email arrives from a *non-subscribed* user | 01:48 |
@iglesiasg | yeah, I am happy to help | 01:48 |
@sonney2k | that is rather rare but occurs | 01:49 |
@iglesiasg | but I still don't have very clear what is it that I would do | 01:49 |
@iglesiasg | aham | 01:49 |
@sonney2k | iglesiasg, basically click accept/reject | 01:49 |
@iglesiasg | ok | 01:49 |
@iglesiasg | sonney2k: so I would get a mail in my account when this happens or do I have to log in somewhere? | 01:50 |
@sonney2k | I have been doing that for some weeks now since I set up the new mail server | 01:50 |
@sonney2k | iglesiasg, exactly you get a mail with a link | 01:50 |
@sonney2k | iglesiasg, I would just make you the list administrator | 01:50 |
@sonney2k | (not now but soon) | 01:50 |
@iglesiasg | sonney2k: ok, I am in then :) | 01:51 |
@sonney2k | that is for our two lists - the support ML and the workshop ml | 01:51 |
@iglesiasg | sonney2k: do you use any filter for them so they go directly to any specific folder in your mailbox? | 01:51 |
@sonney2k | iglesiasg, no but you could do that | 02:05 |
@sonney2k | it is not like a message per day | 02:06 |
@sonney2k | maybe 3 / week | 02:06 |
@iglesiasg | ah ok | 02:13 |
@iglesiasg | I thought they were more | 02:13 |
@sonney2k | iglesiasg, you never know | 02:26 |
* sonney2k ZZzzzz. | 02:26 | |
@iglesiasg | good night | 02:27 |
-!- zxtx [~zv@rrcs-74-62-200-195.west.biz.rr.com] has quit [Ping timeout: 248 seconds] | 03:25 | |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has quit [Ping timeout: 260 seconds] | 03:52 | |
-!- nube [~rho@49.244.14.162] has quit [Quit: Leaving.] | 03:54 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:5c74:1501:97b2:f99d] has joined #shogun | 04:40 | |
-!- nube [~rho@116.90.239.3] has joined #shogun | 05:48 | |
@iglesiasg | time to sleep! :) | 05:59 |
-!- iglesiasg [d58f321f@gateway/web/freenode/ip.213.143.50.31] has quit [Quit: Page closed] | 05:59 | |
-!- hushell [~hushell@c-24-21-141-32.hsd1.or.comcast.net] has joined #shogun | 06:00 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:5c74:1501:97b2:f99d] has quit [Remote host closed the connection] | 06:14 | |
-!- gsomix [~Miranda@5850-AMTS-1-96.dialup.samtel.ru] has joined #shogun | 07:37 | |
gsomix | good morning | 07:37 |
@sonney2k | looks like iglesiasg's night shift is over :D | 08:07 |
@sonney2k | moin thoralf | 08:07 |
@sonney2k | hey gsomix! | 08:08 |
@sonney2k | gsomix, when is your next exam? | 08:08 |
gsomix | sonney2k: next little pass-fail exam at monday... and there is big exam at 19 June. so, I'm at home and working now. | 08:10 |
@sonney2k | ohh man good luck | 08:18 |
gsomix | sonney2k: tnx. :) | 08:20 |
@sonney2k | alrighty brb | 08:25 |
@sonney2k | sonne|work, here I come! | 08:25 |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 08:32 | |
sonne|work | sonney2k: ohh there you are! | 08:38 |
thoralf | moin sonne|work | 09:46 |
sonne|work | thoralf: hey | 09:48 |
thoralf | sonne|work: Did you notice my comments (yesterday) about CKernel's lhs_equals_rhs? | 09:50 |
thoralf | sonne|work: Seems that you introduced it, but did not initialize in init(). | 09:50 |
sonne|work | afk | 09:51 |
-!- gsomix [~Miranda@5850-AMTS-1-96.dialup.samtel.ru] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 09:52 | |
-!- lambday [67157d36@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.54] has joined #shogun | 09:57 | |
lambday | good morning sonney2k :) | 09:58 |
sonne|work | lambda*y! | 10:29 |
sonne|work | hey there | 10:29 |
sonne|work | thoralf: no I didn't whats up? | 10:29 |
lambday | sonne|work: hey :) | 10:30 |
thoralf | sonne|work: You should add "lhs_equals_rhs=false;" to CKernel::init() (Kernel.cpp:962) | 10:31 |
thoralf | sonne|work: Makes about 1/8 of all valgrind warnings ;) | 10:31 |
sonne|work | thoralf: just do it! | 10:31 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 10:36 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 10:36 | |
