--- Log opened Tue May 28 00:00:20 2013 | ||
-!- nube [~rho@49.244.111.24] has quit [Ping timeout: 246 seconds] | 01:38 | |
-!- nube [~rho@49.244.27.0] has joined #shogun | 01:53 | |
-!- foulwall [~foulwall@2001:da8:215:c252:addf:5916:dba8:873b] has quit [Remote host closed the connection] | 01:55 | |
-!- iglesiasg [d58f32c1@gateway/web/freenode/ip.213.143.50.193] has quit [Quit: Page closed] | 02:03 | |
-!- foulwall_ [~foulwall@2001:da8:215:6901:8c32:a3d2:a949:f10d] has joined #shogun | 02:18 | |
-!- foulwall_ [~foulwall@2001:da8:215:6901:8c32:a3d2:a949:f10d] has quit [Ping timeout: 264 seconds] | 02:28 | |
-!- foulwall [~foulwall@2001:da8:215:6902:8d90:876f:7e65:da17] has joined #shogun | 02:29 | |
-!- nube [~rho@49.244.27.0] has quit [Quit: Leaving.] | 03:38 | |
-!- sijin [~smuxi@144.214.222.109] has quit [] | 04:22 | |
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has joined #shogun | 04:49 | |
-!- foulwall [~foulwall@2001:da8:215:6902:8d90:876f:7e65:da17] has quit [Remote host closed the connection] | 05:06 | |
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has quit [Ping timeout: 250 seconds] | 05:31 | |
-!- nube [~rho@116.90.239.3] has joined #shogun | 05:31 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 05:56 | |
shogun-notifier- | shogun: hushell :develop * 9fb0d3c / src/shogun/multiclass/MulticlassOneVsRestStrategy.h: https://github.com/shogun-toolbox/shogun/commit/9fb0d3c1e507c515496f23eff1fd936048c7b592 | 05:56 |
---|---|---|
shogun-notifier- | shogun: fix documentation | 05:56 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 3fae2be / src/shogun/multiclass/MulticlassOneVsRestStrategy.h: https://github.com/shogun-toolbox/shogun/commit/3fae2be16fee3ae4afa06af222d83d86b3e27de0 | 05:56 |
shogun-notifier- | shogun: Merge pull request #1132 from hushell/develop | 05:56 |
shogun-notifier- | shogun: | 05:56 |
shogun-notifier- | shogun: fix documentation | 05:56 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * ae85697 / doc/Doxyfile_cn.in,doc/Doxyfile_en.in: https://github.com/shogun-toolbox/shogun/commit/ae8569738945e05d4fa0cb78a16cf5c6f7b80d08 | 06:06 |
shogun-notifier- | shogun: ignore shogun/ui and external libs in doxygen doc generation | 06:06 |
shogun-buildbot | build #879 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/879 | 06:22 |
shogun-buildbot | build #880 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/880 blamelist: Soeren Sonnenburg <sonne@debian.org> | 06:23 |
-!- gsomix [~gsomix@185.6.239.144] has quit [Ping timeout: 256 seconds] | 07:07 | |
shogun-buildbot | build #1173 of deb3 - modular_interfaces is complete: Failure [failed test ruby_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1173 blamelist: Soeren Sonnenburg <sonne@debian.org> | 07:24 |
-!- erlenda [~erlenda@fw-oslo.intelcom.no] has joined #shogun | 07:44 | |
-!- gsomix [~Miranda@r206-10.smr.ru] has joined #shogun | 07:52 | |
gsomix | good morning | 07:52 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun | 08:02 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 08:03 | |
gsomix | lisitsyn: hey. with AC adapter now? | 08:04 |
@lisitsyn | gsomix: ja | 08:14 |
gsomix | lisitsyn: horosho | 08:15 |
@lisitsyn | shshshsh I need iglesiasg | 08:25 |
sonne|work | gsomix: morning! | 08:46 |
sonne|work | gsomix: already doing speed tests? | 08:46 |
gsomix | sonne|work: nope, sorry. sitting in lab now. | 08:48 |
-!- qwerty [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has joined #shogun | 08:49 | |
-!- qwerty is now known as Guest51873 | 08:50 | |
gsomix | sonne|work: I'll do tests at evening. all ok? | 08:51 |
sonne|work | gsomix: ok. looking forward! | 08:52 |
-!- Guest51873 [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has quit [Client Quit] | 08:53 | |
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has joined #shogun | 08:54 | |
sonne|work | good morning vgorbati | 08:54 |
vgorbati | sonne|work: hello. While you are here: what should I do with test converter...: it produces the same results on build-bots and lisitsyn's machine (linux) while giving another results on my machine (OS X) | 08:55 |
@lisitsyn | vgorbati: I am sorry we didn't find a place for your proposal this year | 08:55 |
@lisitsyn | hope you are not very disappointed! | 08:55 |
sonne|work | lisitsyn: well sure he is | 08:56 |
sonne|work | at least I would be :/ | 08:56 |
vgorbati | lisitsyn: well, I was, but I am a big guy already) | 08:56 |
@lisitsyn | vgorbati: your proposal was definitely strong but we had a tough competition of all proposals, slots and persons | 08:57 |
sonne|work | vgorbati: all I can say is that this is not only about 'how good you are' but also about which task we consider more important currently | 08:57 |
sonne|work | vgorbati: and we would very much still have you around and work with you! | 08:57 |
vgorbati | sonne|work: so, what about the test: I mean, I can just re-submit the data for it using the results, received on build-bots and lisitsyn's machine, and then the test should pass, but it seems like it is not a very good solution.. | 08:57 |
@lisitsyn | vgorbati: yeah that's not really a good solution | 08:58 |
@lisitsyn | sonne|work: vgorbati reports that our random is not very cool | 08:58 |
sonne|work | lisitsyn: what? we have a test for that so it certainly is reproducible | 08:59 |
sonne|work | vgorbati, lisitsyn - anything multithreaded in there? | 08:59 |
@lisitsyn | no | 08:59 |
sonne|work | these functions are not thread safe | 08:59 |
sonne|work | vgorbati: what exactly fails? we really need to debug this then | 09:02 |
vgorbati | sonne|work: gathering links to the proofs of my words) | 09:02 |
vgorbati | sonne|work: okay, here's the travis build for my PR: https://travis-ci.org/shogun-toolbox/shogun/jobs/7428321. Scroll it down to the DIFF for test converter_tdistributedstochasticneighborembedding_modular.py. Now, here is the results on lisitsyn's machine: http://pastebin.com/MxDSrYJS. You may notice, that the GOT results on build-bot are the same as the results on the lisitsyn's machine | 09:05 |
vgorbati | sonne|work: and the EXPECTED is the results, obtained on my machine | 09:06 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 09:06 | |
