--- Log opened Tue Oct 09 00:00:17 2012 | ||
-!- Netsplit *.net <-> *.split quits: @sonney2k | 00:07 | |
-!- Netsplit over, joins: sonney2k | 00:19 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 00:33 | |
-!- heiko [~heiko@host86-177-176-201.range86-177.btcentralplus.com] has joined #shogun | 00:57 | |
-!- heiko [~heiko@host86-177-176-201.range86-177.btcentralplus.com] has left #shogun [] | 00:57 | |
-!- blackburn [~blackburn@188.168.128.167] has quit [Quit: Leaving.] | 01:45 | |
-!- cronor_ [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 02:11 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Read error: Connection reset by peer] | 02:11 | |
-!- cronor_ is now known as cronor | 02:11 | |
-!- cronor_ [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 02:51 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Read error: Connection reset by peer] | 02:51 | |
-!- cronor_ is now known as cronor | 02:51 | |
-!- cronor_ [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 03:08 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Read error: Connection reset by peer] | 03:08 | |
-!- cronor_ is now known as cronor | 03:08 | |
shogun-buildbot | build #127 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/127 | 03:31 |
---|---|---|
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Read error: Connection reset by peer] | 04:36 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 04:37 | |
-!- adoniscik [~emre@c-67-180-103-118.hsd1.ca.comcast.net] has joined #shogun | 06:49 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Quit: cronor] | 07:27 | |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 09:26 | |
-!- ptizoom [~ptizoom@85.210.93.120] has joined #shogun | 10:07 | |
-!- adoniscik [~emre@c-67-180-103-118.hsd1.ca.comcast.net] has quit [Quit: Leaving] | 10:12 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 10:13 | |
-!- cronor_ [~cronor@g226036186.adsl.alicedsl.de] has joined #shogun | 10:29 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Read error: Connection reset by peer] | 10:29 | |
-!- cronor_ is now known as cronor | 10:29 | |
-!- cronor [~cronor@g226036186.adsl.alicedsl.de] has quit [Quit: cronor] | 11:04 | |
wiking | sonne|work: hey, btw: do you have download statistics of shogun 2.0 release? :) | 11:15 |
sonne|work | wiking: no idea... | 11:19 |
wiking | sonne|work: do u have the apache logs?:) | 11:19 |
sonne|work | maybe that http://freecode.com/projects/shogun/date_metrics | 11:19 |
sonne|work | or that http://mloss.org/software/viewstats/2/ | 11:19 |
sonne|work | http://mloss.org/software/downloadstats/2/ | 11:20 |
sonne|work | wiking: no apache on that machine but fapws3 | 11:23 |
wiking | but it has logs right? :) | 11:24 |
sonne|work | no | 11:25 |
wiking | none? :) | 11:25 |
wiking | how come :) | 11:25 |
sonne|work | it sends me emails with error msg if sth goes wrong but no more | 11:26 |
wiking | i see :S | 11:26 |
wiking | btw do we have something like a linked list structure in shogun? | 11:26 |
sonne|work | CList ? | 11:27 |
blackburn | std::list | 11:27 |
blackburn | :D | 11:27 |
sonne|work | *tatatata* | 11:27 |
wiking | :D | 11:27 |
sonne|work | kalaschnikov | 11:27 |
blackburn | sonne|work: copy(vec.begin(), vec.end(), inserter(v, v.begin()); | 11:28 |
blackburn | sonne|work: BOOST:SHARED_PTR | 11:28 |
blackburn | feel da pain? :) | 11:29 |
sonne|work | memmove(blackburn) | 11:29 |
sonne|work | free(blackburn) | 11:29 |
blackburn | memcpy | 11:29 |
wiking | :D | 11:29 |
sonne|work | realloc(blackburn) | 11:29 |
blackburn | noooo | 11:29 |
blackburn | std::allocator<blackburn> | 11:30 |
blackburn | sonne|work: FUNCTORS! | 11:30 |
blackburn | sonne|work: CHAR_TRAITS! | 11:30 |
blackburn | sonne|work: abstract factory! singleton! | 11:33 |
wiking | :> | 11:33 |
blackburn | sonne|work: MVC! dependency injection! | 11:33 |
blackburn | wiking: CLEARCASE! | 11:33 |
wiking | yey | 11:35 |
wiking | mvc is cool | 11:35 |
blackburn | wiking: dependency injection as well | 11:36 |
