| --- Log opened Fri Apr 15 00:00:36 2011 | ||
| alesis-novik | sonney2k, then I'll make one big constructor | 00:01 |
|---|---|---|
| @sonney2k | alesis-novik, don't become demotivated. I always have many many comments in the first iteration | 00:01 |
| alesis-novik | where you pass the matrix and vector explicitly | 00:01 |
| blackburn | lenin! | 00:01 |
| blackburn | OH! warnings are here! | 00:02 |
| blackburn | ../shogun/features/StringFeatures.h:1063: warning: ignoring return value of ‘size_t fread(void*, size_t, size_t, FILE*)’, declared with attribute warn_unused_result | 00:02 |
| alesis-novik | sonney2k, is ASSERTing ok in the constructor? I can't remember | 00:03 |
| @sonney2k | alesis-novik, you need a default constructor (no arguments) and one with the too sets | 00:03 |
| @sonney2k | alesis-novik, well I do it but raising exceptions is probably not (in the constructor) - but what can one do :/ | 00:03 |
| blackburn | it is better than calling constructor in destructor :D | 00:04 |
| alesis-novik | sonney2k, its just that if you pass the matrix and vector explicitly in the constructor, I need to check if the dims match | 00:04 |
| @sonney2k | alesis-novik, or you check only when you call train() or so | 00:05 |
| blackburn | ich bin mude! | 00:06 |
| alesis-novik | sonney2k, but if we set the mean and cov, we don't really need to call train(). There's actually no point in setting the mean then calling train | 00:07 |
| alesis-novik | because it would just get overwritten | 00:07 |
| @sonney2k | alesis-novik, you are right - so the inverse woudl be computed | 00:08 |
| @sonney2k | then just check it | 00:08 |
| blackburn | sonney2k: pull is here | 00:08 |
| @sonney2k | and then call SG_ERROR with some meaningful message | 00:08 |
| blackburn | pull request, i mean | 00:08 |
| @sonney2k | blackburn, for what? | 00:08 |
| blackburn | preproc | 00:08 |
| alesis-novik | train is doing the same things as the constructor, but estimating the mean and cov from data | 00:08 |
| blackburn | sonney2k: preproc is really needs some stalinizing | 00:10 |
| @sonney2k | alesis-novik, I see - so you only compute the inverse of the cov right? | 00:10 |
| blackburn | there are too much freedom | 00:10 |
| alesis-novik | that and the constant | 00:10 |
| dvevre | sonney2k: just finished making a working parser for online features.. its very basic, makes plenty of assumptions, and i haven't integrated it with any learning technique yet. | 00:10 |
| alesis-novik | because doing that for each datapoint would be slow | 00:10 |
| dvevre | should i mail it? | 00:10 |
| CIA-110 | shogun: Soeren Sonnenburg master * r9a54d0e / src/libshogun/features/StringFeatures.h : fix another fread caused error - http://bit.ly/hXvhKl | 00:11 |
| @sonney2k | dvevre, github pullrequrest please | 00:11 |
| @sonney2k | dvevre, otherwise I definitely lose track | 00:12 |
| dvevre | sonney2k: but i haven't integrated it with shogun objects yet | 00:12 |
| serialhex | wtf! i'm here all day and theres nothing goin on, i go out for like 2 hours and you guys have been in here partying without me!!!! | 00:12 |
| dvevre | i can do that, but i'd like to know if i'm on the right track at least | 00:13 |
| @sonney2k | dvevre, I see | 00:13 |
| blackburn | eh | 00:13 |
| @sonney2k | then email | 00:13 |
| @sonney2k | alesis-novik, isn't computing the inverse cov slow too? | 00:14 |
| alesis-novik | Well there's no use crying over every mistake, we'll just keep on trying 'till we run out of cake | 00:14 |
| CIA-110 | shogun: Sergey Lisitsyn master * rb1b5dfd / src/libshogun/features/StringFeatures.h : Merge remote branch 'upstream/master' - http://bit.ly/hBj5CD | 00:15 |