sonne|work | vgorbati: results are completely different so please add some debug code already at the early stages. then you can give sergey the output you get and he can compare - so we will find out | 09:09 |
@lisitsyn | vgorbati: we should check whether it is really a random issue | 09:10 |
@lisitsyn | may be we should just generate a few gaussians | 09:10 |
@lisitsyn | and check | 09:10 |
sonne|work | lisitsyn, vgorbati I am pretty sure it is not | 09:10 |
sonne|work | vgorbati: if you run unit tests? do the random tests fail or work? | 09:10 |
vgorbati | sonne|work: and finally, here is the build results on shogun-buildbot http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1158/steps/test%20python_modular/logs/stdio, and if you scroll down to diff, the GOT is the same as on travis and on lisitsyn's machine | 09:11 |
vgorbati | sonne|work: as far as I remember, the unit tests were fine | 09:11 |
sonne|work | vgorbati: ok so then it is not random* but sth else | 09:12 |
sonne|work | maybe you can print out the output at each iteration or so | 09:12 |
sonne|work | somethign very verbose | 09:12 |
vgorbati | sonne|work: ok, will try | 09:12 |
@lisitsyn | vgorbati: hmm true that should not be random issue then | 09:12 |
vgorbati | lisitsyn: then I am sorry to blame the random, but it was the obvious thing to think of as the reason | 09:13 |
sonne|work | lisitsyn: what other numerics is involved? | 09:14 |
@lisitsyn | yeah it looked as a random thing for me too | 09:14 |
sonne|work | blas/lapack eigen? | 09:14 |
@lisitsyn | sonne|work: no just C-style code actually | 09:14 |
sonne|work | lisitsyn: hmmhh no idea then did you valgrind too? | 09:14 |
@lisitsyn | sonne|work: there is a random shuffle involved actually | 09:15 |
@lisitsyn | but I made it shogun dependent too as I hope | 09:15 |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has joined #shogun | 09:19 | |
sonne|work | lisitsyn: well just compare the output after the shuffle and you will know :D | 09:19 |
gsomix | wow, there is big headlight from audi a6 in lab | 09:35 |
sonne|work | gsomix: your new toy or what? | 09:42 |
gsomix | sonne|work: I think this is toy of my supervisor | 09:44 |
sonne|work | hehe | 09:44 |
@lisitsyn | I'd rather take the other part of that car | 09:44 |
@lisitsyn | I mean the one that is left | 09:46 |
@lisitsyn | vgorbati: I think that's a boring issue | 09:51 |
@lisitsyn | vgorbati: if you are up to have some fun with code we can think of something better | 09:51 |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has quit [Ping timeout: 276 seconds] | 09:52 | |
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has quit [Ping timeout: 250 seconds] | 09:59 | |
-!- foulwall [~foulwall@li379-21.members.linode.com] has joined #shogun | 09:59 | |
-!- foulwall [~foulwall@li379-21.members.linode.com] has left #shogun [] | 09:59 | |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has joined #shogun | 10:00 | |
-!- iglesiasg [d58f3217@gateway/web/freenode/ip.213.143.50.23] has joined #shogun | 10:24 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 10:25 | |
@wiking | btw: gsoc 2014 project idea: interface shogun for spss ;) | 10:42 |
@lisitsyn | wiking: wat? | 10:42 |
@wiking | it's pretty straightforward task... a codegen could create all the modules | 10:42 |
@lisitsyn | lolz | 10:42 |
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has joined #shogun | 10:42 | |
@wiking | lisitsyn: http://www-01.ibm.com/software/analytics/spss/products/modeler/ | 10:42 |
@lisitsyn | ohohoh I am submitting a JMLR paper right now | 10:42 |
@lisitsyn | uploaddding | 10:42 |
@lisitsyn | wiking: I heard about spss | 10:42 |
@lisitsyn | in your twitter :D | 10:43 |
@wiking | lisitsyn: \o/ | 10:43 |
@wiking | lisitsyn: fuckthatshit | 10:43 |
@lisitsyn | very expensive excel right? | 10:43 |
sonne|work | lisitsyn: the never ending story came to an end? | 10:43 |
sonne|work | lisitsyn: haha | 10:43 |
vgorbati | lisitsyn: sorry, so what about " have some fun with code"? | 10:43 |
@lisitsyn | sonne|work: Your manuscript has been successfully submitted. | 10:43 |
sonne|work | vgorbati: what are you interested in doing? | 10:46 |
@wiking | vgorbati: sorry had 0 time to test what you've sent | 10:46 |
sonne|work | wiking: ahh the zerotime guy is back! | 10:47 |
vgorbati | wiking: no problems, it looks like it is not related to random | 10:47 |
vgorbati | sonne|work: well, doing some development tasks would be great, but I do not mind doing some clean up/refactoring | 10:49 |
@lisitsyn | vgorbati: we should couple it with your research | 10:49 |
@lisitsyn | what is you are doing for your thesis? | 10:49 |
vgorbati | lisitsyn: forecasting stuff using neural networks/group method of data handling | 10:50 |
vgorbati | lisitsyn: but doing something opposite would also be great - new knowledge is always new knowledge:) | 10:51 |
@lisitsyn | vgorbati: well may be implementing MGUA stuff is relevant | 10:51 |
@lisitsyn | :) | 10:51 |
@lisitsyn | you may find that useful for your research as shogun provides a platform for that | 10:52 |
vgorbati | lisitsyn: you are speaking about model selection, right? | 11:01 |
@lisitsyn | vgorbati: well we have model selection | 11:02 |
@lisitsyn | but we have no GMDH-like thing | 11:02 |
vgorbati | lisitsyn: I mean "shogun provides a platform for that" - it is about model selection? Because model selection is part of GMDH | 11:02 |
@lisitsyn | ahh | 11:02 |
@lisitsyn | not only that | 11:02 |
@lisitsyn | I mean shogun can be very convenient to try new ideas | 11:03 |
sonne|work | lisitsyn: what is MGUA GMDH? | 11:03 |
@lisitsyn | sonne|work: http://en.wikipedia.org/wiki/Group_method_of_data_handling | 11:04 |
@lisitsyn | well it is too old actually I think :D | 11:04 |
@lisitsyn | but if somebody does research on that | 11:05 |
sonne|work | lisitsyn: kind of multiple kernel learning / group lasso right? | 11:05 |
@lisitsyn | sonne|work: no, no regularization | 11:05 |
@lisitsyn | sonne|work: just make it more complex while you get better results | 11:06 |
@lisitsyn | sonne|work: quite opposite to svm thing actually | 11:06 |