sonne|work | hey lisitsyn :) | 10:44 |
-!- foulwall [~foulwall@2001:da8:215:503:7d52:fdce:12e6:9407] has joined #shogun | 10:47 | |
@lisitsyn | sonne|work: hey | 10:48 |
sonne|work | lisitsyn: any hangovers? | 10:51 |
sonne|work | I slept like 3 hours... | 10:51 |
sonne|work | foulwall: oh hey nice to see you here! | 10:51 |
sonne|work | foulwall: I am wondering if everything is OK? | 10:51 |
sonne|work | foulwall: and btw when is you next exam (again?) | 10:52 |
@lisitsyn | sonne|work: hah no I am totally ok | 10:54 |
@lisitsyn | sonne|work: I don't have to wake up early morning | 10:54 |
@lisitsyn | thankfully no kids yet | 10:55 |
@lisitsyn | :D | 10:55 |
-!- nube [~rho@116.90.239.3] has quit [Quit: Leaving.] | 10:57 | |
@lisitsyn | sonne|work: are you using anything for charts at your job? | 11:06 |
sonne|work | lisitsyn: d3 matplotlib | 11:07 |
@lisitsyn | sonne|work: how do you integrate it with java? | 11:07 |
sonne|work | custom code yes... | 11:07 |
@lisitsyn | sonne|work: ever stuggled with jfreechart? ;) | 11:07 |
@lisitsyn | sonne|work: custom code like webserver or do you integrate it with swing? | 11:08 |
sonne|work | yeah some web server | 11:09 |
@lisitsyn | I see thnaks | 11:14 |
@lisitsyn | sonne|work: then I know what I will push our new employee too *evil laugh* | 11:14 |
@lisitsyn | to* | 11:14 |
foulwall | sonne|work: hey! next exam will be on 11, June, and I almost finished the boring preparation. tomorrow I'll be back to code. these exams are scattered from 11-20. | 11:42 |
foulwall | sonne|work: 4 left exams are on 11, 13, 17, 20. | 11:43 |
sonne|work | foulwall: I fully understand that you will have little time during this period | 11:43 |
sonne|work | so good luck! | 11:43 |
foulwall | sonne|work: thanks. | 11:44 |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 11:47 | |
shogun-notifier- | shogun: Thoralf Klein :develop * 4055e99 / src/shogun/kernel/Kernel.cpp,tests/unit/kernel/CustomKernel_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/4055e99fcc324d85c3badce8df6892deb4dcae5c | 11:47 |
shogun-notifier- | shogun: CKernel: Fixed uninitialized "lhs_equals_rhs" | 11:47 |
shogun-notifier- | shogun: CCustomKernel: Fixed memory leak in unit test. | 11:47 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 87a3fce / src/shogun/kernel/Kernel.cpp,tests/unit/kernel/CustomKernel_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/87a3fcebceaedf8a82785e406b59301f32779687 | 11:47 |
shogun-notifier- | shogun: Merge pull request #1161 from tklein23/develop | 11:47 |
shogun-notifier- | shogun: | 11:47 |
shogun-notifier- | shogun: CKernel: Fixed uninitialized "lhs_equals_rhs", fixed memory leak in CustomKernel test. | 11:47 |
shogun-buildbot | build #924 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/924 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 11:52 |
@wiking | lisitsyn: cmake branch here we go :) | 11:57 |
shogun-buildbot | build #925 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/925 blamelist: Thoralf Klein <thoralf.klein@zib.de> | 12:02 |
-!- HeikoS [~heiko@nat-184-237.internal.eduroam.ucl.ac.uk] has joined #shogun | 12:24 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 12:24 | |
lambday | HeikoS: hi | 12:25 |
@HeikoS | lambday: hi! | 12:25 |
lambday | for json I need json-c, right? | 12:25 |
@HeikoS | I think so, I havent got it installed on my machine | 12:25 |
lambday | I'm a stupid! | 12:25 |
lambday | :( | 12:25 |
@HeikoS | lambday: whats wrong? | 12:25 |
lambday | na its fine now.. last night I lost my mind and was wondering why it isn't compiling.. now I see I forgot to check if json exists even :D | 12:26 |
lambday | it doesn't... installing | 12:26 |
@HeikoS | lambday: never loose your mind over shogun, thats unhealthy :D | 12:27 |
lambday | HeikoS: haha :D | 12:28 |
lambday | actually, lost it to alcohol :D | 12:28 |
