--- Log opened Sun Apr 08 00:00:19 2012 | ||
n4nd0 | sonney2k: covertree sucks :( | 00:10 |
---|---|---|
@sonney2k | ? | 00:11 |
@sonney2k | slow? | 00:11 |
n4nd0 | super slow | 00:11 |
@sonney2k | hehe | 00:11 |
n4nd0 | I am testing it with the files Sergey has sent | 00:11 |
n4nd0 | I am just trying to classify 10 vectors | 00:11 |
@sonney2k | so blackburn slowed everything down with covertrees :D | 00:11 |
n4nd0 | the old KNN method takes about 2 second | 00:11 |
@sonney2k | *lol* | 00:11 |
n4nd0 | with covertree ... it's not finishing ... | 00:12 |
n4nd0 | more than a minute now | 00:12 |
blackburn | sonney2k: no, edrt stuff is faster with covertree | 00:12 |
blackburn | n4nd0: do you precompute distance matrix there? | 00:12 |
n4nd0 | blackburn: any problem with cover trees for high dimensions? | 00:12 |
n4nd0 | blackburn: no | 00:13 |
blackburn | n4nd0: no, it depends only on distance | 00:13 |
blackburn | some intrinsic dimensionality stuff is here | 00:13 |
blackburn | however it is not of interest there I think | 00:13 |
n4nd0 | I don't know | 00:14 |
n4nd0 | but I think that even the covertree impl. we are using is not super good | 00:14 |
n4nd0 | this result is weird | 00:14 |
blackburn | not weird | 00:14 |
blackburn | it computes all distances a few times | 00:14 |
blackburn | it is time demanding | 00:14 |
blackburn | however how long does it take to construct a tree? | 00:15 |
n4nd0 | I am going to check that now | 00:15 |
blackburn | do you separate these processes? | 00:15 |
n4nd0 | I think that the problem is there | 00:15 |
n4nd0 | they are separate | 00:15 |
n4nd0 | construction is in train | 00:16 |
blackburn | yes | 00:16 |
blackburn | so how long train takes? | 00:16 |
n4nd0 | I think that the operation that takes a life time is training | 00:16 |
n4nd0 | it is not ending directly :) | 00:16 |
n4nd0 | we are talking about a huge bunch of vectors though | 00:16 |
n4nd0 | but does cover tree performance takes this into account? | 00:17 |
blackburn | it should be log N | 00:17 |
n4nd0 | I mean, when they say that it is for KNN classification | 00:17 |
n4nd0 | do they take into account construction? | 00:17 |
blackburn | no, construction is not considered there | 00:17 |
blackburn | you would need to construct it only once | 00:17 |
blackburn | n4nd0: I can say that knn (k=10) takes a hour or so to compute outputs with naive quick sort | 00:18 |
blackburn | on that data | 00:18 |
blackburn | all vectors I mean | 00:18 |
n4nd0 | using 1000 vectors for training | 00:19 |
n4nd0 | it took 13 seconds to construct the tree | 00:19 |
blackburn | and 2? | 00:19 |
blackburn | K | 00:19 |
n4nd0 | k = 8 | 00:20 |
n4nd0 | apply takes more time :( | 00:20 |
blackburn | how much? | 00:20 |
n4nd0 | two times more | 00:21 |
n4nd0 | but the number of vectors is really small | 00:21 |
blackburn | actually this apply *can't* take more time than qsort one | 00:21 |
blackburn | I do not understand how can that be | 00:21 |
blackburn | qsort approach computes distances to each vector | 00:21 |
blackburn | and covertree computes a few | 00:22 |
n4nd0 | it makes no sense with intuition but | 00:22 |
n4nd0 | there must sth there | 00:22 |
n4nd0 | I am going to try with more vectors fro apply | 00:22 |
n4nd0 | 500 for apply and 1000 for training | 00:22 |
n4nd0 | I would expect here the cover tree | 00:23 |
n4nd0 | not taking into account training | 00:23 |
blackburn | no need to take train time into account | 00:23 |
n4nd0 | is a shit | 00:23 |
gsomix | good night | 00:23 |
n4nd0 | sorry for the expression :P | 00:24 |
blackburn | nite | 00:24 |
n4nd0 | gsomix: good night! | 00:24 |
n4nd0 | >>>>> Classification with quick sort took 2.95830392838 | 00:24 |
n4nd0 | >>>>> Cover tree construction took 13.0743379593 | 00:24 |
n4nd0 | >>>>> Classification with cover tree took 5.47460794449 | 00:24 |
blackburn | n4nd0: haha | 00:24 |
blackburn | okay 1000 apply? | 00:24 |
n4nd0 | I can tell you | 00:24 |
n4nd0 | 6 sec with quick sort | 00:24 |
n4nd0 | 15 with cover bush | 00:25 |
blackburn | ??! | 00:25 |
blackburn | 15??? | 00:25 |
n4nd0 | just guessing | 00:25 |
n4nd0 | want to make your bet? | 00:25 |
blackburn | it shouldn't be possible | 00:25 |
n4nd0 | >>>>> Classification with quick sort took 6.00689578056 | 00:25 |
n4nd0 | >>>>> Cover tree construction took 13.0408000946 | 00:25 |
n4nd0 | >>>>> Classification with cover tree took 10.9036741257 | 00:25 |
n4nd0 | ^ 1000 in train and 1000 in apply | 00:26 |
blackburn | 2K? | 00:26 |
blackburn | factor of 2 | 00:26 |
n4nd0 | where do you want to end up :P? | 00:26 |
blackburn | both | 00:26 |
n4nd0 | train as well? | 00:26 |
blackburn | 10.90/5.47 | 00:27 |
blackburn | 6.00/2.95 | 00:27 |
blackburn | both ~2 | 00:27 |
blackburn | unbelievable | 00:27 |
blackburn | justinbeliebable lol | 00:27 |
n4nd0 | >>>>> Classification with quick sort took 12.0923390388 | 00:27 |
n4nd0 | >>>>> Cover tree construction took 13.0758159161 | 00:27 |
n4nd0 | >>>>> Classification with cover tree took 21.8708460331 | 00:27 |
n4nd0 | 1000 train, 2000 apply | 00:27 |
n4nd0 | maybe I am doing sth wrong | 00:28 |
-!- gsomix [~gsomix@85.26.233.247] has quit [Ping timeout: 246 seconds] | 00:28 | |
n4nd0 | http://snipt.org/uhpf5 | 00:28 |
PhilTillet | Hmm, I can't find any knn c++ example, is there any? | 00:29 |
blackburn | no | 00:29 |
PhilTillet | that's why I can't find one :p | 00:29 |
n4nd0 | I think that if we use covertree + opencl support | 00:29 |
blackburn | n4nd0: all is ok.. | 00:29 |
blackburn | no | 00:29 |
n4nd0 | maybe we can beat quick sort | 00:29 |
blackburn | n4nd0: nooooooo | 00:30 |
n4nd0 | I was ironic / joking :) | 00:30 |
n4nd0 | I think it must be an error in my integration in KNN | 00:31 |
blackburn | I just glanced over it | 00:32 |
blackburn | all seems to be ok | 00:33 |
@sonney2k | no one beats my quicksort :) | 00:33 |
n4nd0 | sonney2k: hahaha | 00:33 |
@sonney2k | it is 100 times faster than qsort() from libc *lol* | 00:33 |
blackburn | sonney2k: wtf? | 00:34 |
blackburn | how can that be | 00:34 |
n4nd0 | sonney2k: serious? | 00:34 |
n4nd0 | sonney2k: I think you should tell them ... | 00:34 |
blackburn | 100? | 00:34 |
@sonney2k | no | 00:34 |
@sonney2k | not serious | 00:34 |
@sonney2k | it is faster but not sure how much | 00:34 |
blackburn | why it is? | 00:35 |
@sonney2k | it is because they do advances stuff | 00:35 |
@sonney2k | advanced | 00:35 |
@sonney2k | I don't | 00:35 |
n4nd0 | that are never used? | 00:35 |
@sonney2k | n4nd0, just benchmark it ... maybe it got fixed by now | 00:35 |
n4nd0 | sonney2k: what is your intuition about the cover tree? | 00:36 |
blackburn | n4nd0: btw you may use [:,:1000] | 00:36 |
blackburn | instead of range | 00:36 |
blackburn | range is a bad thing | 00:37 |
n4nd0 | you think so? | 00:37 |
n4nd0 | I actually changed :1000 to range | 00:37 |
blackburn | yes, range computes whole range | 00:37 |
blackburn | why? | 00:37 |
n4nd0 | mmm I don't really know why | 00:37 |
@sonney2k | I think they do quicksort first then switch to bubble for small bin sizes | 00:37 |