sonne|work | sure not regularization / I meant the look at different parts of the features parts | 11:07 |
sonne|work | part | 11:07 |
@lisitsyn | sonne|work: depends what point of MKL/lasso you speak about then | 11:08 |
@lisitsyn | I consider regularization most important here :) | 11:08 |
@lisitsyn | sonne|work: IIRC it has nothing to do with real model - just a principle | 11:09 |
@lisitsyn | in wiki they just state some kolmogorov-gabor (wtf is it) :D | 11:09 |
@lisitsyn | looks like very densely connected NN | 11:09 |
vgorbati | lisitsyn: yes, the set of elementary functions can be anything - like linear functions, garmonic etc. You then just make a model more complex by applying the functions over and over: like at the first level you will have z1=a11*x+a12*y, z2=a21*x+a22*y, at the second level - f=a11*z1+a22*z2... | 11:13 |
-!- gsomix [~Miranda@r206-10.smr.ru] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 11:14 | |
@lisitsyn | sonne|work: vgorbati knows more as he is studying in the same university the author of this method was teaching | 11:16 |
vgorbati | lisitsyn: I think developing GDMH for shogun is qiute questionable... I may just work on DR module as it was described in my project, just not in such a dense schedule | 11:16 |
sonne|work | lisitsyn: hmmh in my eyes this is just some combineddotfeatures where the coefficients are learned and yes maybe growing number | 11:16 |
sonne|work | vgorbati: we have no neural network expert - so it could become your topic but as you like | 11:18 |
@lisitsyn | sonne|work: you just don't learn them all | 11:18 |
sonne|work | lisitsyn: well you learn the a's no? | 11:18 |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has quit [Ping timeout: 256 seconds] | 11:18 | |
sonne|work | and then add more and more features | 11:18 |
@lisitsyn | first you learn one layer then +1 more | 11:18 |
@lisitsyn | sonne|work: in case of that K-G polynomial stuff - yes | 11:19 |
sonne|work | yeah ok | 11:19 |
vgorbati | sonne|work: the very basic algorithm: you select elementary functions - say, linear combination of 2 variables; | 11:20 |
vgorbati | sonne|work: you start at the 1 level - try all possible combinations of 2 variables - parameters of models are found using least squares | 11:20 |
sonne|work | ok | 11:21 |
vgorbati | sonne|work: you select the best models via some external criteria - say, the least squares of models at the testing set, there are others as well - the main point is that this "external" criterias should be calculated on the other part of the data | 11:21 |
vgorbati | sonne|work: on the 2 level you take any possible combination of the 1 level models | 11:22 |
sonne|work | makes sense | 11:22 |
vgorbati | sonne|work: and so on | 11:22 |
vgorbati | sonne|work: while external criterias are decreasing | 11:22 |
vgorbati | sonne|work: or until you reach a needed value for external criterias | 11:23 |
vgorbati | sonne|work: with this approach you will drop out the unneeded variables, since only informative models are selected for the next level | 11:23 |
sonne|work | lisitsyn: I think we need some cascades of classifiers btw... | 11:23 |
@lisitsyn | sonne|work: cascades? like weighted set of classifiers? | 11:24 |
sonne|work | lisitsyn: no more in the sense that only if f(x)>threshold at all levels the example is considered yes | 11:25 |
@lisitsyn | ahh so real cascades | 11:26 |
@lisitsyn | got it | 11:26 |
@lisitsyn | yes makes sense | 11:26 |
-!- lambday [67157d37@gateway/web/freenode/ip.103.21.125.55] has joined #shogun | 11:28 | |
lambday | HeikoS: hi | 11:32 |
lambday | lisitsyn: hi :) | 11:33 |
@lisitsyn | lambday: hey | 11:34 |
lambday | sorry couldn't be there in the chat last night at the time of announcement :( urgent lab system adm work :( | 11:35 |
lambday | some nfs lock problem.. really shitty :( | 11:35 |
sonne|work | lambday: did you suffer a raid crash? | 11:35 |
sonne|work | ahh ok so you did reboot everything :D | 11:35 |
lambday | sonne|work: pathetic :( | 11:35 |
lambday | seems like its a bug in the linux kernel | 11:36 |
lambday | something creates deadlock or something... I don't really get it | 11:36 |
sonne|work | I've been a sysadm for a while so I know how you feel :D | 11:37 |
lambday | sonne|work: :( | 11:37 |
lambday | all I do is google it and check stackoverflow solns :D | 11:37 |
sonne|work | I did this 2002-2004 and google / stackoverflow weren't that helpful back then | 11:38 |
sonne|work | but people on mailinglists ... | 11:38 |
lambday | sonne|work: man!! how did you handle it that time then?? :-o | 11:39 |
@lisitsyn | lambday: that's why he never smiles now | 11:39 |
lambday | lisitsyn: lol.. he just did | 11:39 |
@lisitsyn | no no | 11:40 |
lambday | I bet he did :D | 11:40 |
sonne|work | lambday: back then one had to talk to people | 11:40 |
sonne|work | and to add printk's to the linux kernel | 11:40 |
lambday | can't even imagine | 11:41 |
-!- van51 [~van51@195.251.235.71] has joined #shogun | 11:46 | |
van51 | hello | 11:46 |
van51 | sonne|work: sorry for the long delay! whenever you're available to talk, let me know | 11:47 |
sonne|work | van51: good morning and congrats for being accepted | 11:47 |
sonne|work | seems like you left us with no choice ;) | 11:47 |
van51 | sonne|work: thank you for accepting me :) | 11:48 |
sonne|work | van51: I wanted to discuss how to proceed | 11:48 |
sonne|work | van51: IMHO you could easily do the random fourier features / kitchen sink stuff but also the hashing stuff. It. | 11:49 |
sonne|work | van51: so I would like to connect you with Olivier & Quoc for that part | 11:50 |
van51 | sonne|work: ok | 11:50 |
van51 | Quoc I've seen him on the ideas page | 11:50 |
van51 | is Olivier also a mentor? | 11:50 |
sonne|work | van51: yes - have a look at the ideas page again | 11:51 |
sonne|work | it is about the hashing of features. | 11:51 |
van51 | sonne|work: ok yeah, I just saw it | 11:52 |
van51 | and how do I contact them?via email? | 11:52 |
sonne|work | yes please do that and CC me | 11:52 |
sonne|work | I think the hashing idea is related in a sense that it is all about large scale | 11:53 |
sonne|work | and fits well in your expertise | 11:53 |