@HeikoS | lambday: oh, well thats also unhealthy ;) but no comment on that | 12:30 |
lambday | :) | 12:30 |
@HeikoS | lambday: did you send any PRs? | 12:33 |
lambday | HeikoS: no.. facing similar problem with json as ascii | 12:35 |
lambday | trying to figure it out | 12:35 |
@HeikoS | lambday: cool! | 12:35 |
@HeikoS | keep on! | 12:36 |
@HeikoS | your work on this is really appreciated btw! | 12:36 |
lambday | HeikoS: thanks... I wish I could get it done :( | 12:36 |
@HeikoS | lambday: patience | 12:36 |
lambday | HeikoS: https://gist.github.com/lambday/5728461 | 12:39 |
lambday | I'm again doing something wrong | 12:39 |
@HeikoS | trying | 12:40 |
@HeikoS | I first have to install json | 12:40 |
@HeikoS | do you know which debian package that is? | 12:40 |
lambday | json-c-devel | 12:40 |
lambday | umm... I use fedora | 12:40 |
@HeikoS | fedora | 12:40 |
lambday | json-c and json-c-devel you need both probably | 12:40 |
@HeikoS | really? | 12:40 |
lambday | yeah! | 12:40 |
@HeikoS | wow I have a collegue who is using that and has massive problems getting shoung to work | 12:40 |
@HeikoS | a tutorial would be highly appreciated :D | 12:41 |
lambday | HeikoS: I didn't face much :-/ whenever I'm stuck I just got up asking you or lisitsyn :D | 12:41 |
@HeikoS | really? | 12:41 |
@HeikoS | maybe he has to come to irc then | 12:41 |
lambday | hehe | 12:41 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun | 12:42 | |
@HeikoS | lambday: compiling | 12:44 |
lambday | hmm.. compiles fine | 12:44 |
lambday | but reads wrong | 12:44 |
@HeikoS | lambday: I mean recompiling shougn with json support | 12:44 |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 12:44 | |
lambday | okay | 12:44 |
lambday | once this works, I'll add all ptype tests for ascii and json.. won't take long I guess | 12:46 |
lambday | HeikoS: works? | 12:50 |
@HeikoS | json is annoying me | 12:51 |
@HeikoS | path is different in ubuntu | 12:51 |
@HeikoS | its <json/json.h> | 12:51 |
lambday | hmm...worked smooth on fedora :P | 12:51 |
@HeikoS | where is json.h on your machine? | 12:51 |
@HeikoS | lambday: its the accuracy | 12:52 |
@HeikoS | enable debug messages | 12:52 |
@HeikoS | and run the program on its own | 12:52 |
@HeikoS | Ill send a gist | 12:53 |
@HeikoS | https://gist.github.com/karlnapf/5728519 | 12:53 |
lambday | mine is in /usr/include/json/json.h | 12:53 |
lambday | okay | 12:53 |
@HeikoS | lambday: json is crap | 12:54 |
@HeikoS | accuracy of serialisation is 1E-6 | 12:54 |
@HeikoS | come on! | 12:54 |
@HeikoS | sonney2k: ^ | 12:54 |
@HeikoS | what is that? | 12:54 |
lambday | super crap :( | 12:55 |
lambday | ascii works fine with 0.0 | 12:55 |
@HeikoS | lambday: indeed | 12:55 |
@HeikoS | well just set accuracy to 1E-6 and the test will pass | 12:55 |
lambday | yeah.. | 12:55 |
lambday | passed | 12:56 |
lambday | phew! | 12:56 |
@iglesiasg | hushell: hi! | 13:02 |
lambday | HeikoS: char doesn't work for json | 13:42 |
lambday | HeikoS: https://gist.github.com/lambday/5728708 | 13:44 |
lambday | but it worked for ascii | 13:45 |
lambday | couldn't get any of the json tests running that uses "json_object_new_int" :( | 13:55 |
lambday | bool, floats work fine | 13:55 |
@HeikoS | lambday: print the field after loading and have a look yourself to see whats wrong | 13:56 |
@HeikoS | json seems to suck hard | 13:56 |
@HeikoS | sigh | 13:56 |
@HeikoS | annoying | 13:56 |
lambday | HeikoS: I did... the gist that I just pasted.. added the debug msg in a comment | 13:56 |
lambday | :( | 13:56 |
@HeikoS | yeah saw it | 13:56 |
@HeikoS | maybe just put an not implemented there (with a message json currently not supported for complex64 | 13:57 |
lambday | where? | 13:57 |
lambday | complex64_t is not supported but rest are supposed to work, no? | 13:58 |
@HeikoS | yeah | 13:58 |
@HeikoS | you probably modified the json writer? | 13:59 |