n4nd0 | let's focus on doing a toolbox for sorting | 00:38 |
blackburn | n4nd0: what if you take say 10 features? | 00:39 |
n4nd0 | blackburn: train or apply? | 00:39 |
blackburn | is covertree faster then? | 00:40 |
n4nd0 | I guess you say for training | 00:40 |
blackburn | just truncate feature matrices | 00:40 |
blackburn | rows | 00:40 |
n4nd0 | ah ok | 00:40 |
n4nd0 | sorry | 00:40 |
n4nd0 | I am trying first with the whole bunch of sample | 00:41 |
n4nd0 | samples* | 00:41 |
n4nd0 | but that may take a lot | 00:41 |
blackburn | it looks like we have shity covertree :) | 00:42 |
n4nd0 | wow | 00:44 |
n4nd0 | we are almost there | 00:44 |
n4nd0 | >>>>> Classification with quick sort took 0.359225988388 | 00:45 |
n4nd0 | >>>>> Cover tree construction took 0.405721187592 | 00:45 |
n4nd0 | >>>>> Classification with cover tree took 0.390501976013 | 00:45 |
n4nd0 | get worse with more samples ... | 00:45 |
n4nd0 | >>>>> Classification with quick sort took 3.57726192474 | 00:46 |
n4nd0 | >>>>> Cover tree construction took 0.408113956451 | 00:46 |
n4nd0 | >>>>> Classification with cover tree took 4.02059102058 | 00:46 |
@sonney2k | n4nd0, try with few training examples and *lots* for testing | 00:46 |
n4nd0 | >>>>> Classification with quick sort took 0.715730905533 | 00:47 |
n4nd0 | >>>>> Cover tree construction took 0.00958108901978 | 00:47 |
n4nd0 | >>>>> Classification with cover tree took 1.58414793015 | 00:47 |
n4nd0 | 100 training examples, 100000 test vectors | 00:47 |
n4nd0 | using just 10 features | 00:47 |
blackburn | 100000& | 00:47 |
blackburn | where did you get so many | 00:47 |
blackburn | ;) | 00:47 |
n4nd0 | haha | 00:48 |
@sonney2k | ok then screw this cover tree - blackburn, gsomix get JL's to work | 00:48 |
n4nd0 | blackburn: I did testdat[:10, :100000] | 00:48 |
n4nd0 | blackburn: none complained :) | 00:49 |
n4nd0 | blackburn: I don't understand why it does not complain | 00:51 |
n4nd0 | blackburn: testdat.shape = (580, 12630) | 00:51 |
blackburn | n4nd0: it cuts to 100000 | 00:52 |
blackburn | hmm | 00:52 |
blackburn | okay LLE | 00:52 |
blackburn | so just takes all | 00:52 |
blackburn | with 4000 vectors | 00:52 |
blackburn | 0.6s to compute neighborhood matrix | 00:52 |
blackburn | and with 8000 vector | 00:52 |
blackburn | 1.58s | 00:53 |
blackburn | kind of wrong asymptotic... | 00:53 |
blackburn | fuck that | 00:53 |
blackburn | we really need proper covertree then | 00:53 |
blackburn | n4nd0: okay postpone this shit for now | 00:54 |
n4nd0 | blackburn: ok | 00:54 |
n4nd0 | blackburn: fuck, I lost time doing the integration :( | 00:55 |
n4nd0 | waste | 00:55 |
blackburn | n4nd0: no, interface is similar | 00:55 |
n4nd0 | blackburn: at least :) | 00:55 |
blackburn | would be some patch rather than new reintegration | 00:55 |
blackburn | it works in that covertree_point_t way as well | 00:55 |
blackburn | n4nd0: gsomix actually was near to finish that | 00:56 |
blackburn | I hope he didn't delete that stuff | 00:56 |
n4nd0 | blackburn: let's see | 00:56 |
blackburn | okay time to got to bed | 00:57 |
n4nd0 | blackburn: curiosity, do you remember what the author of this covertree said? | 00:57 |
blackburn | n4nd0: hmm? that one we have? | 00:57 |
blackburn | I found that on github | 00:57 |
blackburn | I didn't contact him | 00:57 |
n4nd0 | ok | 00:57 |
n4nd0 | yeah | 00:57 |
n4nd0 | time to sleep | 00:57 |
blackburn | but probably it is a problem | 00:57 |
blackburn | see you | 00:58 |
n4nd0 | good night people! | 00:58 |
blackburn | nite | 00:58 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Quit: Page closed] | 00:58 | |
-!- n4nd0 [~nando@h121n3-ksg-a11.ias.bredband.telia.com] has quit [Quit: leaving] | 00:58 | |
* sonney2k starts to watch revision live stream | 00:58 | |
-!- flxb [~cronor@e177094201.adsl.alicedsl.de] has joined #shogun | 01:39 | |
-!- flxb [~cronor@e177094201.adsl.alicedsl.de] has left #shogun [] | 01:40 | |
PhilTillet | good night everybody! | 01:47 |
-!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has quit [Quit: Leaving] | 01:47 | |
shogun-buildbot | build #460 of java_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/java_modular/builds/460 | 02:06 |
-!- pluskid [~chatzilla@111.120.42.124] has joined #shogun | 02:33 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 02:58 | |
pluskid | why is sonney2k always online? :D | 03:40 |
pluskid | I'm scanning some of your chatting logs last night | 03:40 |
pluskid | concerning cover tree, I accidently encounter this implementation the other day: https://github.com/mbrucher/CoverTree/ | 03:41 |
pluskid | The author also has a blog post on this: http://feedproxy.google.com/~r/eifelle/CPPV/~3/eN8ZuHOetUY/ | 03:41 |
pluskid | Maybe you could compare with this implementation to identify where the problem is? | 03:42 |
-!- pluskid [~chatzilla@111.120.42.124] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 11.0/20120312181643]] | 04:06 | |
-!- pluskid [~chatzilla@111.120.42.124] has joined #shogun | 04:08 | |
-!- pluskid [~chatzilla@111.120.42.124] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 11.0/20120314124128]] | 04:14 | |
-!- pluskid [~chatzilla@111.120.42.124] has joined #shogun | 04:16 | |
-!- pluskid [~chatzilla@111.120.42.124] has quit [Ping timeout: 252 seconds] | 04:21 | |
-!- pluskid [~chatzilla@173.254.214.60] has joined #shogun | 04:21 | |
@sonney2k | pluskid, which timezone are you in? | 06:19 |
pluskid | GMT+8 | 06:19 |
pluskid | sonney2k: I'm in Asia | 06:20 |
@sonney2k | pluskid, so it is sth like 14 hrs at your place? | 06:26 |
pluskid | 12:16 PM actually | 06:27 |
pluskid | 26 | 06:27 |
@sonney2k | ahh ok | 06:27 |
pluskid | are you here for all 24 hours ? :D | 06:27 |
pluskid | It seems you are always online | 06:27 |
@sonney2k | btw, the buildbot figured out there are a few warnings with your patch | 06:27 |
@sonney2k | http://shogun-toolbox.org/buildbot/builders/octave_static/builds/629/steps/compile/logs/warnings%20%288%29 | 06:27 |
pluskid | OK | 06:28 |
pluskid | I will fix it | 06:28 |
@sonney2k | pluskid, yeah well znc :) | 06:28 |
@sonney2k | thanks | 06:28 |
@sonney2k | pluskid, if you can come up with some other small patches within the next weeks it would be helpful | 06:29 |
pluskid | OK, I see | 06:29 |
pluskid | I will try to find something to work on | 06:29 |
@sonney2k | we don't know #slots yet but we will try to come up with some kind of pre-selection by the end of tomorrow | 06:30 |
pluskid | if I didn't find interesting small stuffs, I will come and ask here | 06:30 |
@sonney2k | ok gtg | 06:30 |
@sonney2k | cu | 06:30 |
@sonney2k | thanks | 06:30 |
pluskid | that's cool | 06:30 |
pluskid | cu | 06:30 |
-!- gsomix [~gsomix@85.26.235.174] has joined #shogun | 08:40 | |
gsomix | good morning :) | 08:46 |
pluskid | morning | 08:48 |
-!- pluskid [~chatzilla@173.254.214.60] has quit [Remote host closed the connection] | 08:49 | |
-!- pluskid [~pluskid@173.254.214.60] has joined #shogun | 08:53 | |
-!- n4nd0 [~nando@h121n3-ksg-a11.ias.bredband.telia.com] has joined #shogun | 09:20 | |
n4nd0 | pluskid: hey! I just read your mail, thank you for your findings :) | 09:20 |