vgorbati | lisitsyn, sonne|work: so you guys really confused me what to do. What if I try to work on non-linear multilayered perceptron? | 11:53 |
sonne|work | van51: in short - all you do with the hashing stuff is compute h(i) (i being an index into a feature) | 11:54 |
sonne|work | van51: so say if you have 1000 features and a hash dictionary of size 8 bit you compress these 1000 features down to 256 | 11:54 |
sonne|work | so think of some really massively dimensional space | 11:54 |
sonne|work | like something the polynomial kernel could be made of | 11:55 |
sonne|work | this hashing trick helps one to learn with kernel approximations but in the primal | 11:55 |
sonne|work | so it is kind of a solution to the same problem you do with kitchen sinks etc | 11:56 |
sonne|work | approximate some kernel space | 11:56 |
van51 | sonne|work: ok, I got the idea | 11:56 |
sonne|work | vgorbati: well - you can do everythign :D But we want you to do what you would have most fun with. Because then it is fun for us all. | 11:56 |
sonne|work | vgorbati: and you are most productive too | 11:56 |
van51 | sonne|work: but I think it'd be better if I read a bit more before contacting them? | 11:57 |
sonne|work | van51: no please contact them now already | 11:57 |
sonne|work | it is kind of stupid that I am assigned as mentor for you but hey sure I will help were I can | 11:58 |
van51 | okay! | 11:58 |
van51 | will contact them right away | 11:59 |
-!- nube [~rho@116.90.239.3] has quit [Quit: Leaving.] | 12:08 | |
@lisitsyn | vgorbati: if you are eager to publish we should think of something, implement and publish it | 12:23 |
@lisitsyn | I like that idea as it promotes shogun in papers | 12:23 |
* lambday is listening to shogun mix | 12:25 | |
@lisitsyn | lambday: shogun mix? :D | 12:29 |
@lisitsyn | this is dj sonne and a shogun mix? | 12:29 |
@lisitsyn | the state of shogun? | 12:30 |
@lisitsyn | by dj wiking | 12:30 |
lambday | lisitsyn: lol | 12:30 |
lambday | http://www.youtube.com/watch?v=ItvzR6y_kjA | 12:30 |
lambday | (y) for "state of shogun" ! :D | 12:31 |
@lisitsyn | lambday: armin van shoguneer | 12:31 |
lambday | lisitsyn: hahahahaha | 12:31 |
lambday | he came to bombay few weeks back btw | 12:32 |
@lisitsyn | shogun van dyk | 12:32 |
lambday | buuren I meant :D | 12:32 |
@lisitsyn | dj shoguniesto | 12:32 |
-!- foulwall [~foulwall@li379-21.members.linode.com] has joined #shogun | 12:32 | |
@lisitsyn | what else? | 12:33 |
@lisitsyn | shogun and beyond | 12:33 |
lambday | lisitsyn: ah, LOVE this one :D | 12:33 |
@lisitsyn | I like earth wind and shogun band | 12:33 |
* lambday bows | 12:34 | |
@lisitsyn | have you listened to frank shogunatra? | 12:34 |
vgorbati | lisitsyn: to publish something this something needs to be not ordinary - do you have some ideas of such a thing?) | 12:34 |
@lisitsyn | vgorbati: yes that's a long way for a non-shameful paper but we are not in hurry right? | 12:35 |
lambday | lisitsyn: nope :D | 12:35 |
@lisitsyn | vgorbati: someone said think hard is enough | 12:35 |
@lisitsyn | lets think hard enough! | 12:35 |
@lisitsyn | lambday: the black sun shogun may be? | 12:37 |
@lisitsyn | and their famous 'arrakis' track that sounds exactly like my notebook training an SVM | 12:38 |
lambday | lisitsyn: lol :D | 12:40 |
lambday | I just googled it and it sounds awfully similar to the "powerpuff girls" soundtrack cartoon on cartoon network.. | 12:41 |
@lisitsyn | lambday: black sun empire sounds like powerpuff girls? what? :) | 12:44 |
lambday | the first part ! :D | 12:45 |
@lisitsyn | ahh but not the ear-bleeding-bass part right? ;) | 12:45 |
lambday | lisitsyn: thankfully :D | 12:47 |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has joined #shogun | 13:01 | |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has quit [Quit: Leaving] | 13:07 | |
@HeikoS | iglesiasg, hushell, votjakovr, lambday, foulwall, pickle27 congratulations to all of you! | 13:15 |
@iglesiasg | thank you HeikoS ! :) | 13:16 |
foulwall | thanks HeikoS:D | 13:16 |
sonne|work | lisitsyn: oh the entertainer is back :D | 13:19 |
@HeikoS | entertainer? | 13:19 |
sonne|work | HeikoS: you missed van51 :) | 13:19 |
@HeikoS | sonne|work: let me check :) | 13:20 |
van51 | HeikoS: I am offended! | 13:20 |
sonne|work | iglesiasg: btw congratulations for getting git push rights to shogun! | 13:20 |
@iglesiasg | sonne|work: wow thanks! | 13:21 |
sonne|work | van51: you should be - must be HeikoS GERMAN attitude ;) | 13:21 |
@HeikoS | GERMAN :) | 13:22 |
sonne|work | foulwall: hey and congrats to | 13:22 |
@HeikoS | I didnt do anything here ;) | 13:22 |
sonne|work | HeikoS: you know van51 is from Greece | 13:22 |
foulwall | sonne|work: thank you! | 13:22 |
@HeikoS | sonne|work: me being rude is actually not related to my nationality but to the fact that I am just very bad in text-based communication! ;) | 13:23 |
sonne|work | iglesiasg: so get your patch in yourself and make sure it works :D | 13:24 |
@iglesiasg | sonne|work: will do! | 13:24 |
sonne|work | foulwall: alright so welcome to the team and lets get to work... | 13:30 |
foulwall | sonne|work: ok~~~, shall we talk some hours later? | 13:32 |
@lisitsyn | HeikoS wo bist du? | 13:32 |
@lisitsyn | HeikoS: :D | 13:32 |
@lisitsyn | HeikoS: am I progressing? ;) | 13:32 |
@HeikoS | lisitsyn: hier hier! | 13:32 |
@lisitsyn | da da | 13:32 |
@lisitsyn | HeikoS: wie machen sie da? | 13:33 |
@lisitsyn | HeikoS: ich mache arbeiten hier! | 13:33 |
@HeikoS | sehr gut! | 13:34 |
sonne|work | HeikoS: don't give him false hopes :D | 13:34 |
@lisitsyn | sonne|work: bist du glueklich? | 13:35 |
@lisitsyn | :D | 13:35 |
@HeikoS | So wird das niemals etwas Sergey. Mehr üben! | 13:35 |
@lisitsyn | waaat | 13:35 |
sonne|work | van51: my suggestion would be that you take https://github.com/shogun-toolbox/shogun/pull/901 as a baseline and implement the RF features | 13:36 |
@lisitsyn | HeikoS: I don't give a word :D | 13:36 |
@lisitsyn | ehr | 13:36 |
@lisitsyn | google translate is not helpful | 13:37 |
@lisitsyn | HeikoS: is that I should practice more? | 13:37 |
@lisitsyn | :) | 13:37 |
@HeikoS | lisitsyn: yes | 13:37 |