@HeikoS | you could put a warning in there | 13:59 |
lambday | for all the types that I couldn't get it worked with? | 13:59 |
lambday | basically its all types that json uses int for | 13:59 |
lambday | writing is fine | 14:00 |
lambday | reading is the problem I think | 14:00 |
lambday | sending the PR with what I've added so far.. rest, will add as you suggest | 14:01 |
@HeikoS | okay! | 14:02 |
-!- foulwall [~foulwall@2001:da8:215:503:7d52:fdce:12e6:9407] has quit [Remote host closed the connection] | 14:03 | |
lambday | HeikoS: please check | 14:05 |
lambday | HeikoS: I'll be back in a min... going to have a tea (chai :P) | 14:08 |
@HeikoS | enjoy! :) | 14:08 |
lambday | back | 14:12 |
lambday | ah, its raining here :D | 14:12 |
lambday | I want to add some tests for xml too | 14:12 |
lambday | I added the serializations stuff for xml and wanna see if they work | 14:13 |
lambday | for complex I mean | 14:13 |
-!- nube [~rho@49.244.7.173] has joined #shogun | 14:27 | |
sonne|work | HeikoS: he regarding your email. I think this is exactly what I heard from researchers... | 14:29 |
sonne|work | lisitsyn, HeikoS, wiking - we should have such discussion/planning sessions once per month or so. I think that was very useful / productive | 14:30 |
@HeikoS | sonne|work: agreed! | 14:31 |
shogun-notifier- | shogun: lambday :develop * 68690e2 / tests/unit/io/Serialization_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/68690e2d868e0bec77db017256f36c05bfacc002 | 14:31 |
shogun-notifier- | shogun: addded unit-tests for serialization | 14:31 |
shogun-notifier- | shogun: Heiko Strathmann :develop * b0873e1 / tests/unit/io/Serialization_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/b0873e1a83200071d7ac4cfc145b94e9a3728563 | 14:31 |
shogun-notifier- | shogun: Merge pull request #1162 from lambday/develop | 14:31 |
shogun-notifier- | shogun: | 14:31 |
shogun-notifier- | shogun: addded unit-tests for serialization | 14:31 |
@HeikoS | sonne|work: I still have this re-write idea floating around in my head ... | 14:33 |
sonne|work | HeikoS: rewrite what? | 14:33 |
@HeikoS | mplayer2 like | 14:34 |
@HeikoS | sonne|work: but later, | 14:34 |
sonne|work | you mean total redesign and start from scratch? | 14:34 |
@HeikoS | not really re-design | 14:34 |
sonne|work | or smart pointers or what? | 14:34 |
@HeikoS | design is in our heads | 14:34 |
@HeikoS | but rather re-write and implement all the nice things we know | 14:34 |
sonne|work | I don't understand what you mean | 14:35 |
-!- nube [~rho@49.244.7.173] has quit [Quit: Leaving.] | 14:37 | |
@lisitsyn | sonne|work: like starting with a better core | 14:44 |
@lisitsyn | that installs everywhere | 14:44 |
sonne|work | lisitsyn: core as in? | 14:48 |
shogun-buildbot | build #926 of cyg1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/926 blamelist: lambday <heavensdevil6909@gmail.com> | 14:49 |
shogun-buildbot | build #927 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/927 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 14:50 |
@wiking | HeikoS: how's clone? :) | 15:06 |
@HeikoS | wiking: wokring on it | 15:07 |
@lisitsyn | sonney2k: something very core like no algorithm at all | 15:34 |
@lisitsyn | sonne|work: have you ever considered shogun as python - C++ - matlab bridge? | 15:35 |
@lisitsyn | X - C++ - Y | 15:35 |
sonne|work | lisitsyn: yes | 15:35 |
@lisitsyn | sonne|work: may be that's the way | 15:36 |
sonne|work | yes that is one of its key strengths | 15:36 |
@lisitsyn | sonne|work: writing in C++ is pain | 15:36 |
@lisitsyn | sonne|work: well if we had something to do that | 15:36 |
@lisitsyn | currently we have * - C++ | 15:36 |
sonne|work | now I am lost | 15:37 |
sonne|work | X -> C++ -> Y ? | 15:37 |
@lisitsyn | if we had base classes/functions for matlab, python and etc | 15:37 |
@lisitsyn | sonne|work: yes | 15:37 |
@lisitsyn | we just take matlab code and adapt it to our predesigned interface | 15:37 |