pluskid | n4nd0: :) | 09:22 |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun | 09:22 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] | 09:22 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 09:22 | |
pluskid | hope the problem could be solved soon | 09:22 |
-!- pluskid [~pluskid@173.254.214.60] has quit [Quit: Konversation terminated!] | 09:22 | |
-!- pluskid [~chatzilla@173.254.214.60] has joined #shogun | 09:23 | |
n4nd0 | pluskid: but it is tru that it might because of the dataset | 09:23 |
pluskid | yes, it might be | 09:24 |
pluskid | but if it is the case | 09:24 |
pluskid | at least we could figure out cover tree is good for what kind of dataset | 09:24 |
pluskid | and document that | 09:24 |
n4nd0 | I thought that if we use for our test any of the one referenced here | 09:24 |
n4nd0 | http://hunch.net/~beygel/cover-tree.pdf | 09:24 |
n4nd0 | on the last page | 09:24 |
n4nd0 | cover tree should be better then | 09:25 |
pluskid | yes, that would be a good try | 09:26 |
pluskid | they used larger dataset | 09:27 |
pluskid | 5000-point | 09:27 |
pluskid | as mentioned in the paper | 09:27 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 09:53 | |
-!- n4nd0 [~nando@h121n3-ksg-a11.ias.bredband.telia.com] has quit [Ping timeout: 260 seconds] | 10:02 | |
-!- gsomix [~gsomix@85.26.235.174] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 10:19 | |
-!- gsomix [~gsomix@85.26.235.174] has joined #shogun | 10:20 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun | 10:22 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] | 10:22 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 10:22 | |
-!- hpsoar [7c108901@gateway/web/freenode/ip.124.16.137.1] has joined #shogun | 10:44 | |
-!- hpsoar [7c108901@gateway/web/freenode/ip.124.16.137.1] has quit [Client Quit] | 10:47 | |
-!- gsomix [~gsomix@85.26.235.174] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 11:04 | |
-!- gsomix [~gsomix@85.26.235.174] has joined #shogun | 11:04 | |
-!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun | 12:08 | |
gsomix | blackburn, yo | 12:15 |
blackburn | gsomix: hi, I'll merge your stuff in a min | 12:15 |
blackburn | gsomix: as for covertree I have one suggestion | 12:16 |
blackburn | gsomix: did you delete your covertree modifications? I hope not | 12:16 |
gsomix | blackburn, I did not delete. | 12:17 |
blackburn | nice | 12:17 |
blackburn | see KNN sources | 12:17 |
blackburn | actually all you would need | 12:18 |
blackburn | is to change block at KNN.cpp:113 | 12:18 |
blackburn | where covertree is constructed | 12:18 |
gsomix | blackburn, do you mean KNN in shogun? | 12:19 |
blackburn | yes | 12:19 |
blackburn | and 426 | 12:19 |
blackburn | where neighbors are enqueued | 12:20 |
gsomix | blackburn, and..? | 12:21 |
blackburn | gsomix: hmm my suggestion is to use your adopted covertree there :) | 12:22 |
pluskid | blackburn: hi, what's the status of regression support in shogun? | 12:22 |
pluskid | it seems incomplete? | 12:22 |
blackburn | as we measured some stuff and it turns that dncrane's covertree is rather slow | 12:22 |
blackburn | pluskid: in which means? | 12:22 |
pluskid | for example | 12:23 |
pluskid | why train_machine() in LinearRidgeRegression protected? | 12:23 |
pluskid | and LeastSquareRegression is almost empty ? | 12:24 |
blackburn | pluskid: in shogun machines train delegates all the stuff to virtual protected train_machine() | 12:24 |
blackburn | oh I see | 12:24 |
pluskid | I mean, when a method is protected, it cannot be invoked by others | 12:25 |
blackburn | train can | 12:25 |
blackburn | that's ok | 12:25 |
pluskid | oh, train invokes train_machine ? | 12:25 |
blackburn | yes | 12:25 |
pluskid | ok | 12:25 |
blackburn | ahah | 12:25 |
blackburn | got it | 12:25 |
blackburn | pluskid: that's ok with leastsquares | 12:25 |
blackburn | it is LinearRidge with tau = 0 | 12:26 |
pluskid | got it :) | 12:26 |
-!- gsomix [~gsomix@85.26.235.174] has quit [Ping timeout: 246 seconds] | 12:26 | |
blackburn | that would be hard to approach any large scale with it though | 12:28 |
pluskid | I'm considering to add LARS, if n4nd0 hasn't started to work on it yet | 12:31 |
CIA-64 | shogun: Evgeniy Andreev master * r126a1de / examples/undocumented/python_modular/kernel_custom_modular.py : minor fix - http://git.io/W8JYdw | 12:31 |
CIA-64 | shogun: Evgeniy Andreev master * rc5c7117 / (3 files in 2 dirs): fixed set_triangle_kernel_matrix_from_full in custom kernel - http://git.io/JGPY7Q | 12:31 |
blackburn | no idea | 12:31 |
pluskid | mlpack seems to got an implementation of LARS | 12:31 |
pluskid | with elasticnet | 12:32 |
CIA-64 | shogun: Sergey Lisitsyn master * r79a91ad / src/shogun/evaluation/ClusteringEvaluation.cpp : Merge branch 'normalized-mutual-info' of git://github.com/pluskid/shogun - http://git.io/eSsHvg | 12:32 |
-!- gsomix [~gsomix@85.26.164.224] has joined #shogun | 12:33 | |
-!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has joined #shogun | 12:35 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 13:12 | |
pluskid | lapack is horrible... | 13:19 |
pluskid | where can I find a function list, with doc on which parameters is what? | 13:19 |
pluskid | for lapack & clbas | 13:19 |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun | 13:36 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] | 13:36 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 13:36 | |
wiking | happy easter y'all, except blackburn, you'll have it later :P | 13:46 |
pluskid | what is easter y' ? | 13:58 |
PhilTillet | lol | 13:58 |
wiking | :D | 14:01 |
gsomix | >> except blackburn | and me. I live in a floor lower than blackburn. :] | 14:03 |
gsomix | *on | 14:04 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds] | 14:05 | |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 14:17 | |
-!- V[i]ctor [~victor@host-176-100-246-254.masterbit.su] has joined #shogun | 14:31 | |
-!- gsomix [~gsomix@85.26.164.224] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 14:41 | |
CIA-64 | shogun: Heiko Strathmann master * rb7b8742 / (21 files in 7 dirs): | 14:49 |
CIA-64 | shogun: Merge pull request #428 from karlnapf/master | 14:49 |
CIA-64 | shogun: subset stacks (+24 more commits...) - http://git.io/u8aL2w | 14:49 |
-!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has quit [Read error: Connection reset by peer] | 14:54 | |
blackburn | pluskid: http://www.netlib.org/lapack/explore-html/index.html | 14:59 |
-!- PhilTillet [~Philippe@vir78-1-82-232-38-145.fbx.proxad.net] has joined #shogun | 15:04 | |
-!- blackburn [~qdrgsm@83.234.54.186] has quit [Ping timeout: 240 seconds] | 15:06 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 15:11 | |
-!- V[i]ctor [~victor@host-176-100-246-254.masterbit.su] has left #shogun [] | 15:13 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 15:25 | |
pluskid | blackburn: thanks!! | 15:26 |
-!- gsomix [~gsomix@188.168.5.161] has joined #shogun | 15:35 | |
shogun-buildbot | build #694 of libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/694 blamelist: heiko.strathmann@gmail.com | 15:39 |