@lisitsyn | HeikoS: how and why :) | 13:37 |
van51 | sonne|work: Ok, I'll start looking that | 13:37 |
-!- erlenda [~erlenda@fw-oslo.intelcom.no] has quit [Ping timeout: 256 seconds] | 13:37 | |
@lisitsyn | HeikoS: what is the first part? | 13:38 |
sonne|work | it might be a bit messy and certainly misses a test but the goal should be to create some class CRandomFourierDotFeatures that gets a DenseFeatures<float64> as input and then implements all the add / dense_dot operations with tests | 13:38 |
@lisitsyn | gtranslate said I will never do something - makes no sense | 13:38 |
van51 | sonne|work: got it! thanks for the suggestion | 13:40 |
van51 | I'll start looking at it in about an hour after a finish with a small assignment | 13:40 |
van51 | hopefully the other will get back to me soon | 13:41 |
@lisitsyn | sonne|work: did you know these guys like van51 presented us an alphabet? | 13:41 |
@lisitsyn | I should be thankful ;) | 13:41 |
van51 | lisitsyn: you are welcome :P | 13:42 |
@lisitsyn | van51: ah!! the most important question | 13:42 |
@lisitsyn | van51: how do you name variables and functions? | 13:42 |
sonne|work | maybe I should too but hey we got them trapped in the mighty €€€€€€€ | 13:42 |
van51 | lisitsyn: you know the typical.. a,b,c etc | 13:43 |
@lisitsyn | sonne|work: no you have latin alphabet ;) | 13:43 |
@lisitsyn | van51: hmm we use greek letters | 13:43 |
@lisitsyn | van51: what is the 'sin' function? | 13:43 |
@lisitsyn | just 'sin'? | 13:43 |
@lisitsyn | I always think it is vice versa :) | 13:43 |
sonne|work | so I guess van51 we are even. | 13:43 |
@lisitsyn | sonne|work has got a serious euro-ache | 13:44 |
van51 | lisitsyn: what do you mean? | 13:44 |
van51 | about the sin? i lost you somewhere | 13:44 |
@lisitsyn | van51: I have seen in some book they wrote 'sin' with greek letters | 13:44 |
@lisitsyn | and variables were latin variables | 13:44 |
@lisitsyn | ehhr | 13:45 |
@lisitsyn | latin letters I mean | 13:45 |
van51 | lisitsyn: so they had written "συν" and "ημ"? | 13:46 |
@lisitsyn | yes probably! | 13:46 |
van51 | hehe | 13:46 |
@lisitsyn | van51: can that be that? | 13:46 |
@lisitsyn | it was a book on some typesetting or so | 13:46 |
@lisitsyn | by greeks | 13:46 |
@lisitsyn | van51: what is \nu \mu? | 13:46 |
-!- gsomix [~Miranda@r206-10.smr.ru] has joined #shogun | 13:47 | |
van51 | well it could be, but it's not really used in cs | 13:47 |
van51 | we use them as in english | 13:47 |
van51 | cos,sin | 13:47 |
@lisitsyn | van51: I see | 13:49 |
van51 | but if it was in typesetting then yea they were showing the language model,right? | 13:49 |
@lisitsyn | van51: I remember they said like 'we greeks use that notation' | 13:50 |
@lisitsyn | van51: a while ago we used cyrillic stuff :D син кос | 13:51 |
@lisitsyn | way too funny | 13:51 |
van51 | hehe what for? :P | 13:51 |
@lisitsyn | van51: син sin | 13:52 |
@lisitsyn | good it was quite short and not common | 13:52 |
@lisitsyn | before that markov wrote things like 'найдём minimum функции' | 13:53 |
lambday | HeikoS: there? | 13:53 |
@HeikoS | lambday: just getting lunch, will be back in 20 min | 13:53 |
@lisitsyn | like minimum is a very foreign word :) | 13:53 |
lambday | HeikoS: actually I too am leaving for lab.. you'll be here at night? it might take 1-2 hrs :( | 13:54 |
van51 | lisitsyn: i may write something in greek(or greeklish) and leave it here as my legacy | 13:56 |
@lisitsyn | van51: where? :) | 13:57 |
@lisitsyn | in doc? | 13:57 |
@lisitsyn | :) | 13:57 |
van51 | altough some germans(like sonne|work) will say no | 13:57 |
van51 | doc,comments | 13:57 |
van51 | code :p | 13:57 |
@lisitsyn | van51: ahh sonne|work is just a gsoc student | 13:57 |
@lisitsyn | he pretends to be a boss here :( | 13:58 |
van51 | he's bullying you? | 13:58 |
@lisitsyn | yeah I was original author of shogun and he changed all copyrights | 13:58 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun | 13:59 | |
sonne|work | van51: dont' listen to him | 13:59 |
van51 | lisitsyn: well if he managed to top you like that | 13:59 |
sonne|work | votjakovr: congrats too! | 13:59 |
van51 | he kind of deserves it | 13:59 |
sonne|work | true true | 13:59 |
-!- sonne|work is now known as sonne|shogun-kin | 13:59 | |
-!- sonne|shogun-kin is now known as shogun-king | 13:59 | |
shogun-king | van51: only issue is that I am at work and lisitsyn as irc superpowers. | 14:00 |
@lisitsyn | shogun-king: this is not going to happen I'll sue you | 14:00 |
van51 | someone got a bit carried away.. | 14:00 |
votjakovr | shogun-king: Oh, thanks :) | 14:00 |
shogun-king | lisitsyn: try harder kid! | 14:00 |
@lisitsyn | van51: he always acts like that after our divorce in 1997 | 14:01 |
shogun-king | pfffffffffffffffffffffffh | 14:02 |
@lisitsyn | we've been together since our first date at Seoul olympics | 14:02 |
@lisitsyn | in 1988 | 14:02 |
@lisitsyn | it was a cloudy day and we were watching some horse sports | 14:03 |
-!- foulwall [~foulwall@li379-21.members.linode.com] has quit [Remote host closed the connection] | 14:05 | |
@lisitsyn | van51: I don't know may be he is just jealous | 14:05 |
van51 | lisitsyn: some people have a hard time moving on | 14:06 |
@lisitsyn | van51: yes yes true | 14:07 |
@lisitsyn | van51: he took over my house already I don't know what else he needs | 14:08 |
@lisitsyn | I am living there in broken bathtub | 14:08 |
@lisitsyn | that's cruel but that's life | 14:08 |
-!- F3V3R [~CRM114@77.49.62.176.dsl.dyn.forthnet.gr] has joined #shogun | 14:12 | |
@HeikoS | votjakovr: congrats also from me :) | 14:13 |
votjakovr | HeikoS: thank you :) | 14:15 |
-!- shogun-king is now known as sonne|divorced | 14:17 | |
sonne|divorced | lisitsyn: you cruel bastard | 14:18 |
sonne|divorced | you took the kids and everything | 14:18 |
sonne|divorced | and now you blame me | 14:18 |
@lisitsyn | sonne|divorced: that's a lie! you always bully me | 14:19 |
sonne|divorced | lisitsyn: Now that you say it - I think I even enjoy it :P | 14:19 |
@lisitsyn | sonne|divorced: you are unforgivable for taking away my cat | 14:21 |
@lisitsyn | I put a photo to my github profile that's the only thing that is left | 14:21 |