blackburn | wiking: and boost and stl and all the things sonne|work hate :D | 11:36 |
wiking | hehehe | 11:36 |
wiking | yeah boost is cool | 11:36 |
wiking | it's a bit bloated though | 11:36 |
wiking | but still cool | 11:37 |
blackburn | I learned it a bit when doing test task for the company | 11:37 |
blackburn | enjoyed to do some things | 11:37 |
sonne|work | BLOOOOOOATED | 11:39 |
sonne|work | using boost increases shogun compile time by factor 10 | 11:40 |
blackburn | sonne|work: without libraries we've got a lot or redundant code as well | 11:41 |
wiking | sonne|work: but just compile time :) | 11:41 |
sonne|work | blackburn: I don't mind libraries | 11:47 |
blackburn | sonne|work: especially standard template one :) | 11:48 |
sonne|work | ...to link against | 11:51 |
blackburn | and the nobel prize goes to quantum guys | 11:53 |
sonne|work | btw | 12:05 |
sonne|work | http://wbailer.wordpress.com/2012/01/17/shogun-machine-learning-toolbox-under-win64/ | 12:05 |
wiking | "applications exists on exceptions from SG_ERROR" | 12:07 |
wiking | this is the normal case :) | 12:07 |
wiking | it's not a limitation | 12:07 |
wiking | hahah who's up for this one: "MemoryMappedFile is not implemented (needs reimplementation based on Win API | 12:08 |
wiking | :D | 12:08 |
blackburn | I don't think it is implemented even in linux :) | 12:10 |
sonne|work | wiking: well no - in cmdline interface it should just return to command prompt | 12:12 |
sonne|work | but command line ansi characters can be disabled | 12:13 |
wiking | sonne|work: what's with that elwms interface? they had it in the previous macports file | 12:50 |
wiking | so i wonder how do we support it in 2.0 | 12:50 |
wiking | i've never used it.... | 12:50 |
wiking | blackburn: have u managed clang_complete to complete for stl? | 13:04 |
blackburn | wiking: it was working IIRC | 13:19 |
blackburn | wiking: I wish clang_complete showed doc | 13:19 |
blackburn | wiking: do you know whether it is possible? | 13:19 |
blackburn | maybe it is not that hard to patch it for that, I don't know | 13:20 |
sonne|work | wiking: probably not too important... it is one .so file that can be used as R extension, python extension, octave & matlab extensiuon at the same time while even enabling to transfer matrices etc across languages (e.g. one can run python from octave etc) | 13:48 |
sonne|work | wiking: I used that to slowly transition away from matlab - but I wanted to call my newly written python code | 13:50 |
sonne|work | and e.g. use matplotlib | 13:50 |
wiking | sonne|work: oh | 13:59 |
wiking | sonne|work: i think i'll remove it then :D | 13:59 |
wiking | sonne|work: none the less the port file is now in a good shape | 13:59 |
wiking | wonder how long it'll take to apply it to the repo | 13:59 |
wiking | somebody already tested it and gave some feedback so i've chnaged some minor stuff | 13:59 |
sonne|work | wiking: probably faster than getting into debian | 14:13 |
blackburn | sonne|work: 6 minutes support reaction time! faster than all that enterprise :D | 14:34 |
blackburn | (see mailing list) | 14:34 |
sonne|work | but your answer doesn't help | 14:35 |
sonne|work | he used exactly that right? | 14:36 |
sonne|work | I think he is missing lapack/blas/atlas | 14:36 |
blackburn | sonne|work: no, he used wrong names | 14:37 |
blackburn | only one is right but yes you are right he must be missing blas | 14:37 |
sonne|work | LIBLINEAR_L2R_L1LOSS_SVC_DUAL else if (strcmp(name,"LIBLINEAR_L2R_L1LOSS_SVC_DUAL")==0) | 14:38 |
sonne|work | ok | 14:38 |
blackburn | sonne|work: we should have a QA | 14:45 |
blackburn | :D | 14:45 |
sonne|work | blackburn: we have :) | 14:46 |
blackburn | sonne|work: users? | 14:46 |
blackburn | they could give up too early | 14:46 |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] | 15:05 | |
-!- blackburn [~blackburn@37.61.179.175] has joined #shogun | 17:25 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 260 seconds] | 19:45 | |
-!- heiko [~heiko@host86-176-125-28.range86-176.btcentralplus.com] has joined #shogun | 21:00 | |