| alesis-novik | sonney2k, it is, but you need it when computing the PDF (in the simple method I'm using at least) | 00:15 |
| @sonney2k | alesis-novik, OK then | 00:15 |
| * sonney2k really hopes to be replaced by 5 (in words five) mentors | 00:16 | |
| alesis-novik | I'm not sure what Vojtech has in his matlab implementation, maybe this won't be that useful, but it's still a popular distribution, I'm sure there can be some use for it | 00:16 |
| @sonney2k | alesis-novik, I mean when you do combineddistribution - you could have mixtures of guasians | 00:16 |
| @sonney2k | gaussians | 00:16 |
| blackburn | sonney2k: how will you choose mentors? | 00:16 |
| @sonney2k | blackburn, look at the ideas page | 00:17 |
| @sonney2k | they area already known | 00:17 |
| blackburn | sonney2k: eh, I mean how will you choose 5 of 12 | 00:17 |
| @sonney2k | blackburn, cwidmer and mlsec woudl be your man btw | 00:17 |
| serialhex | sonney2k: have you talked about student selection between you & the mentors yet? | 00:18 |
| blackburn | sonney2k: I had about 15-20 emails with cwidmer :) | 00:18 |
| alesis-novik | sonney2k, the idea said that there is a matlab implementation for EM and that the student should base the project on it. I know for a simple version of EM you need to be able to compute the PDF | 00:18 |
| blackburn | he really helped me with coming up with a proposal | 00:18 |
| @sonney2k | serialhex, no meeting will be on monday | 00:18 |
| serialhex | ok, cool | 00:18 |
| alesis-novik | sonney2k, I tried to get the matlab version, but the letter I sent him might have gotten lost somewhere | 00:19 |
| dvevre | sonney2k: mailed.. right now, it's just compile and run, along with that sample data file | 00:19 |
| @sonney2k | serialhex, we have to choose which subjects we want to see developed and then check if we have good enough students for them | 00:20 |
| alesis-novik | well, I still got to use lapack and blas, so that should be useful | 00:20 |
| blackburn | sonney2k: why mlsec will? | 00:20 |
| serialhex | sonney2k: ahh, ok cool | 00:20 |
| @sonney2k | alesis-novik, send again and CC me | 00:20 |
| serialhex | sonney2k: sounds like fun! :D | 00:20 |
| @sonney2k | dvevre, do you know whcih data format vw uses? | 00:22 |
| @sonney2k | (seems to be pretty inefficient format) | 00:23 |
| dvevre | yeah.. i've read about it | 00:23 |
| dvevre | they say its similar to svmlight's | 00:23 |
| @sonney2k | dvevre, it is | 00:23 |
| dvevre | different namespaces and all | 00:24 |
| @sonney2k | but dense features ? | 00:24 |
| dvevre | not sure really | 00:24 |
| @sonney2k | dvevre, when did you start to code this? | 00:24 |
| dvevre | today morning or so | 00:24 |
| @bettyboo | moin | 00:24 |
| dvevre | its around 4 am here :) | 00:24 |
| @sonney2k | dvevre, I see | 00:25 |
| @sonney2k | dvevre, what is not good is the mixing of spaces and tabs everywhere | 00:25 |
| dvevre | hmm.. i thought emacs was upto it | 00:25 |
| @sonney2k | so it is hard to read (not indented) | 00:25 |
| blackburn | another one bites the dust | 00:25 |
| blackburn | 4 am, are you crazy? :D | 00:25 |
| dvevre | strange.. i used emacs with stroustrup indentation | 00:25 |
| dvevre | and it looks great on mine | 00:25 |
| blackburn | *like me* | 00:25 |
| dvevre | :( | 00:25 |
| dvevre | blackburn: a man's got to do what a man's got to do :D | 00:26 |
| blackburn | dvevre: how true it is | 00:26 |
| blackburn | just like any other X like X :D | 00:26 |
| blackburn | tfu | 00:26 |
| blackburn | X what X | 00:27 |