@lisitsyn | I cry nights | 14:21 |
gsomix | lisitsyn: cry? | 14:21 |
gsomix | lisitsyn: I did not notice it before | 14:22 |
sonne|divorced | lisitsyn: you didn't attend your cat for way to long. no wonder I got it. the cat's new hairdresser did find uncountable numbers of empty vodka bottles hidden in its hair | 14:24 |
@lisitsyn | sonne|divorced: I always had troubles with alcohol you knew that | 14:25 |
sonne|divorced | I thought I can get used to it but hey violence is NOT an option | 14:26 |
@lisitsyn | sonne|divorced: I've been beating you up twice a week MAX - what are you unhappy with.. | 14:27 |
sonne|divorced | lisitsyn: yes it is true I still miss you and me TEETH | 14:29 |
@lisitsyn | what teeth :D | 14:30 |
sonne|divorced | mine | 14:30 |
@lisitsyn | ahh | 14:31 |
@lisitsyn | well we put some wooden stuff there | 14:31 |
@lisitsyn | do you find it difficult to chew? | 14:31 |
sonne|divorced | lisitsyn: yes and ohh in this case please take me back | 14:31 |
@lisitsyn | haha | 14:31 |
-!- sonne|divorced is now known as sonne|work | 14:31 | |
@lisitsyn | good show | 14:32 |
@lisitsyn | :D | 14:32 |
sonne|work | lisitsyn: that happens to my brain when I have to code java ;) | 14:32 |
@lisitsyn | sonne|work: we all do java :( | 14:33 |
sonne|work | o_O | 14:33 |
@lisitsyn | sonne|work: err I mean we all need food :D | 14:34 |
-!- erlenda [~erlenda@fw-oslo.intelcom.no] has joined #shogun | 14:34 | |
sonne|work | erlenda: oslo? | 14:36 |
-!- gsomix [~Miranda@r206-10.smr.ru] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 14:57 | |
-!- zeller [~zeller@embln.embl.de] has joined #shogun | 14:58 | |
zeller | hi there | 14:58 |
@lisitsyn | hey georg | 14:58 |
@lisitsyn | zeller: how are you? | 14:58 |
zeller | hi sergey, i'm fine thanks | 14:58 |
zeller | how are things going? | 14:59 |
@lisitsyn | zeller: students do celebrate here ;) | 14:59 |
zeller | hehe well deserved! :-) | 14:59 |
@lisitsyn | zeller: and we with iglesiasg finally submitted that lifelong paper | 14:59 |
zeller | for tapkee? | 15:00 |
@lisitsyn | yes | 15:00 |
zeller | cool congrats! | 15:00 |
@lisitsyn | lets see if editors are happy with it ;) | 15:00 |
zeller | I'll keep my fingers crossed | 15:00 |
@lisitsyn | hah thanks | 15:00 |
@iglesiasg | zeller: thanks! | 15:00 |
-!- zeller [~zeller@embln.embl.de] has left #shogun [] | 15:01 | |
-!- erlenda [~erlenda@fw-oslo.intelcom.no] has quit [Quit: Leaving] | 15:33 | |
-!- gsomix [~Miranda@188.168.2.121] has joined #shogun | 15:42 | |
gsomix | I'm at home finally | 15:43 |
-!- van51 [~van51@195.251.235.71] has left #shogun ["QUIT :Leaving."] | 15:54 | |
-!- iglesiasg [d58f3217@gateway/web/freenode/ip.213.143.50.23] has quit [Ping timeout: 250 seconds] | 15:55 | |
-!- zeller_ [~zeller@embln.embl.de] has joined #shogun | 16:01 | |
-!- zeller_ [~zeller@embln.embl.de] has quit [Client Quit] | 16:01 | |
lambday | ah, back | 16:21 |
lambday | HeikoS: hi | 16:21 |
lambday | lisitsyn: sonne|work you guys rock :D | 16:21 |
sonne|work | lambday: what did we do? | 16:23 |
* sonne|work is innocent. | 16:23 | |
lambday | lol :D | 16:23 |
@lisitsyn | I have no idea | 16:24 |
-!- F3V3R [~CRM114@77.49.62.176.dsl.dyn.forthnet.gr] has quit [Ping timeout: 264 seconds] | 16:24 | |
lambday | hahaha | 16:25 |
lambday | I am confused about the serialization for hdf5file and json for complex64_t :-/ | 16:32 |
lambday | should I just put error messages there for complex64_t like SGObject for those two? | 16:33 |
lambday | they need it to be converted to native types for those two | 16:34 |
gsomix | cu later, guys | 16:37 |
gsomix | forgot ac adapter in the lab | 16:37 |
gsomix | :( | 16:37 |
sonne|work | lambday: SG_NOT_IMPLEMENTED maybe for now yes | 16:41 |
lambday | sonne|work: okay | 16:41 |
-!- ozan [~ozan@ee-167-078.eee.metu.edu.tr] has joined #shogun | 16:42 | |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun | 16:46 | |
-!- foulwall [~foulwall@2001:da8:215:c252:3cda:6e80:3f71:5b6] has joined #shogun | 16:50 | |
foulwall | sonney2k: just back from lab, begin tonight's work. | 16:50 |
-!- gsomix [~Miranda@188.168.2.121] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 16:53 | |
@sonney2k | foulwall, excellent! | 16:59 |
@sonney2k | lisitsyn, any idea why gsomix suffers from the same problems like you do? | 16:59 |
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has quit [Quit: Page closed] | 17:00 | |
@sonney2k | two AC adapters forgotten in <24hrs | 17:00 |
@lisitsyn | sonney2k: next time he should forget keys | 17:02 |
@lisitsyn | sonney2k: I forgot AC in another city though! :) | 17:03 |
@HeikoS | lambday: hi! | 17:09 |
@HeikoS | lambday: but if you do SG_NOTIMPLENENTED please put a warning before that | 17:09 |
@HeikoS | sgnotimplemented gives no output | 17:09 |
@HeikoS | which is a bad experience for users | 17:10 |
@sonney2k | HeikoS, well no we intend to fix this rather soon | 17:13 |
@sonney2k | HeikoS, but he has to get started | 17:14 |
@HeikoS | sonney2k: writing such a messages takes 1minute and massivle reduces frustration by users | 17:14 |
@sonney2k | HeikoS, c'mon you see the exact line of the error so ... | 17:14 |
@HeikoS | many people cannot read source code of serialization stuff | 17:15 |
lambday | HeikoS: hi | 17:16 |
@HeikoS | lambday: hi! | 17:16 |
lambday | yes | 17:16 |
lambday | making the changes | 17:17 |
lambday | I should login from a different irc client -this one gives no audio alert for msgs :-/ | 17:17 |
lambday | brb | 17:18 |
-!- lambday [67157d37@gateway/web/freenode/ip.103.21.125.55] has quit [] | 17:18 | |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has joined #shogun | 17:18 | |
lambday | back | 17:19 |
@sonney2k | gtg | 17:19 |
lambday | sonney2k: ciao ;) | 17:19 |
@HeikoS | sonney2k: bye! | 17:20 |
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 17:21 | |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has joined #shogun | 17:22 | |
-!- hoijui [~hoijui@wlan-141-23-100-225.tubit.tu-berlin.de] has quit [Quit: Leaving] | 17:32 | |
-!- ozan [~ozan@ee-167-078.eee.metu.edu.tr] has quit [Quit: Leaving] | 17:33 | |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has quit [Quit: Leaving.] | 17:34 | |