-!- harshit_ [~harshit@182.68.113.64] has joined #shogun | 15:53 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 16:08 | |
n4nd0 | pluskid: I have not started with LARS yet, I am planning to fix the testsuite so fell free to work on that if you want | 16:11 |
pluskid | n4nd0: OK, I'm starting on this | 16:11 |
pluskid | btw: Lapack and blas is really hard to use ... | 16:12 |
n4nd0 | you think so? | 16:13 |
n4nd0 | I am not an expert but I managed to make it work for the couple of things I have used it around here | 16:14 |
blackburn | I don't think it is :) | 16:14 |
pluskid | haven't got familiar with it yet | 16:15 |
pluskid | doing something like matrix multiple | 16:15 |
pluskid | and solving linear system | 16:15 |
pluskid | makes me headache | 16:15 |
n4nd0 | matrix multiple? | 16:15 |
pluskid | C = A * B this easy | 16:16 |
n4nd0 | ah ok multiplication ;) | 16:16 |
pluskid | ah, yeah | 16:16 |
pluskid | maybe I need some time to get familiar with its naming convention | 16:17 |
pluskid | it could be better if their doc are supplied with examples | 16:18 |
n4nd0 | now that we are talking about it | 16:18 |
pluskid | for example | 16:18 |
pluskid | in a function for multiplication of two matrices | 16:19 |
n4nd0 | blackburn: what is the main difference between lapack and BLAS? | 16:19 |
blackburn | n4nd0: BLAS is basic ops | 16:19 |
n4nd0 | I think that lapack includes BLAS but I am not sure about it | 16:19 |
blackburn | like Ax=b | 16:19 |
pluskid | Lapack built on BLAS | 16:19 |
blackburn | and Lapack is solving/etc | 16:19 |
blackburn | eigenproblems | 16:19 |
blackburn | factorization | 16:19 |
blackburn | svd | 16:19 |
blackburn | etc | 16:19 |
pluskid | they provide options to indicate that one of the matrix is transposed or not | 16:19 |
blackburn | BLAS is all about three levels | 16:19 |
blackburn | vector-vector operations | 16:19 |
blackburn | like dot | 16:19 |
pluskid | and you have to indicate the shape (dimension) of the matrix | 16:19 |
blackburn | that is Level 1 | 16:20 |
pluskid | but I don't know whether the shape should be before transpose or after... | 16:20 |
blackburn | matrix-vector operations | 16:20 |
blackburn | Level 2 | 16:20 |
n4nd0 | and matrix-matrix level 3 | 16:20 |
n4nd0 | :) | 16:20 |
shogun-buildbot | build #695 of libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/695 | 16:20 |
blackburn | exactly | 16:20 |
shogun-buildbot | build #632 of r_static is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/632 blamelist: heiko.strathmann@gmail.com | 16:21 |
n4nd0 | pluskid: I think that options are actually to change the operations like | 16:21 |
n4nd0 | pluskid: if you want to do A*v or A'*v you use the same function but you use an option for that | 16:21 |
shogun-buildbot | build #654 of cmdline_static is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/654 blamelist: heiko.strathmann@gmail.com | 16:21 |
pluskid | n4nd0: yes, but in the function, you have to pass the shape (m * n) of A | 16:21 |
blackburn | pluskid: in the doc they say that sizes are of Op(A) | 16:21 |
blackburn | so Op(A) is A or A' | 16:22 |
pluskid | if A is transposed, shall I pass m, n or n, m ? | 16:22 |
pluskid | OK | 16:22 |
n4nd0 | pluskid: I'd say you pass those parameters before transposed | 16:22 |
pluskid | I'll read the doc again | 16:22 |
pluskid | didn't find the detailed doc before blackburn gave me that link | 16:23 |
n4nd0 | pluskid: I have been using this one too http://www.gnu.org/software/gsl/manual/html_node/Level-2-GSL-BLAS-Interface.html | 16:23 |
pluskid | blackburn: you are right, I should use size of Op(A) | 16:24 |
pluskid | n4nd0: according to the doc, we should use the size *after* transposed :) | 16:25 |
shogun-buildbot | build #633 of octave_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/633 blamelist: heiko.strathmann@gmail.com | 16:25 |
pluskid | n4nd0: the API of GSL-BLAS seems slightly different from what we are using? | 16:25 |
blackburn | yeah totally different | 16:25 |
pluskid | it seems GSL is doing some more wrapping | 16:26 |
pluskid | while the package we are using is straightforward fortran translation | 16:26 |
blackburn | no we are tied to basic API | 16:27 |
blackburn | it would stay the same for scalapck | 16:27 |
blackburn | or cuda lapack | 16:27 |
blackburn | or any other lapack | 16:27 |
blackburn | it is standard api while gsl's is not | 16:27 |
pluskid | OK, I see | 16:28 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Remote host closed the connection] | 16:29 | |
shogun-buildbot | build #620 of python_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/620 blamelist: heiko.strathmann@gmail.com | 16:29 |
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun | 16:29 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 16:32 | |
shogun-buildbot | build #633 of r_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/633 | 16:34 |
shogun-buildbot | build #655 of cmdline_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/655 | 16:38 |
shogun-buildbot | build #634 of octave_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/634 | 16:43 |
shogun-buildbot | build #621 of python_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/621 | 16:48 |
shogun-buildbot | build #485 of octave_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/485 blamelist: heiko.strathmann@gmail.com | 17:01 |
n4nd0 | benchmarking cover tree with letter subset ... | 17:03 |
n4nd0 | in the paper on cover trees they say cover tree should be about ten times faster here | 17:04 |
blackburn | n4nd0: btw I checked things | 17:04 |
blackburn | and in my LLE implementation | 17:04 |
blackburn | covertree is 3-4 times faster than quicksort | 17:04 |
blackburn | on a range 4K - 8K 3d vectors | 17:04 |
n4nd0 | mmm | 17:04 |
n4nd0 | something fishy going on here then | 17:04 |
n4nd0 | it sucks | 17:05 |
n4nd0 | >>>>> Classification with quick sort took 51.0038590431 | 17:05 |
n4nd0 | >>>>> Cover tree construction took 99.7238249779 | 17:05 |
n4nd0 | >>>>> Classification with cover tree took 106.712816 | 17:05 |
blackburn | yeah | 17:05 |
blackburn | n4nd0: actually gsomix is on the way to integrate JL's | 17:05 |
n4nd0 | that is with 10000 vectors for training and classifcation | 17:05 |
n4nd0 | but I don't understand why is it faster with your tests | 17:06 |
n4nd0 | I am afraid I may have done something wrong :S | 17:06 |
blackburn | no I don't think so | 17:07 |
blackburn | pluskid measured distance evaluations and it is computed too often | 17:07 |
n4nd0 | aham | 17:07 |
n4nd0 | it needs to compute distances quite often to apply queries too | 17:08 |
shogun-buildbot | build #472 of csharp_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/472 blamelist: heiko.strathmann@gmail.com | 17:08 |
pluskid | do you see JL's reply in the mailing list? | 17:17 |
blackburn | did he reply?? | 17:18 |
blackburn | can't see anything | 17:18 |
pluskid | yes | 17:18 |
n4nd0 | no, I don't | 17:18 |
pluskid | he said: | 17:18 |
pluskid | Can you compute the expansion constant spectrum as in the original | 17:18 |