sonne|work | that is the elwms hack crap | 15:37 |
@lisitsyn | sonne|work: well then we won't have to write research code in C++ | 15:38 |
@lisitsyn | like matlab plugins | 15:38 |
@lisitsyn | python plugins | 15:38 |
@lisitsyn | sonne|work: I am fascinated by that idea | 15:39 |
sonne|work | I am lost now | 15:39 |
@lisitsyn | sonne|work: shogun as a *bridge* between target languages | 15:39 |
sonne|work | what will it do then? | 15:39 |
@lisitsyn | sonne|work: handle all the data things etc | 15:39 |
@lisitsyn | sonne|work: like platform | 15:40 |
@lisitsyn | to run other code and give you results in target language | 15:40 |
sonne|work | you mean you start python | 15:40 |
sonne|work | and run something via shogun in matlab | 15:40 |
@lisitsyn | yes | 15:40 |
sonne|work | and get the result back? | 15:40 |
@lisitsyn | yes | 15:40 |
sonne|work | or vice versa | 15:40 |
sonne|work | yeah that is elwms | 15:40 |
@lisitsyn | well that doesn't mean we drop C++ code | 15:41 |
@lisitsyn | but we just write well C++ code | 15:41 |
@lisitsyn | for something really relevant | 15:41 |
@lisitsyn | that really needs speed | 15:41 |
@lisitsyn | sonne|work: well that's for training mainly | 15:41 |
sonne|work | lisitsyn: I thought someone needs this | 15:44 |
sonne|work | but hmmh very rarely people do | 15:44 |
@lisitsyn | sonne|work: I see that attaching an algorithm implemented in matlab is easy | 15:44 |
sonne|work | lisitsyn: people usually work in *one* language | 15:44 |
sonne|work | they don't care at all about others | 15:44 |
@lisitsyn | sonne|work: well I don't use matlab | 15:44 |
@lisitsyn | but I have to use matlab code sometimes | 15:45 |
@lisitsyn | to try something or so | 15:45 |
sonne|work | yeah and the issue is that you cannot do python -> matlab | 15:45 |
sonne|work | only matlab -> python | 15:45 |
@lisitsyn | why not python matlab? | 15:45 |
@lisitsyn | I find that easy | 15:45 |
sonne|work | you cannot run it embedded | 15:45 |
@lisitsyn | nobody needs it to be run embedded | 15:45 |
@lisitsyn | you just store your data in files then | 15:45 |
@lisitsyn | and load it | 15:45 |
sonne|work | so you would start a matlab process every time you do somethign? | 15:46 |
@lisitsyn | and compute and save again | 15:46 |
@lisitsyn | yes | 15:46 |
sonne|work | slow! | 15:46 |
@lisitsyn | sonne|work: well what is it comparing to time you would spend rewriting it? | 15:46 |
@lisitsyn | sonne|work: I would like to try LMNN but it is >1k lines of matlab | 15:47 |
sonne|work | sure but *I* don't have a use case for that | 15:47 |
@lisitsyn | I could just set matrices with some predefined functions | 15:47 |
@lisitsyn | and run it via that platform | 15:47 |
@lisitsyn | HeikoS: do you have such cases? | 15:47 |
sonne|work | I wrote exactly that for the languages, R, octave, matlab, python in elwms | 15:48 |
sonne|work | only that you cannot call matlab because embedded is not possible | 15:48 |
sonne|work | lisitsyn: anyway we have all the code for that in shogun/ui - and I would rather prefer killing our static interfaces having some kind of embedded front-end for the old sg() command | 15:50 |
@lisitsyn | sonne|work: no sg() command, this should be object-oriented | 15:51 |
sonne|work | lisitsyn: you want to keep the old users around | 15:51 |
@lisitsyn | sonne|work: yes sure | 15:52 |
@lisitsyn | sonne|work: what HeikoS pushes more | 15:52 |
@lisitsyn | is that we should be terribly easy to install | 15:52 |
sonne|work | yes but is that even possible with C++? | 15:52 |
sonne|work | I mean all we can do is add executables for each and every system | 15:53 |
sonne|work | or packages | 15:53 |
sonne|work | like the vlc guys are doing | 15:53 |
@lisitsyn | sonne|work: well people can wait, they just want to do less | 15:53 |
@lisitsyn | sonne|work: you know installing any python package with its dependencies is easy | 15:55 |