-!- pickle27 [~androirc@d67-193-243-174.home3.cgocable.net] has joined #shogun | 17:36 | |
-!- pickle27 [~androirc@d67-193-243-174.home3.cgocable.net] has quit [Ping timeout: 252 seconds] | 17:51 | |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-nsunvhvtlukipqcp] has quit [Remote host closed the connection] | 18:08 | |
-!- gsomix [~Miranda@185.6.239.62] has joined #shogun | 18:11 | |
gsomix | hey, hey. | 18:11 |
gsomix | my battery is full now | 18:12 |
-!- lisitsyn [~blackburn@185.6.239.62] has joined #shogun | 18:47 | |
-!- mode/#shogun [+o lisitsyn] by ChanServ | 18:47 | |
-!- gsomix [~Miranda@185.6.239.62] has quit [Read error: Connection reset by peer] | 19:17 | |
-!- gsomix [~gsomix@185.6.239.62] has joined #shogun | 19:19 | |
@sonney2k | gsomix, ohh men with full battery are sexy! | 19:47 |
-!- hushell [~hushell@c-67-189-100-116.hsd1.or.comcast.net] has quit [Ping timeout: 264 seconds] | 19:50 | |
gsomix | sonney2k, lol | 19:59 |
gsomix | sonney2k, just have received data from github. slowly... | 20:00 |
@sonney2k | gsomix, well for testing you can generate a data set yourself. just write some few lines inpython | 20:18 |
gsomix | sonney2k, yes, of course. | 20:18 |
@sonney2k | gsomix, and as an example implement a line count | 20:18 |
gsomix | sonney2k, doing it now. :) | 20:18 |
@sonney2k | and byte count based on string length or so | 20:18 |
@sonney2k | very good | 20:18 |
@sonney2k | gsomix, btw when you do benchmarks (it should be as fast as cat file >/dev/null) | 20:20 |
@sonney2k | gsomix, you should flush caches | 20:20 |
@sonney2k | gsomix, sync; echo 3 > /proc/sys/vm/drop_caches | 20:20 |
@sonney2k | gsomix, I am on debian since the dawn of ages | 20:20 |
-!- sanyam [uid10602@gateway/web/irccloud.com/x-tqyfqhhucroulpqk] has joined #shogun | 20:22 | |
gsomix | sonney2k, stable/testing/sid? I plan to try - it would be useful. | 20:23 |
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has joined #shogun | 20:26 | |
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has quit [Ping timeout: 248 seconds] | 20:33 | |
lambday | HeikoS: things are becoming problematic :( | 20:36 |
@sonney2k | gsomix, stable | 20:37 |
@sonney2k | lambday, what's going on? | 20:37 |
lambday | std::complex<T> doesn't support overloaded operators for say [T=int, double].. if I add this as a ptype, compiler complains about this at many places | 20:38 |
lambday | things I did so far - added in common, datatype... added serialization things... added methods in Parameter class.. | 20:40 |
lambday | also edited class_list.cpp.py | 20:41 |
lambday | added in types | 20:41 |
lambday | shall we add this as a struct type or something? | 20:42 |
@sonney2k | lambday, I don't know what you are exactly doing... where do you get this overload error? | 20:43 |
@sonney2k | lambday, overloading in SGVector? | 20:46 |
@sonney2k | or where? | 20:46 |
lambday | sonney2k: for example, https://gist.github.com/lambday/5665110 | 20:47 |
lambday | I'm checking... got a few errors in compare_ptype() too in Parameter class.. trying to fix it | 20:50 |
@sonney2k | lambday, the only way to get around these errors is to define specialized templates for that particular type that do nothing or SG_ERROR | 20:52 |
lambday | sonney2k: ummm.. like? | 20:53 |
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Leaving] | 20:54 | |
@sonney2k | wait a second but sum += vec1[m_idx[i]] * vec2[dsf->m_idx[i]] | 20:55 |
@sonney2k | so this dot() operation expects a float64_t as result | 20:55 |
@sonney2k | so either you don't define subsetfeatures for your complex type | 20:56 |
lambday | but how's that possible? we need SGVectors<complex64_t>, and also SGMatrix<complex64_t>... | 20:57 |
lambday | when I do it for SGVectors, doesn't that automatically do for that stuff for those template classes | 20:57 |
lambday | that uses SGVector<T> | 20:58 |
@sonney2k | lambday, what it wants to do is this: float64_t x = 0; x+= y*z (with y and z being complex types) | 20:59 |
@sonney2k | what should the result be? | 20:59 |
@sonney2k | x+=y is not defined | 20:59 |
lambday | yes | 20:59 |
@sonney2k | lambday, so what you could do is overload the dot() class to do nothing for complex64_t | 21:01 |
@sonney2k | err dot() function | 21:01 |
lambday | sonney2k: but I get similar type of error in many many places... | 21:01 |
@sonney2k | lambday, I guess all dot features like dense/sparse... | 21:01 |
lambday | like for example, ‘static T shogun::CMath::abs(T) [with T = std::complex<double>]’ | 21:01 |
@sonney2k | lambday, yeah ... it makes totally sense again... how do you define abs for complex... | 21:02 |
@sonney2k | you have to implement that | 21:02 |
lambday | sonney2k: ohkay | 21:03 |
lambday | hmmm.. | 21:03 |
lambday | but say, we don't want to have abs for complex64_t | 21:03 |
@sonney2k | lambday, then you still have to implement a dummy one doing an SG_NOT_IMPLEMENTED at least | 21:04 |
lambday | sonney2k: hmmm.. okay | 21:04 |
@sonney2k | lambday, I had the same issue when I did stringfeatures (which are basically just varialbe length vectors) | 21:04 |
@sonney2k | and some functions that only make sense for string but not float | 21:04 |
@sonney2k | I did stuff like | 21:05 |
@sonney2k | template<> void CStringFeatures<float32_t>::compute_symbol_mask_table(int64_t max_val) | 21:05 |
@sonney2k | { | 21:05 |
@sonney2k | } | 21:05 |
@sonney2k | just to please the compiler. | 21:05 |
lambday | okay... | 21:05 |
@sonney2k | (in features/StringFeatures.cpp) | 21:05 |
@sonney2k | no idea how to improve that | 21:05 |
lambday | but then this has to be done for many places... pleasing the compiler is a heavy task then | 21:06 |
lambday | sonney2k: but this helps a lot | 21:06 |
lambday | :) | 21:06 |
@sonney2k | I don't know what to suggest... it was all designed with just a scalar in mind. so now some operations just don't make sense for complex types | 21:07 |
@sonney2k | and while the container is still reasonable some functions are not | 21:07 |
@sonney2k | so only solution would be to drop these functions altogether | 21:08 |
@sonney2k | and implement only for the types that support it this function | 21:08 |