blackburn | hey heiko | 21:11 |
heiko | hey blackburn | 21:11 |
blackburn | heiko: got you message, where should I answer? :) | 21:12 |
heiko | email best, I have food on the oven and have to go there now :) | 21:12 |
heiko | I just found the pyramid match kernel that might be worth a try | 21:12 |
blackburn | heiko: is it 1/0 image? | 21:13 |
blackburn | or black/white means no color? | 21:13 |
blackburn | heiko: pyramid thing seems to be good idea but I'd recommend to think about features construction | 21:13 |
heiko | 8bit image | 21:15 |
heiko | ok | 21:16 |
heiko | i am back | 21:16 |
blackburn | heiko: back? then lets chat :) | 21:16 |
heiko | ok | 21:16 |
blackburn | heiko: any assumptions on the 2d relations? | 21:16 |
heiko | energy fields | 21:16 |
heiko | so spatial relationship or at least distance is important | 21:16 |
blackburn | anything about neighborhood? | 21:17 |
blackburn | can you say neighborhood of one 'pixel' radius matter | 21:17 |
blackburn | or more? | 21:17 |
heiko | no that should be fine | 21:18 |
heiko | the radius | 21:18 |
blackburn | first idea came to me is LBP | 21:18 |
heiko | what is that? | 21:18 |
blackburn | it is a method of constructing features of images | 21:18 |
blackburn | useful for textures and faces | 21:18 |
heiko | dont know | 21:18 |
heiko | I mean, these fields are not really images | 21:18 |
blackburn | let me briefly describe | 21:19 |
heiko | but ok | 21:19 |
blackburn | for each 'pixel' | 21:19 |
blackburn | we compare it with its neighbors | 21:19 |
blackburn | so we've got result of 8 compares | 21:19 |
blackburn | like left pixel is greater than center | 21:19 |
heiko | k | 21:19 |
blackburn | right pixel is less than center | 21:19 |
blackburn | then we compute histograms of such results | 21:20 |
blackburn | like in block B we've got mostly left greater than right | 21:20 |
blackburn | so the histogram describes pattern in the block | 21:20 |
blackburn | got what I mean? | 21:21 |
heiko | yes get it | 21:21 |
blackburn | good thing for you is that we have it in shogun :D | 21:22 |
heiko | any experience with that? :) | 21:22 |
blackburn | heiko: I know it works good on faces, not that good on road signs (but ok) | 21:22 |
blackburn | heiko: face detectors works very good with it | 21:22 |
heiko | I see | 21:23 |
heiko | work on the histograms? | 21:23 |
heiko | with svm for example | 21:23 |
blackburn | heiko: for histograms best way is to map they explicitly using HomogeneousKernelMap | 21:24 |
blackburn | and work like in typical linear space | 21:24 |
blackburn | with no kernel | 21:24 |
blackburn | fast and nearly no accuracy loss | 21:24 |
heiko | HomogeneousKernelMap just takes a bunch of histograms and produces a sparse representation of them? | 21:25 |
blackburn | heiko: no, unfortunately not sparse and of 3x size | 21:25 |
blackburn | but linear | 21:25 |
blackburn | say you have kernel | 21:25 |
blackburn | <x,y> = \sum_i min(x_i, y_i) | 21:25 |
blackburn | then HKM is a map M | 21:25 |
blackburn | <Mx,My> = k(x,y) | 21:26 |
blackburn | flaw is that Mx is have to be at least 3 times bigger | 21:27 |
heiko | I see | 21:27 |
blackburn | it comes from fourier | 21:27 |
blackburn | so sin, cos and 1 | 21:27 |
heiko | well that sounds like a good approach | 21:27 |
heiko | do you have any lbp code? | 21:27 |
blackburn | heiko: shogun/features/LBPPyrDotFeatures.cpp | 21:27 |
blackburn | :) | 21:27 |
blackburn | you could try that first | 21:28 |
heiko | no documentation :( | 21:28 |
blackburn | unfortunately | 21:28 |
heiko | but dense_dot is implemented | 21:29 |
heiko | blackburn, looks like the code does what you said | 21:30 |
blackburn | heiko: yeah I know :D | 21:30 |
blackburn | heiko: did you plot these fields? | 21:31 |
heiko | no, just reading the code | 21:31 |
blackburn | heiko: I mean ones you are classifying | 21:31 |
heiko | oh, yes I did | 21:31 |
heiko | very sparse | 21:32 |
heiko | and I cannot find a pattern | 21:32 |
heiko | with my eye and brain | 21:32 |