pluskid | covertree paper? | 17:18 |
blackburn | did he answer directly? | 17:18 |
blackburn | expansion constant spectrum? wtf? | 17:18 |
pluskid | he replied to shogun-list | 17:18 |
blackburn | :D | 17:18 |
n4nd0 | shouldn't it be here http://news.gmane.org/gmane.comp.ai.machine-learning.shogun? | 17:19 |
pluskid | I don't know | 17:19 |
pluskid | maybe the gmane archive has delay | 17:19 |
pluskid | he replied 3 hours ago | 17:19 |
pluskid | if you subscribed to the mailing list | 17:19 |
n4nd0 | didn't get in my inbox either | 17:19 |
pluskid | you should receive one in your inbox | 17:19 |
blackburn | haha | 17:19 |
n4nd0 | yeah, I am subscribed | 17:19 |
blackburn | yeah I have subscribed recently | 17:19 |
pluskid | strange | 17:20 |
pluskid | anyway | 17:20 |
blackburn | year ago or so | 17:20 |
blackburn | :D | 17:20 |
pluskid | that's what he said | 17:20 |
pluskid | though I don't understand | 17:20 |
pluskid | I haven't read the paper | 17:20 |
blackburn | can you paste 'To'? | 17:20 |
pluskid | To: Chiyuan Zhang <pluskid@gmail.com> | 17:21 |
pluskid | Cc: shogun-list@shogun-toolbox.org | 17:21 |
pluskid | shogun-list doesn't accept CC? | 17:21 |
blackburn | it does | 17:21 |
pluskid | who implemented our current cover tree? maybe we can ask him what is the expansion constant spectrum? | 17:22 |
n4nd0 | well, I think the cover tree is taken from somewhere else | 17:22 |
blackburn | yes | 17:22 |
pluskid | hmm | 17:22 |
pluskid | blackburn: you mentioned that gsomix is merging JL's ? | 17:22 |
blackburn | yes | 17:22 |
pluskid | so JL has an official implementation now? | 17:22 |
blackburn | yes | 17:23 |
pluskid | cool | 17:23 |
pluskid | maybe we could wait to see what about that one | 17:23 |
n4nd0 | yes | 17:23 |
pluskid | since none of us know the internal of how cover tree works | 17:23 |
n4nd0 | I just wanted to check whether the test data was affecting | 17:23 |
blackburn | JL's one is better anyway | 17:23 |
pluskid | is JL's implementation publicly available now? | 17:24 |
blackburn | yes | 17:24 |
pluskid | I think we should at least reply him in the mail | 17:25 |
blackburn | yes could you please? | 17:25 |
pluskid | sure | 17:25 |
blackburn | because we can't see the email :D | 17:25 |
pluskid | maybe the mail arrived at my inbox because he also sent to me | 17:26 |
pluskid | and the shogun-list is super slow | 17:26 |
pluskid | maybe | 17:26 |
blackburn | yes could be | 17:26 |
n4nd0 | pluskid: in his web-page there is an implementation of isomap that includes cover tree | 17:26 |
pluskid | I would say one of us is porting JL's implementation now, if we still have problem with his implementation, we will contact him to try to figure out the problem | 17:26 |
blackburn | pluskid: yes thanks | 17:27 |
pluskid | n4nd0: cool, he is still updating his Isomap code! | 17:27 |
n4nd0 | pluskid: I think is another's guy isomap | 17:27 |
pluskid | n4nd0: ?? | 17:28 |
pluskid | JL is one of the authors of Isomap | 17:28 |
n4nd0 | pluskid: look under application | 17:28 |
n4nd0 | http://hunch.net/~jl/projects/cover_tree/cover_tree.html | 17:28 |
n4nd0 | another's guy isomap's implementation :) | 17:28 |
n4nd0 | I don't really know if that sense makes sense using normal English | 17:28 |
n4nd0 | but I think you got the idea, or? | 17:29 |
n4nd0 | :D | 17:29 |
pluskid | Oh, I see | 17:29 |
pluskid | :D | 17:29 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 17:30 | |
blackburn | one great thing about us is we are able to write some meanless things | 17:30 |
blackburn | while english is foreign for us :) | 17:30 |
pluskid | english foreign for me, too | 17:31 |
n4nd0 | if some native speaker was here, he/she would probably get a headache :D | 17:31 |
pluskid | n4nd0: JL's page provided some dataset for cover tree testing, maybe you could try out your KNN on that | 17:31 |
blackburn | pluskid: yes we know :) | 17:32 |
blackburn | we have no native speakers there | 17:32 |
blackburn | at all | 17:32 |
pluskid | haha | 17:32 |
blackburn | the only active at the ML native speaker is Evan | 17:32 |
blackburn | we should consult him about language too as well as git :D | 17:33 |
pluskid | his IRC id is ? | 17:33 |
blackburn | I've never seen him there | 17:33 |
n4nd0 | me neither | 17:33 |
blackburn | pluskid: is it true there is no first/last name separation in china? | 17:34 |
pluskid | not ture | 17:34 |
pluskid | but the first/last is reversed | 17:34 |
blackburn | so Zhang is a? | 17:35 |
pluskid | actually, I'm always confused | 17:35 |
pluskid | I'd rather prefer given name vs. family name | 17:35 |
pluskid | Zhang is family name, Chiyuan is given name | 17:35 |
blackburn | aha! I see | 17:35 |
pluskid | English people write Chiyuan Zhang | 17:35 |
pluskid | but in China, we always use Zhang Chiyuan | 17:35 |
blackburn | we tend to reverse that stuff too | 17:35 |
blackburn | in russian | 17:36 |
pluskid | I don't know which one is first which is last... | 17:36 |
blackburn | last name is family name I think :) | 17:36 |
blackburn | we also make use of patronymes there | 17:36 |
blackburn | that's why "o" letter is in my email address | 17:37 |
blackburn | pluskid: my brother works for huawei there | 17:37 |
pluskid | cool! | 17:37 |
pluskid | huawei is a big company in China | 17:37 |
blackburn | yeah | 17:38 |
pluskid | biggest local network device company according to my knowledge | 17:38 |
blackburn | not only local afaik | 17:38 |
blackburn | even backbones are constructed with huawei equipment | 17:38 |
pluskid | wow | 17:39 |
blackburn | actually they have beaten cisco here I think | 17:39 |
blackburn | biggest mobile operator, megafon, uses mainly huawei stuff | 17:40 |
pluskid | here ? means where? | 17:40 |
blackburn | russia | 17:40 |
pluskid | wow! | 17:40 |
pluskid | I didn't know they have made such achievements | 17:41 |
n4nd0 | they have two very big countries then | 17:41 |
pluskid | can't find "patronymes" in my dictionary | 17:41 |
blackburn | pluskid: hmm we include father's name | 17:42 |
n4nd0 | I think they say patronymic | 17:42 |
blackburn | yeah | 17:42 |
blackburn | probably my bad | 17:42 |
pluskid | Oh, I see now | 17:43 |
blackburn | n4nd0: are your middle names related to your family's members? | 17:43 |
blackburn | :) | 17:43 |
pluskid | so where are you from n4nd0 ? | 17:43 |
n4nd0 | pluskid: Spain, resident in Sweden | 17:43 |
n4nd0 | blackburn: they are, but it is just a choice | 17:44 |
n4nd0 | I mean, some people say like I will call my son with my father's name | 17:44 |
blackburn | n4nd0: so Jos? is your father's name? | 17:44 |
n4nd0 | but we don't use patronyms or matronyms | 17:44 |
n4nd0 | blackburn: no, Jos? is the masculin for my grandmother's name | 17:44 |
blackburn | oooh | 17:45 |
blackburn | :) | 17:45 |
n4nd0 | blackburn: Fernando is my godfather's | 17:45 |
pluskid | haha, I heard that in some restaurant at Spain, people can use some kind of Chinese dialect to order dishes :D | 17:45 |
blackburn | n4nd0: what about third one then? | 17:45 |
n4nd0 | pluskid: probably is they way in which we would speak Chinese :P | 17:45 |