@lisitsyn | we've got to strive for the same | 15:55 |
sonne|work | lisitsyn: yeah but we have to compile | 15:55 |
sonne|work | and that needs tons of dependencies | 15:55 |
@lisitsyn | sonne|work: because HeikoS reported quite a few times people turn over because they don't know how to install that | 15:55 |
sonne|work | like a working compiler | 15:55 |
sonne|work | and people don't want to do that | 15:56 |
sonne|work | lisitsyn: I had the very same issues with people in my lab back then. they all were only able to start matlab and write A'X | 15:59 |
sonne|work | well not all of course | 15:59 |
sonne|work | but you know what I mean :) | 15:59 |
thoralf | :D | 15:59 |
sonne|work | if you can give them something that works out of the box they will use it | 16:00 |
thoralf | sonne|work: In your lab? You're not at TUB any more, right? | 16:01 |
sonne|work | thoralf: past tense yes, but it is the same everywhere in ML labs | 16:01 |
-!- vgorbati [~vgorbati@212.2.159.34] has joined #shogun | 16:15 | |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Quit: Page closed] | 16:18 | |
@wiking | sonne|work: ok now i have cmakefile | 16:23 |
sonne|work | wiking: doing what? | 16:38 |
@wiking | sonne|work: doing the same as ./configure | 16:49 |
sonne|work | wiking: then commit and let us try it. | 16:50 |
sonne|work | I cannot believe that though :D | 16:50 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 17:05 | |
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has quit [Quit: Leaving.] | 17:05 | |
@wiking | sonney2k: well it actually detects everything apart from cpuid | 17:13 |
-!- nube [~rho@49.244.7.173] has joined #shogun | 17:16 | |
@sonney2k | wiking, and it then generates what? | 17:28 |
@sonney2k | I mean can we compile libshogun / interfaces already? | 17:28 |
@sonney2k | lambday, wow | 17:28 |
@sonney2k | lambday, so serialization works for everything except char/int json right? | 17:28 |
@wiking | sonney2k: it generates libshogun | 17:30 |
@wiking | sonney2k: it detects java, python etc + swig | 17:30 |
@wiking | but the swig compilation and the option to turn on/off an interface is not yet ready | 17:30 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 17:31 | |
@sonney2k | wiking, ok. That is the easy part though! | 17:33 |
@wiking | well cpuid i've seen somewhere integrated into cmake | 17:35 |
@wiking | and the interfaces with cmake+swig is here: http://www.swig.org/Doc1.3/Introduction.html#Introduction_build_system | 17:35 |
@sonney2k | wiking, cpuid is not *that* important with new gcc versions | 17:36 |
@sonney2k | they all support -march=native | 17:36 |
@sonney2k | which automatically selects the best options | 17:36 |
@sonney2k | there might be some issues still with not enabling all sse variants | 17:37 |
thoralf | sonney2k: -march doesn't really matter as we're using -O0 anyway | 17:46 |
thoralf | :) | 17:46 |
lambday | sonney2k: yes | 17:48 |
lambday | I added more tests for SGVector and SGMatrix | 17:48 |
lambday | that too works | 17:48 |
lambday | except json that is | 17:48 |
lambday | man! that's a pretty huge test file :( | 17:49 |
thoralf | SGVector has an issue (i think) - try adding "ASSERT(j < len)" just in front of "if (vec[j] > maxv)" SGVector.cpp, line 996 | 17:50 |
thoralf | Two unit tests will fail then. | 17:50 |
thoralf | Failing tests are: mMCLDA.train_and_apply and mQDA.train_and_apply | 17:52 |
lambday | sonney2k: HeikoS: will be back after dinner | 17:54 |
thoralf | Can anyone confirm that (j<len) is a valid assertion in SGVector::arg_max, line 996? | 17:59 |
@HeikoS | thoralf: man this code is so hard to read ;) | 18:10 |
thoralf | HeikoS: Multiple indices rock ;) | 18:11 |
@HeikoS | sonney2k, btw is there any reason why SGMatrix and SGVector are distinguished? Why dont they have a joint type? would make some things easier? | 18:11 |