@sonney2k | e.g. in the CMath::abs case | 21:08 |
@sonney2k | you would implement it for all types except complex_t | 21:08 |
lambday | hmm... | 21:09 |
lambday | okay.. | 21:09 |
@sonney2k | with the solution I used in string features I was lazy and said - well I don't need this for just these few types (float*) so I am just giving it NOP functiosn | 21:09 |
@sonney2k | IMHO it is not too bad to do the same again for complex_t but it certainly is a bit hacky | 21:10 |
lambday | sonney2k: okay then I'm checking where do we have to do this for complex types.. | 21:10 |
lambday | so you did this for string features | 21:10 |
lambday | I'm checking | 21:10 |
lambday | https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/features/StringFeatures.cpp#L1869 | 21:13 |
lambday | here onwards, right | 21:13 |
lambday | hmmm | 21:13 |
@sonney2k | yeah | 21:19 |
@sonney2k | lambday, maybe you can do it better than that with some macro | 21:19 |
lambday | sonney2k: nice idea.. where shall I define that? | 21:20 |
lambday | this at least requires for most of the math functions to be rewritten for complex64_t | 21:21 |
@sonney2k | lambday, it needs to be some central place lib/common.h or lib/DataType.h? | 21:21 |
lambday | sonney2k: hmm | 21:21 |
@sonney2k | lambday, they might make sense though | 21:21 |
@sonney2k | I didn't know that you are planning something that intrusive | 21:22 |
lambday | I didn't know this complex64_t is gonna make it that complex :'( | 21:22 |
@lisitsyn | hey I am just to note that | 21:22 |
@sonney2k | I was just thinking you need that for SGVector/Matrix and that's it | 21:22 |
@sonney2k | lambday, yes? | 21:22 |
@lisitsyn | dot in case of complex is still float64 | 21:22 |
lambday | sonney2k: we need SGVector/Matrix.. and math functions for complex | 21:23 |
@sonney2k | lisitsyn, true but it still needs to be implemented then | 21:23 |
@lisitsyn | sonney2k: I once tried complex svm :D | 21:23 |
@sonney2k | lambday, maybe not all feature types already? | 21:23 |
@lisitsyn | fourier iirc | 21:23 |
@lisitsyn | sonney2k: I told that already but I get exactly the same results | 21:24 |
lambday | :'( | 21:24 |
@sonney2k | lisitsyn, same results? | 21:25 |
@lisitsyn | sonney2k: yeah if you just FFT features you get the same | 21:25 |
@lisitsyn | as dot products are preserved :) | 21:25 |
@sonney2k | lisitsyn, now that you have your paper submitted - could you please fix vgorbati's test? | 21:29 |
@lisitsyn | sonney2k: t-sne? | 21:29 |
@sonney2k | lisitsyn, dont' remember exactly the thing from this morning | 21:29 |
@lisitsyn | sonney2k: yes that | 21:29 |
gsomix | sonney2k, about SGString. for line count I want use object of SGString like a buffer. but I can't because length of string in SGString and size of internal array in memory are same: many free/mallocs are needed for reading. | 21:46 |
@sonney2k | gsomix, errm? | 21:48 |
@sonney2k | gsomix, I thought you return a copy of the line? | 21:48 |
@sonney2k | and have some line buffer internally? | 21:48 |
gsomix | sonney2k, yes. | 21:48 |
@sonney2k | so yes one free / line | 21:48 |
@sonney2k | that should be insignificant compared to reading that line from disk | 21:49 |
@sonney2k | like 100 times faster | 21:49 |
gsomix | sonney2k, hm, ok. | 21:51 |
@sonney2k | gsomix, and how fast is it? | 22:01 |
gsomix | sonney2k, there are some memory related issues. сan you wait a little? | 22:10 |
-!- lisitsyn [~blackburn@185.6.239.62] has quit [Ping timeout: 256 seconds] | 22:17 | |
gsomix | sonney2k, http://pastebin.com/a1GgMeH5 | 22:19 |
gsomix | just one run | 22:20 |
gsomix | for chunk_size = 255 and buffer_size=16*1024*1024. | 22:20 |
gsomix | Human.dat | 22:20 |
@sonney2k | gsomix, did you flush buffers? | 22:23 |
gsomix | sonney2k, time strongly depends on chunk_size: Human.dat contains many ~141 symbol lines. | 22:23 |
gsomix | sonney2k, caches? yes, of course. | 22:23 |
@sonney2k | how big is the file 100MB right? | 22:24 |
gsomix | sonney2k, yep. ~90MB. | 22:24 |
gsomix | I think I should write something like cyclic internal buffer for _fully_ using of all read data. | 22:25 |
@sonney2k | gsomix, yes that makes a lot of sense. I recall that we had something like that for the streamingfeatures. but better start clean and convert all readers to this framework | 22:27 |
@sonney2k | gsomix, how do you do it now? | 22:28 |
@sonney2k | I mean when buffersize is say 1MB? | 22:28 |
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client] | 22:30 | |
gsomix | sonney2k, didn't parse. "do" what? :( | 22:31 |
@sonney2k | gsomix, when you have a buffersize of 1MB | 22:39 |
@sonney2k | what would be done when you read lines of size 141 bytes? | 22:39 |
gsomix | read 1MB at begin of buffer -> find delimiter -> fseek to delimiter -> copy line of size 141 into SGString | 22:41 |
gsomix | it's thriftily, yep | 22:41 |
gsomix | *it's not | 22:42 |
@sonney2k | gsomix, ahh ok so you don't use the 1MB buffer at all? | 22:44 |
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has joined #shogun | 22:44 | |
@sonney2k | but have to seek if the buffer is too big for every string? | 22:44 |
gsomix | sonney2k, yep. | 22:45 |
gsomix | but buffer_size is configurable. so if we know information about lines in file, reader can work more thriftily and faster | 22:47 |
@sonney2k | gsomix, bah - it should work faster for bigger buffer size. so do the rolling thing. I would do it the following way: | 22:48 |
@sonney2k | write some rolling buffer class | 22:48 |
@sonney2k | (based on dynarray) then whenever one reaches the end of the rolling buffer without finding the delimiter copy the remaining bytes to the front | 22:49 |
@sonney2k | and fill buffer again | 22:49 |
gsomix | sonney2k, got it. | 22:50 |
@sonney2k | gsomix, alright please try to get this to work such that we can move on... | 22:56 |
gsomix | time to sleep now | 23:12 |
gsomix | nite | 23:12 |
@sonney2k | gsomix, nite! | 23:19 |
-!- iglesiasg [d58f321d@gateway/web/freenode/ip.213.143.50.29] has joined #shogun | 23:42 | |
-!- mode/#shogun [+o iglesiasg] by ChanServ | 23:42 | |
--- Log closed Wed May 29 00:00:22 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!