n4nd0 | blackburn: well Iglesias and Garc?a are just surnames, my father's family name and my mother's | 17:46 |
blackburn | ah! | 17:46 |
pluskid | but that kind of dialect is very different from our official language, actually, I totally can't understand | 17:46 |
pluskid | complicated name system! | 17:46 |
n4nd0 | pluskid: aham, it sounds something more serious then what you are talking about | 17:46 |
pluskid | IRC IDs would be enough :p | 17:47 |
pluskid | there are many many dialects in Southern China | 17:47 |
blackburn | pluskid: from which part of chinae you are? | 17:48 |
blackburn | china* | 17:48 |
pluskid | around Western | 17:48 |
pluskid | previously I thought many here would be Germany :) | 17:49 |
pluskid | as shogun is originally developed from MPI, am I right? | 17:49 |
blackburn | pluskid: well yes | 17:50 |
blackburn | Soeren is from Berlin | 17:50 |
pluskid | blackburn: if you are in Russia, you must be in some similar timezone as me | 17:52 |
blackburn | pluskid: wow, why? | 17:52 |
blackburn | we have 9 timezones :D | 17:52 |
pluskid | er... | 17:52 |
pluskid | we are using a unified timezone in china | 17:53 |
pluskid | everywhere is Beijin time | 17:53 |
pluskid | GMT+8... | 17:53 |
blackburn | western one is kaliningrad - GMT+1 | 17:53 |
pluskid | big difference | 17:53 |
blackburn | and eastern is some kamchatka GMT+9 probably | 17:53 |
blackburn | let me check | 17:53 |
shogun-buildbot | build #486 of octave_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/486 | 17:54 |
pluskid | never mind | 17:54 |
pluskid | so what's your local time now? | 17:54 |
pluskid | it's 23:54 here | 17:55 |
blackburn | 19/55 | 17:55 |
blackburn | 19.55 | 17:55 |
blackburn | I'm in GMT+3 timezone | 17:55 |
blackburn | or +4? | 17:55 |
blackburn | :D | 17:55 |
pluskid | hmm, quite different | 17:55 |
n4nd0 | 17:55 here | 17:55 |
pluskid | so you use a dot to separate time in Russia? | 17:56 |
pluskid | n4nd0: even earlier | 17:56 |
blackburn | not really | 17:56 |
blackburn | both are ok I think :) | 17:57 |
blackburn | : or . | 17:57 |
pluskid | hmm | 17:57 |
blackburn | and kaliningrad is +3 | 17:57 |
blackburn | I was wrong | 17:57 |
blackburn | okay anadyr is +12 | 17:57 |
blackburn | 8 timezones still | 17:57 |
blackburn | would be bad to have night at say 14:30 and day at 3:30 :) | 17:58 |
shogun-buildbot | build #473 of csharp_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/473 | 18:01 |
-!- pluskid [~chatzilla@173.254.214.60] has quit [Read error: Connection reset by peer] | 18:01 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has joined #shogun | 18:06 | |
-!- wiking [~wiking@78-23-191-201.access.telenet.be] has quit [Changing host] | 18:06 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 18:06 | |
-!- harshit_ [~harshit@182.68.113.64] has quit [Remote host closed the connection] | 18:07 | |
-!- pluskid [~chatzilla@111.120.42.124] has joined #shogun | 18:08 | |
pluskid | night | 18:08 |
n4nd0 | good night | 18:08 |
blackburn | nite | 18:09 |
-!- pluskid [~chatzilla@111.120.42.124] has quit [Client Quit] | 18:09 | |
-!- gsomix [~gsomix@188.168.5.161] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 18:11 | |
-!- gsomix [~gsomix@188.168.5.161] has joined #shogun | 18:11 | |
gsomix | hi all | 18:11 |
n4nd0 | gsomix: hey! | 18:12 |
blackburn | hmm am I here? | 18:19 |
gsomix | yep | 18:19 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 18:21 | |
n4nd0 | I will see you later guys | 18:22 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 18:23 | |
wiking | :> | 18:24 |
wiking | blackburn_: r u here? :) | 18:24 |
blackburn_ | ??? | 18:24 |
blackburn_ | yes | 18:24 |
blackburn_ | :) | 18:24 |
wiking | :D | 18:24 |
gsomix | blackburn_, ??? | 18:24 |
wiking | anyhow i have to test now mkl :> | 18:24 |
wiking | blackburn_: any suggestions which mkl to use for multi class case? | 18:25 |
blackburn_ | mkl? whoa that's nice :D | 18:25 |
blackburn_ | I do not know anything about mkl :D | 18:26 |
wiking | ah ok | 18:26 |
wiking | i'll have to try how it works for me | 18:26 |
wiking | i have 3 different sets of features | 18:26 |
wiking | and seems like a good idea | 18:27 |
wiking | to use different kernels for different feature sets | 18:27 |
blackburn_ | yeah may be | 18:27 |
wiking | since i've tried it with only concatenating all the features | 18:27 |
wiking | but i wonder if it really makes a big difference | 18:27 |
wiking | if i use mkl | 18:27 |
wiking | i guess it should :P | 18:28 |
blackburn_ | I think it depends whether they have rather different structure | 18:28 |
wiking | yeah they doo | 18:28 |
wiking | one of it is a very sparse feature set... that is coming from text processing | 18:29 |
wiking | while the other is like a histogram feature from images | 18:29 |
wiking | until now what i've done is to concatenate them l2-norm them and then use polykernel | 18:29 |
wiking | and as we know using intersection or jensen shannon kernel for those histogram features are giving me better results... | 18:30 |
blackburn_ | in my experience it is not very good to use any dot product on histograms | 18:30 |
wiking | yeah... | 18:30 |
wiking | so i'm thinking about now the mkl | 18:30 |
wiking | but i guess the only way to see the outcome is to actually try it :>>> | 18:31 |
wiking | just quite lazy | 18:31 |
blackburn_ | what about structured output latent model multiple kernel learning svm? | 18:31 |
blackburn_ | :D | 18:31 |
wiking | well | 18:32 |
wiking | that'd be amazng | 18:32 |
wiking | :D | 18:32 |
wiking | i wonder where's alex | 18:32 |
wiking | i guess he is taking his easter holidays... | 18:32 |
blackburn_ | wiking: afaik it uses gmnp | 18:32 |
wiking | blackburn_: yeah i've just checked the code | 18:33 |
wiking | i mean the headers... | 18:33 |
blackburn_ | while gmnp is very good it should be nice too | 18:33 |
wiking | well i guess now it's time for me to do the coding | 18:34 |
wiking | the reading in the various features | 18:34 |
wiking | and then do the setting up of the mkl | 18:34 |
wiking | and see what i get | 18:34 |
wiking | ;) | 18:34 |
wiking | ok i need to switch to vpn | 18:34 |
wiking | brb | 18:34 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 18:35 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 18:35 | |
wiking | re | 18:35 |
wiking | :) | 18:35 |
wiking | heheh it's funny to see the amount of members now being decreased of the #shogun channel.... almost back to it's normal 8 members state before the gsoc fever :) | 18:36 |
blackburn_ | yeah | 18:37 |
blackburn_ | however it will increase tomorrow I think | 18:40 |
wiking | why? :) | 18:41 |
blackburn_ | well it is sunday today | 18:41 |
wiking | hahahah your are an optimist :> | 18:43 |
wiking | blackburn_: remember i've mentioned about my patch for mahout a couple of months ago | 18:47 |
wiking | they've managed to merge it now like 2 days ago.... | 18:47 |
blackburn_ | yes | 18:47 |
blackburn_ | oooh fast | 18:47 |
wiking | took them like 2-2.5 months | 18:47 |