| @sonney2k | dvevre, so the parser starts a new thread and reads the data | 00:27 |
| blackburn | sonney2k: I have to make some graph class, where should I put it? | 00:27 |
| blackburn | base or lib? | 00:27 |
| @sonney2k | blackburn, lib/ | 00:27 |
| dvevre | sonney2k: yeah,, that's the main part | 00:27 |
| blackburn | okay | 00:27 |
| dvevre | separate threads | 00:27 |
| @sonney2k | how do you access vectors from the outside? | 00:28 |
| dvevre | as in? | 00:28 |
| @sonney2k | ahh it is in main() | 00:29 |
| dvevre | example->feature_vector[i] | 00:29 |
| dvevre | :) | 00:29 |
| alesis-novik | found some what appears to be memory leaks | 00:29 |
| @sonney2k | dvevre, I mean the idea is that you start reading to some memory buffer example by example right? | 00:29 |
| dvevre | yeah.. the buffer is like a ring | 00:30 |
| dvevre | i've made the buffer size 100 or something | 00:30 |
| dvevre | the aim is to not incur any memory shortage | 00:30 |
| @sonney2k | some constant that one should be able to specify yes | 00:30 |
| dvevre | as only those many examples are loaded in the memory | 00:30 |
| dvevre | dvevre: yes, yes.. it can be changed later.. | 00:30 |
| dvevre | i've just done some hard coding for testing purposes | 00:31 |
| @sonney2k | sure | 00:31 |
| dvevre | vw uses a larger buffer | 00:31 |
| dvevre | and dynamically allocates more memory for the buffer if it finds it necessary | 00:31 |
| dvevre | that's a pretty good method | 00:31 |
| @sonney2k | dvevre, well but they should change their data format to get a speedup of factor 2 - much more than their dynamic buffer size will do | 00:33 |
| @sonney2k | still it seems optimized to the max | 00:33 |
| dvevre | sonney2k: i think i'll have to find out more abt their format.. haven't tried to find bottlenecks there yet | 00:34 |
| dvevre | they're tried to make it too flexible | 00:34 |
| dvevre | they've | 00:34 |
| @sonney2k | dvevre, I think you should encapsulate the format a bit | 00:35 |
| @sonney2k | some function that just reads 1 example | 00:35 |
| @sonney2k | then you could have a class that does e.g. binaryfile/asciifile etc | 00:35 |
| dvevre | sonney2k: yeah.. like make the functions format independent as far as possible | 00:35 |
| @sonney2k | and only ever loads example by example | 00:35 |
| @sonney2k | yes | 00:36 |
| dvevre | you're right. but i still am not clear about one thing | 00:36 |
| dvevre | once streamingfeatures is made, what will its relation be with the CFeatures class? | 00:36 |
| @sonney2k | dvevre, I guess you need to assume that the number of features is known and that you have to read that many bytes or so to get the next vector | 00:36 |
| @sonney2k | dvevre, it will derive from CFeatures I think | 00:37 |
| dvevre | sonney2k: yes.. in the current program, i have taken the number of features as = the number of features in the first example | 00:37 |
| @sonney2k | dvevre, though there will be no way to say how many examples there are so it will just return 0 or -1 (not sure what is better) | 00:37 |
| dvevre | deriving from CFeatures.... i considered that, but there seem to be so many things which may be different | 00:37 |
| @sonney2k | dvevre, I only see get_num_vectors for which you just have to return -1 | 00:39 |
| dvevre | you're right.. i think i was confusing with dotfeatures and simplefeatures which have methods like get_feature_matrix | 00:42 |
| @sonney2k | dvevre, yes no way to do this with them | 00:42 |
| @sonney2k | your code needs quite some polishing though... hard to read currently | 00:43 |