blackburn | what are the classes? | 21:32 |
heiko | binary | 21:32 |
heiko | svm with RBF kernel gives about 60% accuracy | 21:32 |
heiko | but using vectors of the images is not good | 21:32 |
blackburn | yes bad idea always | 21:33 |
heiko | only 1D local structure | 21:33 |
heiko | how to do this best with shogun | 21:33 |
heiko | there is only this dense_dot method | 21:33 |
blackburn | heiko: so looks like noise for you? | 21:33 |
heiko | I have to create a matrix of all the dot products and then use as custom kernel matrix right? | 21:34 |
blackburn | heiko: why? add is here to | 21:34 |
heiko | blackburn, more or less, yes | 21:34 |
blackburn | o | 21:34 |
heiko | add? | 21:34 |
blackburn | yeah line 286 | 21:34 |
heiko | what can I do with add? :) | 21:34 |
blackburn | it is COFFIN features in general | 21:34 |
blackburn | add and dense dot is the best a cat can get | 21:35 |
heiko | how to use then? | 21:35 |
blackburn | with liblinear or ocas | 21:35 |
blackburn | just like dense features | 21:35 |
heiko | so just pass these features to liblinear and thats it? | 21:35 |
blackburn | yeah if there is no bug there yes | 21:35 |
blackburn | other way is to employ opencv but I'd recommend shogun first | 21:36 |
blackburn | heiko: did you try dimreduction? | 21:36 |
heiko | why? | 21:36 |
heiko | no I didnt but this should be pointless here | 21:37 |
heiko | svm are designed to work in high dimensions | 21:37 |
blackburn | heh no, not for features construction | 21:37 |
blackburn | for visualization | 21:37 |
heiko | i see | 21:37 |
heiko | how would you do that? | 21:38 |
blackburn | heiko: is data big? | 21:38 |
heiko | never tried this | 21:38 |
heiko | individual images are not | 21:38 |
heiko | but I have 20000 of them :) | 21:38 |
blackburn | do you have 4-5 gb of free ram then? | 21:38 |
blackburn | :D | 21:38 |
heiko | I have | 21:39 |
heiko | but what would you do? | 21:39 |
blackburn | well I'd try MDS | 21:39 |
blackburn | just to see whether some compactness is here | 21:40 |
blackburn | heiko: create MultidimensionalScaling() | 21:40 |
blackburn | call set_landmark(true) | 21:40 |
blackburn | set_landmark_number(2000) | 21:41 |
blackburn | uhh I am bad | 21:41 |
blackburn | it would compute all 20000x20000 distances | 21:41 |
blackburn | still if you don't mind call set_target_dim(2) | 21:42 |
blackburn | and call embed on your features | 21:42 |
heiko | what do I get? | 21:42 |
blackburn | it will output 2d embedding (dense features) | 21:42 |
heiko | what does it do? | 21:42 |
blackburn | scatter plot it to see how data is distributed according to distance | 21:42 |
heiko | I see | 21:42 |
heiko | well, | 21:42 |
blackburn | it attempts to save distances | 21:43 |
heiko | one thing I know is: | 21:43 |
heiko | I can successfully do two-sample tests on the data | 21:43 |
heiko | so there is a difference | 21:43 |
heiko | and this already happens when using image vectors | 21:43 |
heiko | maybe I should dig a bit in there first ... | 21:43 |
blackburn | MDS tries to find 2d embedding w.r.t. distances | 21:43 |
heiko | ok | 21:44 |
heiko | distances of what? | 21:44 |
blackburn | so far vectors are far and close are close | 21:44 |
blackburn | between each other | 21:44 |
heiko | image columns? | 21:44 |
blackburn | no, vectors | 21:44 |
blackburn | between images in general | 21:44 |
heiko | these are vectors | 21:44 |
blackburn | then yes | 21:44 |
heiko | ok, thanks for the hints | 21:44 |
heiko | Ill let you know what I got | 21:44 |
blackburn | heiko: okay, will wait for that | 21:45 |
heiko | probably will take a while - you know I am slow :D | 21:45 |
blackburn | heiko: did you try to find anything about energy fields? | 21:45 |
heiko | no, but I also dont want to invest too much time now | 21:45 |
heiko | first want to get some usable results | 21:45 |
heiko | and then refine etc | 21:45 |
heiko | so imaging stuff might be a good start | 21:45 |
blackburn | I think MRF could be a good idea but I am a bit lame in graphical models | 21:46 |