@HeikoS | thoralf: man I am too lazy to try to understand this | 18:11 |
@HeikoS | :D | 18:11 |
thoralf | HeikoS: Forget the crazy loop - maybe you cann tell if length(vec) == len should be true or not :) | 18:13 |
-!- nube [~rho@49.244.7.173] has quit [Quit: Leaving.] | 18:13 | |
@HeikoS | yes | 18:14 |
@HeikoS | but I think I dont get the question | 18:14 |
thoralf | HeikoS: If I add "ASSERT(j<len)" to the loop, then two tests fail. | 18:14 |
thoralf | HeikoS: I was debugging something and adding assertions - and found this. | 18:15 |
@HeikoS | thoralf: I see | 18:15 |
thoralf | HeikoS: And now I need to know: Am I wrong or shogun. ;) | 18:15 |
@HeikoS | maybe we should just rewrite this horrible method? | 18:15 |
@HeikoS | its totally old anyway | 18:15 |
@HeikoS | what is inc? | 18:16 |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 18:17 | |
shogun-notifier- | shogun: lambday :develop * 8acb5bc / tests/unit/io/Serialization_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/8acb5bc56796bb41c63c884ce6b582a7aeeb3ebb | 18:17 |
shogun-notifier- | shogun: added more unit-tests serialization for ptype and ctype | 18:17 |
shogun-notifier- | shogun: Heiko Strathmann :develop * 24b2ef0 / tests/unit/io/Serialization_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/24b2ef02075a7e1bef2fed30ac6de35a4db13a77 | 18:17 |
shogun-notifier- | shogun: Merge pull request #1163 from lambday/develop | 18:17 |
shogun-notifier- | shogun: | 18:17 |
shogun-notifier- | shogun: added more unit-tests serialization for ptype and ctype (SGVector, SGMatrix) | 18:17 |
thoralf | HeikoS: I don't know - but I think this method even fails when inc=1 | 18:17 |
@HeikoS | well at least it is not documented :D | 18:18 |
@HeikoS | no seriously, no idea, I think it should be remove and rewritten for the new SGVector stuff | 18:18 |
@HeikoS | without inc | 18:18 |
thoralf | What should we do now? Adding an assertion breaks the build. For writing tests for pathological tests its too late now ;) | 18:21 |
thoralf | I'll file a bug report. | 18:21 |
@HeikoS | thoralf: thats best | 18:21 |
thoralf | Feel free to comment https://github.com/shogun-toolbox/shogun/issues/1164 | 18:24 |
-!- HeikoS [~heiko@nat-184-237.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.] | 18:25 | |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun | 18:28 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 18:29 | |
shogun-buildbot | build #928 of cyg1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/928 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 18:34 |
shogun-buildbot | build #929 of cyg1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/929 blamelist: lambday <heavensdevil6909@gmail.com> | 18:46 |
@sonney2k | thoralf, the arg_max function is correct though | 19:34 |
@sonney2k | iglesiasg, any idea why multiclass lda fails? | 19:34 |
@iglesiasg | sonney2k: I didn't see it | 19:35 |
lambday | sonney2k: now that all tests run for complex64_t.. shall I update the news file? | 19:35 |
lambday | by the way, I just saw a python... a real one! | 19:36 |
lambday | our campus is right beside a national sanctuary sort of place.. campus is pretty wild... literally I mean | 19:37 |
@sonney2k | iglesiasg, https://github.com/shogun-toolbox/shogun/issues/1164 | 19:37 |
@sonney2k | iglesiasg, the arg_max is ok though | 19:37 |
@sonney2k | lambday, woa! | 19:37 |
@sonney2k | lambday, don't let her get a bite! | 19:37 |
lambday | few days back a leopard was spotted right beside my hostel :D | 19:37 |
@sonney2k | lambday, sure update! | 19:37 |
lambday | hahaha | 19:37 |
lambday | sonney2k: should I add it in the features? | 19:38 |
lambday | that shogun supports complex64_t now | 19:38 |
@iglesiasg | sonney2k: I will take a look asap | 19:38 |
@sonney2k | lambday, yes! | 19:39 |
@sonney2k | lambday, and put your name in brackets behind it. you deserve some credit! | 19:39 |