wiking | and it was like a 5 liner patch... nothing serious ... :( | 18:48 |
blackburn_ | yeah they are serious guys | 18:48 |
blackburn_ | :) | 18:48 |
wiking | :)))) | 18:49 |
* gsomix hates photometrics stuff. | 18:55 | |
@sonney2k | ahh nice and quiet :) | 19:00 |
gsomix | sonney2k, at home? in chat? | 19:01 |
@sonney2k | at home | 19:02 |
@sonney2k | :) | 19:02 |
@sonney2k | kids are asleep and I am going to shogun mode now :) | 19:03 |
gsomix | ^_^ | 19:03 |
@sonney2k | gsomix, I see that you patch got merged already? so did you have time to think about what to do next? | 19:07 |
@sonney2k | I mean your preferences? | 19:07 |
gsomix | sonney2k, I want to work on SG_Vector. | 19:08 |
@sonney2k | gsomix, ok - so then first thing to do is to convert everything from returning copies of SGVector to references | 19:10 |
@sonney2k | i.e. functions doing SGVector foo() should be SGVector& foo() | 19:10 |
@sonney2k | same for input arguments | 19:11 |
blackburn_ | INTRUSIVE | 19:11 |
@sonney2k | next step is then to just copy & paste the ref/unref code from SGObject | 19:11 |
@sonney2k | HIGHLY INTRUSIVE | 19:11 |
@sonney2k | blackburn_, that is why we should do this ASAP | 19:11 |
blackburn_ | sonney2k: ah that is kind of painful | 19:12 |
blackburn_ | penetration lol | 19:12 |
@sonney2k | blackburn_, it is not as bad as last years double* -> SGVector transition though | 19:12 |
blackburn_ | sonney2k: what about typemaps? | 19:13 |
@sonney2k | gsomix, then the next step would be to remove destroy_vector / etc | 19:13 |
blackburn_ | any changes are required? | 19:13 |
@sonney2k | blackburn_, yes sure add a & | 19:13 |
blackburn_ | nothing more? | 19:13 |
@sonney2k | gsomix, and to go through all the code and use the SG_REF / UNREF macro on sgvectors | 19:14 |
blackburn_ | gsomix: what about covertree? :) | 19:14 |
@sonney2k | that is sth we probably all have to help | 19:14 |
@sonney2k | blackburn_, isn't pluskid working on this? | 19:14 |
blackburn_ | sonney2k: no | 19:14 |
gsomix | blackburn_, in spare time. | 19:14 |
blackburn_ | gsomix: ok | 19:14 |
n4nd0 | sonney2k: he's focusing on LARS I think | 19:14 |
blackburn_ | yes | 19:14 |
n4nd0 | I can work on cover tree if it is ok | 19:15 |
blackburn_ | huh | 19:15 |
blackburn_ | :D | 19:15 |
blackburn_ | why not | 19:15 |
n4nd0 | I don't think I am doing so much with the testsuite | 19:15 |
blackburn_ | gsomix: can you introduce your pretty ready work to n4nd0 ? | 19:15 |
gsomix | blackburn_, yep. But it's a little ugly. ^_^'' | 19:17 |
blackburn_ | gsomix: it doesn't really matter if it is faster | 19:18 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has left #shogun [] | 19:18 | |
n4nd0 | gsomix: what is the state of it? | 19:18 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 19:18 | |
blackburn_ | oops | 19:18 |
gsomix | n4nd0, many functions and global variables. And some stuff for v_array. | 19:19 |
n4nd0 | gsomix: what is the implementation based on? JL's paper or another implementation? | 19:20 |
gsomix | n4nd0, JL code. | 19:20 |
gsomix | *JL's | 19:20 |
n4nd0 | the one here right? http://hunch.net/~jl/projects/cover_tree/cover_tree.html | 19:21 |
n4nd0 | first link after the enumerate with 3 points | 19:21 |
gsomix | n4nd0, yep. templated version. | 19:21 |
n4nd0 | gsomix: ok | 19:21 |
gsomix | n4nd0, I even think it's better to start anew. | 19:22 |
n4nd0 | gsomix: oh, do you think so? | 19:22 |
gsomix | it's seems my code contains errors. | 19:22 |
blackburn_ | gsomix: can you remember any pitfalls you have struggled with? | 19:22 |
gsomix | blackburn_, yep. | 19:23 |
gsomix | it's related with differences in v_arrays, our ans JC's. | 19:23 |
gsomix | wait a sec | 19:24 |
blackburn_ | ans sounds a little french ;) | 19:24 |
gsomix | huh :) | 19:25 |
gsomix | n4nd0, http://dl.dropbox.com/u/19029407/to_n4nd0.zip | 19:25 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 19:26 | |
n4nd0 | gsomix: cool, that's your implementation so far right? | 19:26 |
gsomix | n4nd0, yep. | 19:26 |
n4nd0 | gsomix: what did you do more or less? take the methods in JL's code and add template for the point object | 19:27 |
n4nd0 | more changes? | 19:27 |
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 19:28 | |
gsomix | n4nd0, and formatting. :) | 19:29 |
gsomix | n4nd0, there are little changes to our v_array support. | 19:30 |
n4nd0 | gsomix: and what's the problem of this implementation? | 19:30 |
gsomix | n4nd0, memory leaks. | 19:31 |
gsomix | in push and alloc | 19:31 |
n4nd0 | gsomix: ok | 19:31 |
n4nd0 | but for the rest, is it functional? | 19:31 |
n4nd0 | I mean, is it complete? | 19:31 |
gsomix | n4nd0, iirc, yes. | 19:32 |
n4nd0 | gsomix: cool then | 19:32 |
@sonney2k | well ask John Langford about the leaks then (with some small example code) | 19:33 |
gsomix | n4nd0, maybe I'm wrong. Just try to run. | 19:34 |
n4nd0 | gsomix: ok, I will take a look to understand a bit how it works and later run | 19:34 |
n4nd0 | sonney2k: is it v_array better than std::vector too? :) | 19:35 |
@sonney2k | I am not a big fan of any std:: stuff .... simply because std::vector is not even specified to be something contiguous etc | 19:36 |
@sonney2k | so *everything* is better :D | 19:36 |
gsomix | sonney2k, I will start to work tomorrow. | 19:36 |
gsomix | At now it's time to optics. :) | 19:37 |
@sonney2k | gsomix, you mean on shogun or some job :) | 19:37 |
n4nd0 | sonney2k: I am a big fan of STL but you are starting to make me doubt about it :O | 19:37 |
gsomix | sonney2k, shogun, of course. | 19:37 |
@sonney2k | gsomix, ok | 19:37 |
@sonney2k | n4nd0, maybe I am just not knowledgeable enough about it | 19:38 |
n4nd0 | sonney2k: I have never analyzed its efficiency to tell the truth, I think that's the part you critize about it | 19:40 |
blackburn | actually it is as well efficient afaik | 19:41 |
n4nd0 | criticize* | 19:41 |
@sonney2k | blackburn, where is it specified that vector is contiguous in memory? | 19:45 |
blackburn | it is not probably however that does not implies it is very uncontiguous :) | 19:47 |
n4nd0 | blackburn: why wouldn't it be contiguous? | 19:47 |
blackburn | n4nd0: because of resize | 19:48 |
blackburn | hmm however some says it is | 19:49 |
blackburn | sonney2k: http://stackoverflow.com/questions/672352/how-is-dynamic-memory-managed-in-stdvector | 19:49 |
n4nd0 | blackburn: this v_array class we have in shogun, have you worked with it? | 19:53 |
blackburn | n4nd0: yeah had used it somewhere.. any problem? | 19:54 |
n4nd0 | blackburn: yeah, for example the method pop | 19:55 |
n4nd0 | inline T pop() { return *(--end); } | 19:55 |
n4nd0 | shouldn't be a check around there? | 19:55 |
blackburn | it isnot safe :) | 19:55 |
n4nd0 | because it is meant to be so? | 19:55 |
gsomix | sonney2k, can I ask you? :) | 19:56 |
@sonney2k | gsomix, please always ask - not ask if you can ask! | 19:57 |
@sonney2k | n4nd0, well speed trade off | 19:58 |
@sonney2k | checks cost quite a bit of performance when you do them *often* | 19:58 |
gsomix | sonney2k, it's offtop. :) what have you studied in your minor/major courses at the university? | 19:58 |