| @sonney2k | btw, what does john do when his examples have a different number of features? | 00:43 |
| dvevre | he initializes them to a default | 00:44 |
| dvevre | the format specifies the default as zero or one | 00:44 |
| @sonney2k | dvevre, but they are not dense right? | 00:44 |
| @sonney2k | (that would be wasteful) | 00:44 |
| dvevre | no, i don't think they are | 00:45 |
| @sonney2k | do I understand the code right that all features of a block have to be processed before you read the next chunk? | 00:45 |
| dvevre | by block do you mean the buffer as a whole? | 00:46 |
| @sonney2k | yes | 00:47 |
| dvevre | currently, yes. | 00:47 |
| dvevre | i can make this dynamic also | 00:48 |
| dvevre | like make a note of indices which have been processed, and read into those memory locations | 00:48 |
| @sonney2k | dvevre, I think you can assume that they are sequentially processed | 00:49 |
| @sonney2k | so just rotating (or saving one index) is sufficient | 00:49 |
| @sonney2k | you should problably program that data structure independently in lib/ | 00:50 |
| dvevre | sonney2k: yeah.. instead of having to write those unnecessary increment functions and so many other things in the parser code | 00:50 |
| dvevre | some kind of io_buf data structure used in vw | 00:51 |
| dvevre | the code would be much cleaner | 00:51 |
| @sonney2k | just steal that :) | 00:51 |
| @sonney2k | err borrow | 00:51 |
| dvevre | sonney2k: oops! :D | 00:51 |
| dvevre | i think i will. i just wanted to start from scratch without borrowing actual code from vw initially :) | 00:52 |
| @sonney2k | but ok the principle is clear - you can simplify the code drastically when you use other code (we need some line reader function for all the datatypes that need support) and then thes snake (rolling thing) data structure | 00:53 |
| @sonney2k | ok | 00:53 |
| dvevre | and it would have taken too long to get an implementation that could directly be merged into the repo.. i thought i'll send this tiny thing | 00:53 |
| dvevre | sonney2k: yeah.. that's the proper way to do it | 00:53 |
| @sonney2k | dvevre, I guess it is better to have some sleep now | 00:53 |
| dvevre | i guess.. i've got some things to show john as well | 00:54 |
| dvevre | so have to start working on that soon... | 00:54 |
| blackburn | ubernachten is good! | 00:54 |
| blackburn | ist gut :D | 00:54 |
| @sonney2k | not sure if you have time in the next few days to program - but if so try to do it within shogun and if possible re-use some code. | 00:54 |
| @sonney2k | I definitely have to sleep now | 00:55 |
| blackburn | me too but can't ;) | 00:55 |
| dvevre | good night! | 00:55 |
| @sonney2k | blackburn, daway daway... | 00:55 |
| dvevre | blackburn: of all people you can afford to | 00:55 |
| @sonney2k | blackburn, dobrui notsche or so | 00:56 |
| serialhex | so am i the only person in here who only speeks one language? | 00:56 |
| blackburn | sonney2k: spokoynoy nochi | 00:56 |
| serialhex | (not counting programming laguages of course) | 00:56 |
| @sonney2k | prekrasnoe notsche | 00:56 |
| blackburn | sonney2k: :D | 00:56 |
| @sonney2k | anyway | 00:56 |
| * serialhex laments my amerikan ejumakatun | 00:57 | |
| blackburn | sonney2k: see you | 00:57 |
| serialhex | nite blackburn | 00:57 |
| dvevre | serialhex: now you're speaking something else too! | 00:57 |
| @bettyboo | nite | 00:57 |
| @sonney2k | cu tomorrow - and guys sleep is important. dvevre if you can afford it program some more (tommorrow!) | 00:57 |
| @sonney2k | l8r | 00:57 |
| serialhex | yep, dat b propa amerikun engrish | 00:57 |
| blackburn | serialhex: I don't :D | 00:57 |