lambday | sonney2k: :) | 19:40 |
lambday | thanks :) | 19:40 |
lambday | okay I'm changing | 19:40 |
@sonney2k | southern germany is halfway under water - flooded | 19:44 |
lambday | whoa! | 19:44 |
thoralf | sonney2k: Okay, so "vec" may be shorter than len? | 19:44 |
thoralf | s/shorter/longer/ | 19:46 |
lambday | sonney2k: I don't know what else to write :( | 19:54 |
-!- hushell [~hushell@c-24-21-141-32.hsd1.or.comcast.net] has quit [Ping timeout: 260 seconds] | 20:00 | |
@sonney2k | thoralf, yes | 20:10 |
shogun-notifier- | shogun: lambday :develop * 69447db / src/NEWS: https://github.com/shogun-toolbox/shogun/commit/69447dbdd535f1e3e7a2d1b60d732dfd66da651a | 20:10 |
shogun-notifier- | shogun: NEWS edited with complex64_t support | 20:11 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 5aed350 / src/NEWS: https://github.com/shogun-toolbox/shogun/commit/5aed350a60aa4140d68259398cc6eff1821f33b7 | 20:11 |
shogun-notifier- | shogun: Merge pull request #1165 from lambday/develop | 20:11 |
shogun-notifier- | shogun: | 20:11 |
shogun-notifier- | shogun: NEWS edited with complex64_t support | 20:11 |
@sonney2k | lambday, no idea if you replied already but am I right that serialization works in your tests for everything except json int / char? | 20:11 |
lambday | sonney2k: yes | 20:11 |
lambday | int/char stores it using json int only | 20:12 |
lambday | and they all fail | 20:12 |
@sonney2k | lambday, weird | 20:12 |
lambday | storing is fine I guess the problem is with reading | 20:12 |
lambday | wait let me show you a gist | 20:12 |
thoralf | sonney2k: Feel free to close the ticket. | 20:12 |
@sonney2k | thoralf, no iglesiasg is investigating what's wrong | 20:12 |
@iglesiasg | not yet actually, I must finish something first | 20:13 |
@sonney2k | iglesiasg, then soon :) | 20:13 |
@iglesiasg | sonney2k: BTW, I didn't get any mail about this mailing list thing | 20:14 |
@iglesiasg | should I have received something? | 20:14 |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has joined #shogun | 20:17 | |
lambday | sonney2k: https://gist.github.com/lambday/5728708 | 20:17 |
lambday | sonney2k: check the comment... I added the debug msgs | 20:17 |
@sonney2k | argh I am too tired now | 20:21 |
@sonney2k | cu! | 20:21 |
lambday | sonney2k: alright | 20:21 |
lambday | ciao :) | 20:21 |
thoralf | Bye sonney2k | 20:25 |
shogun-buildbot | build #930 of cyg1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/930 blamelist: Soeren Sonnenburg <sonne@debian.org> | 20:29 |
shogun-buildbot | build #931 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/931 blamelist: lambday <heavensdevil6909@gmail.com> | 20:30 |
-!- vgorbati [~vgorbati@212.2.159.34] has quit [Quit: vgorbati] | 20:37 | |
lambday | good night guys | 20:41 |
-!- lambday [67157d36@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.54] has quit [Quit: lambday] | 20:41 | |
@iglesiasg | ok time to go home, see you later people | 20:47 |
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Quit: Page closed] | 20:50 | |
-!- thoralf [~thoralf@enki.zib.de] has quit [Quit: Konversation terminated!] | 20:54 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 246 seconds] | 20:57 | |
-!- lisitsyn [~blackburn@109-226-114-235.clients.tlt.100megabit.ru] has joined #shogun | 21:15 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 21:15 | |
@lisitsyn | ha pure friday | 21:15 |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has quit [Ping timeout: 276 seconds] | 22:00 | |
-!- gsomix [~gsomix@80.234.54.193] has joined #shogun | 22:32 | |
gsomix | good evening/night | 22:32 |
gsomix | hm, not many people | 22:33 |
gsomix | so, night party^W coding session | 22:33 |
-!- zxtx [~zv@rrcs-74-62-200-195.west.biz.rr.com] has joined #shogun | 22:58 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 23:11 | |
--- Log closed Sat Jun 08 00:00:36 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!