heiko | yeah same here | 21:46 |
heiko | actually, there is this new kernel belief propagation which can do inference in pairwise MRF | 21:46 |
heiko | this could be wortha try | 21:46 |
blackburn | heiko: Ising model was created for something like that | 21:47 |
heiko | but I havent yet understood it :) | 21:47 |
blackburn | in 1934 :D | 21:47 |
heiko | lol :) | 21:47 |
blackburn | heiko: belief propagation kernel? interessant to see | 21:47 |
heiko | I am planning to implement that into graphlab btw :) | 21:47 |
blackburn | darf ich ansehen? | 21:47 |
heiko | If I find some time | 21:47 |
blackburn | what is graphlab? | 21:48 |
heiko | http://www.gatsby.ucl.ac.uk/~gretton/kernelBP/kernelBP.htm | 21:48 |
heiko | an open-source distriuted graph-based ML software | 21:48 |
heiko | can do stuff like pagerank in a distributed fashion | 21:48 |
blackburn | I see | 21:48 |
heiko | and graphical models fit in there very well | 21:48 |
blackburn | heiko: did I say that right in german? :D | 21:49 |
heiko | nein :) | 21:49 |
blackburn | argh | 21:49 |
heiko | what did you want to say exactly? | 21:49 |
blackburn | what is wrong? | 21:49 |
heiko | Darf ich mal sehen - but that is really slangy :) | 21:49 |
blackburn | well I tried to change typical 'darf ich vorstellen' (I learned that from a book :D) | 21:50 |
heiko | Kannst du mir das zeigen | 21:50 |
heiko | would be better | 21:50 |
blackburn | can you show me? | 21:50 |
blackburn | ^ | 21:50 |
blackburn | I know du and kannst and das | 21:50 |
heiko | there is the subject missing thats also wrong in english :) | 21:50 |
blackburn | :D | 21:50 |
blackburn | heiko: where? | 21:51 |
heiko | can you show me what? | 21:51 |
blackburn | ah well not that wrong I thought | 21:51 |
blackburn | I mean when we are talking and you got what I mean :) | 21:52 |
heiko | yes, its just a language thing | 21:52 |
heiko | you always have a subject in any roman language | 21:52 |
blackburn | yeah I know | 21:52 |
blackburn | but it is lost sometimes still, right? | 21:53 |
heiko | yes | 21:53 |
heiko | especially in chatrooms | 21:53 |
blackburn | heiko: I am now developing that test task thing - got totally crazy with design | 21:55 |
blackburn | EntityManager + EntityFactory + EntityLoader + EntityKeeper | 21:55 |
blackburn | :D | 21:55 |
heiko | lol :D | 21:55 |
blackburn | I should stop describing new entities of entities | 21:56 |
blackburn | or I will end with a metamodel | 21:56 |
blackburn | of a metamodel | 21:56 |
blackburn | heiko: may be it is good that sonney2k is here to stop my crazy intentions :D | 21:58 |
heiko | hehe | 21:58 |
blackburn | heiko: if I get that job I am going to become commercial counterpart of alex binder :D | 21:59 |
blackburn | mostly the same that he does | 22:00 |
heiko | wow | 22:00 |
heiko | cool thing | 22:00 |
heiko | you two then can join each other to create a super ML monster :LD | 22:00 |
blackburn | we should all meet and constuct a MLdzilla | 22:01 |
heiko | hehe | 22:03 |
blackburn | heiko: I recently noticed that not only shogun slowed down, scikits is being developed by other people now | 22:05 |
heiko | shogun slowed down? | 22:06 |
blackburn | well two commits per week | 22:06 |
blackburn | or so | 22:06 |
heiko | I see | 22:07 |
heiko | ok | 22:07 |
heiko | but we got the gsoc stuff | 22:07 |
heiko | that was massive this time | 22:07 |
blackburn | yeah | 22:07 |
heiko | I am still working quite a bit, but the streaming framework is so broken that I cannot commit :) | 22:07 |
heiko | spent almost all yesterday trying to fix the stuff | 22:07 |
blackburn | I will become active once I get or not get the job | 22:07 |
heiko | but hope to be done soon | 22:07 |
blackburn | now I am thinking how to design that thing not to look stupid | 22:08 |
blackburn | :D | 22:08 |
blackburn | however I have to hurry up, I said it will be ready by 11 | 22:08 |
blackburn | of oct | 22:08 |
heiko | ok | 22:09 |
heiko | then gogogog :) | 22:09 |
blackburn | it works already but they asked multithreading | 22:09 |