blackburn | n4nd0: yeah it tries to be as fast as it could be | 19:59 |
n4nd0 | blackburn: ok | 19:59 |
@sonney2k | gsomix, well major: computer science, minor: physics | 19:59 |
blackburn | I never knew you studied physics :) | 20:01 |
gsomix | sonney2k, thanks. :) do you like physics? | 20:02 |
@sonney2k | gsomix, back then yes - now I like machine learning and traffic :D | 20:02 |
gsomix | I have major in applied math and physics. It's long. | 20:08 |
gsomix | I cut up to "physics" in my proposal :) | 20:08 |
gsomix | because blackburn is more a mathematician than I. :3 | 20:13 |
@sonney2k | alright guess I've requested 8 slots for gsoc (I think we really have 8 very very good students that applied) and 9 if there was an inf amount of slots available | 20:13 |
@sonney2k | spare time for me now! | 20:14 |
* sonney2k goes watching the revision demo party live stream | 20:14 | |
blackburn | demoscene memories? ;) | 20:15 |
@sonney2k | yeah | 20:15 |
@sonney2k | I am too old :D | 20:15 |
@sonney2k | I mean I went to 'scene' parties since 1993 or so | 20:15 |
blackburn | sonney2k: really? I was 2 :D | 20:16 |
n4nd0 | blackburn: I doubled your age at that time :P | 20:17 |
gsomix | now it is clear how the CPU Fire Applet appeared. demoscene! :) | 20:17 |
blackburn | yeah kind of | 20:17 |
blackburn | lorenz stuff as well I guess | 20:18 |
n4nd0 | is any of you guys fan of topcoder or programming contests in general? | 20:20 |
n4nd0 | just curious | 20:20 |
blackburn | oh I do not like that stuff pretty much :) | 20:20 |
gsomix | n4nd0, just codeforces (russian analog). but I'm not very good in sports programming. | 20:21 |
n4nd0 | I think it is quite fun :) | 20:22 |
n4nd0 | but kind of difficult, lot of good people around there | 20:22 |
@sonney2k | yeah I wrote lorenz at a demoscene party :) | 20:23 |
blackburn | n4nd0: my main point is that these guys can hardly fit to real programming | 20:23 |
@sonney2k | gsomix, yes... | 20:24 |
gsomix | n4nd0, http://codeforces.ru/?locale=en this system differs from topcoder. five tasks, two hours. | 20:24 |
n4nd0 | gsomix: I have participated in codeforces ;) | 20:24 |
@sonney2k | blackburn, it is also the reason I know the uint and int operations have same costs | 20:24 |
gsomix | n4nd0, ah :) | 20:24 |
blackburn | sonney2k: yeah I know that | 20:24 |
@sonney2k | and all that low level assembly programming stuff | 20:25 |
@sonney2k | anyways have a nice evening | 20:25 |
gsomix | n4nd0, what is your nickname? | 20:25 |
blackburn | sonney2k: you too | 20:25 |
* sonney2k party time *yay* | 20:25 | |
n4nd0 | gsomix: fj.iglesias | 20:26 |
blackburn | wow it seems you are finally happy with spare time :) | 20:26 |
gsomix | n4nd0, thanks. | 20:26 |
blackburn | (that you have that background) | 20:26 |
n4nd0 | gsomix: I suck at sports programming, but I think it is fun :P | 20:27 |
PhilTillet | Hello everybodyyyy | 20:30 |
blackburn | hey | 20:30 |
PhilTillet | what was the name of that gay video yesterday ? | 20:33 |
blackburn | azis? | 20:34 |
PhilTillet | aaah yes | 20:34 |
PhilTillet | thanks | 20:34 |
PhilTillet | this is seriously completely wtf music | 20:36 |
blackburn | music is not as wtf | 20:40 |
blackburn | but video.. | 20:40 |
PhilTillet | true | 20:43 |
PhilTillet | Golden WTF Awards. | 20:43 |
gsomix | wtf music is: http://www.youtube.com/watch?v=W7PxdpBWdwA :) | 20:45 |
PhilTillet | Jaaaaapaaaaan = WTF | 20:45 |
PhilTillet | Ok this music is completely f*cked up. XD | 20:46 |
PhilTillet | Perfect background for coding. | 20:46 |
gsomix | japan and breakcore :] | 20:46 |
PhilTillet | http://www.youtube.com/watch?v=Q16KpquGsIc wtf music+video | 20:46 |
PhilTillet | :D | 20:46 |
PhilTillet | (japan...) | 20:47 |
gsomix | One my friend loves this music | 20:47 |
PhilTillet | :D | 20:47 |
PhilTillet | me too, I can actually dance it ... | 20:47 |
PhilTillet | well I could | 20:48 |
PhilTillet | didn't have time to practice last months.. | 20:48 |
gsomix | I can not imagine how you can dance to this music :) | 20:51 |
PhilTillet | it's the epilepsy dance | 20:52 |
gsomix | ou, I prefer something a little more slowly | 20:55 |
PhilTillet | :D | 20:56 |
PhilTillet | where are you from gsomix ? | 20:56 |
gsomix | Russia | 20:56 |
PhilTillet | Russia is so big | 20:57 |
PhilTillet | western Russia? | 20:58 |
gsomix | PhilTillet, yep. Samara city | 20:58 |
PhilTillet | I see :) | 21:03 |
shogun-buildbot | build #203 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/203 | 21:08 |
gsomix | shogun-buildbot, thanks. :) | 21:16 |
PhilTillet | I love this buildbot | 21:17 |
PhilTillet | blackburn, i'm pushing the nearest centroid in a few minutes I think :p | 21:21 |
shogun-buildbot | build #202 of nightly_all is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/202 | 21:26 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 21:27 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 21:51 | |
PhilTillet | blackburn, i've just sent the pull request for nearest centroids | 21:57 |
blackburn | okay nice | 21:57 |
PhilTillet | as far as i've tested it, it worked | 21:57 |
blackburn | I guess wrong branch | 21:58 |
PhilTillet | ah true, I had some problem with git, i forgot to put it in another branch sorry :/ | 21:58 |
blackburn | it is included to opencl | 21:59 |
PhilTillet | aaaaaah true | 21:59 |
PhilTillet | >_< | 21:59 |
blackburn | yeah better add new branch :) | 21:59 |
PhilTillet | you are right, i'll fix that | 22:01 |
blackburn | ok | 22:01 |
PhilTillet | but it might take some time :D | 22:02 |
PhilTillet | i wish i was a git expert | 22:02 |
PhilTillet | okay, done :) | 22:07 |
PhilTillet | now i must find how to cancel the one on the opencl branch | 22:10 |
PhilTillet | XD | 22:10 |
PhilTillet | I think I can't, could you delete that commit in the wrong branch? | 22:11 |
blackburn | no I can't | 22:13 |
blackburn | you can | 22:13 |
blackburn | you can revert it actually | 22:14 |
PhilTillet | yes | 22:14 |
PhilTillet | i've just figured it out | 22:14 |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 22:24 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun | 22:41 | |
PhilTillet | blackburn, what did you mean with "May be it would be better to use matrix there and then construct features" ? | 22:45 |
blackburn | PhilTillet: you construct features from matrix and then enqueue its feature vectors | 22:46 |
blackburn | why not to construct matrix and then construct features? | 22:46 |
PhilTillet | Okay, I'll do that | 22:49 |
-!- genix [~gsomix@188.168.13.216] has joined #shogun | 22:56 | |
-!- gsomix [~gsomix@188.168.5.161] has quit [Ping timeout: 264 seconds] | 22:56 | |
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has quit [Ping timeout: 245 seconds] | 23:03 | |
genix | good night, guys | 23:24 |
PhilTillet | good night :) | 23:24 |
n4nd0 | good night | 23:29 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 272 seconds] | 23:46 | |
-!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun | 23:57 | |
--- Log closed Mon Apr 09 00:00:19 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!