blackburn | thinking how to avoid redundant locks.. | 22:09 |
heiko | wow | 22:10 |
heiko | lots of things | 22:10 |
blackburn | yes | 22:10 |
blackburn | and JSON | 22:11 |
blackburn | for loading | 22:11 |
blackburn | they didn't ask for JSON but it seems to be easier for me | 22:11 |
blackburn | than handcrafting my own format | 22:11 |
blackburn | :D | 22:11 |
blackburn | ahh and UML | 22:12 |
heiko | cool !: :) | 22:12 |
heiko | alothough | 22:13 |
heiko | although I hate UML | 22:13 |
blackburn | class diagram is ok | 22:13 |
blackburn | I hate sequence diagram | 22:13 |
blackburn | I've never seen a process that is simplified by a sequence diagram | 22:14 |
blackburn | wasting paper maybe | 22:14 |
blackburn | or spending time to avoid doing things at job :) | 22:14 |
heiko | I also never saw this | 22:15 |
heiko | when the processes are easy, it works fine, but then you dont need a diagram | 22:15 |
heiko | when the processes are complicated, its a mess | 22:16 |
heiko | better use plain blackboard :) | 22:16 |
blackburn | heiko: the only thing is that there are some big companies | 22:18 |
blackburn | but they should rather disappear | 22:18 |
blackburn | oracle especially | 22:18 |
heiko | what do you mean? | 22:20 |
blackburn | they have to work with all that IDEF0, UML, anything | 22:20 |
blackburn | because they are so big that person means nothing | 22:21 |
blackburn | it should work even if robots come | 22:21 |
heiko | hehe you got a quite bad image of these guys dont you? :) | 22:30 |
blackburn | heiko: do SAP and oracle deserve respect? :) | 22:36 |
heiko | lol SAP | 22:37 |
heiko | I heard a talk of this statistical analysis software | 22:37 |
heiko | how is it called again? | 22:37 |
heiko | the guy was in our uni early this year | 22:37 |
heiko | SAS | 22:37 |
heiko | dude, this software is ridiculus | 22:37 |
blackburn | yes, SAS | 22:37 |
blackburn | heiko: no idea about that, what does it contain? kmeans and linear regression? | 22:38 |
heiko | yes | 22:38 |
heiko | and programming in comic sans :) | 22:38 |
blackburn | oracle data mining is the same thing | 22:38 |
blackburn | it has kmeans, svm and apriori | 22:38 |
blackburn | not even fp-tree, just apriori | 22:38 |
blackburn | hey 70s | 22:39 |
heiko | yes | 22:40 |
heiko | but this will change | 22:40 |
heiko | for all the new methods, you really need to know what you are doing | 22:40 |
blackburn | yeah once we get super semi svm they will get MKL | 22:40 |
heiko | yeah maybe | 22:41 |
heiko | I mean these things require quite some knowledge | 22:41 |
heiko | but KMeans does not | 22:41 |
blackburn | sure but I don't think they aim data mining products for people who doesn't know what data mining is | 22:42 |
heiko | MKL, man, this is so specialised knowledge | 22:42 |
blackburn | I understand, just some example | 22:42 |
heiko | yeah | 22:42 |
heiko | I prefer open-source stuff anyway | 22:42 |
heiko | but companies need reliable things that are known to work | 22:42 |
heiko | and these products offer this | 22:43 |
blackburn | no they need money :) | 22:43 |
heiko | they also can generate power-point presentations from data :) | 22:43 |
blackburn | yeah that's true business thing | 22:43 |
heiko | btw just running two-sample tests on the vectorised images with gaussian kernel | 22:44 |
heiko | and I get 0% type 2 error | 22:44 |
heiko | strange that the SVM fails | 22:44 |
heiko | at least now I know which kernel to use to maximise differences between the classes | 22:44 |
blackburn | heh yeah statistical approach would work too I think | 22:45 |
blackburn | however you are the expert here | 22:45 |
heiko | this stuff is related | 22:45 |
heiko | there is a formal connection between linear hard margin SVM and MMD test | 22:45 |
heiko | so if two-sample test works good, svm should work well too | 22:46 |
blackburn | there are connections between svm and everything :) | 22:46 |
heiko | indeed :) | 22:47 |
--- Log closed Wed Oct 10 00:00:13 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!