| dvevre | sonney2k: i have to! i'll try to send you another version tomorrow | 00:58 |
| serialhex | riiite blackburn... | 00:58 |
| blackburn | serialhex: nachinaem urok russkogo yazyka | 00:58 |
| * serialhex bookmarks google translate | 00:58 | |
| blackburn | интересно сможешь ли ты понимать меня с помощью этой штуки | 00:59 |
| alesis-novik | Ah yes, every night, only in #shogun | 00:59 |
| serialhex | blackburn: the first thing didnt come up with anything intelligible in google trans. | 00:59 |
| blackburn | serialhex: what about second in cyrillic? | 00:59 |
| serialhex | yes, "wondering if you can understand me with this thing" | 01:00 |
| blackburn | hehe | 01:00 |
| @bettyboo | ;D | 01:00 |
| * serialhex sereously lamenting his amerikun edukatun | 01:00 | |
| serialhex | what did the first say? | 01:00 |
| blackburn | serialhex: начинаем урок русского языка | 01:00 |
| blackburn | serialhex: this time it will translate :) | 01:01 |
| serialhex | yes it did "begin the lesson of russian language" | 01:01 |
| serialhex | not exactly perfect but intelligible :P | 01:01 |
| dvevre | someone should make bettyboo translation-capable :) | 01:02 |
| @bettyboo | dvevre: python_modular? | 01:02 |
| blackburn | bettyboo: иди отсюда на хрен | 01:02 |
| @bettyboo | blackburn: can you explain more? i don't get it. | 01:02 |
| blackburn | oh, you don't :D | 01:02 |
| dvevre | hhehe | 01:02 |
| * serialhex is listening to rammstein and reading russian :P | 01:02 | |
| blackburn | links | 01:02 |
| blackburn | links zu drei vier | 01:03 |
| blackburn | :D | 01:03 |
| serialhex | blackburn: "Go here to horseradish"? | 01:03 |
| alesis-novik | :D | 01:03 |
| blackburn | serialhex: not sure I can explain it :D | 01:03 |
| alesis-novik | Ah the little things about the language | 01:03 |
| * serialhex shakes his head in confusion | 01:03 | |
| blackburn | serialhex: we have a PLENTY of dirty words :D | 01:04 |
| alesis-novik | It's a mild form of GTFO | 01:04 |
| serialhex | oh, that's the most fun thing to learn in a new language!!! | 01:04 |
| serialhex | gtfo: get the fuck out? | 01:04 |
| blackburn | serialhex: some things like пиздоблядская мудопроёбина google translate will not translate :D | 01:04 |
| @bettyboo | blackburn, he? | 01:04 |
| serialhex | yeah... is it censoring it for me? or is there nothing like it in english? | 01:05 |
| blackburn | serialhex: I don't know how to explain it in russian, and no way in english ;) | 01:06 |
| alesis-novik | blackburn, that was interesting... | 01:06 |
| serialhex | hmm... ok | 01:06 |
| blackburn | alesis-novik: heheheh | 01:06 |
| alesis-novik | I know one, not really dirty, but | 01:06 |
| * serialhex thinks blackburn was cussing at him in russian :D | 01:07 | |
| alesis-novik | Alkan buhailovich zalivaila | 01:07 |
| blackburn | ahahahaha | 01:07 |
| serialhex | yeah, that comes up with nothing :( | 01:07 |
| blackburn | serialhex: that two word combo was something like ''cuntwhore fuckedpuke" :D | 01:08 |
| serialhex | niiiiice!!! | 01:08 |
| alesis-novik | But it sounds much nicer in Russian | 01:08 |
| serialhex | i'm sure it does | 01:08 |
| blackburn | we are able to construct 3-4-5 words in one super-dirty word ;) | 01:09 |
| serialhex | i mean, 'dhumkoph' (i probably mis-spelled it) translates from german to something like 'stupid-head' but is quite severe in german :D | 01:09 |
| blackburn | dummkopf, as I remember | 01:09 |
| serialhex | oooh!!! so it's like a meta-cussing language :D | 01:09 |
| serialhex | yeah that's probably it blackburn! | 01:10 |
| alesis-novik | There are different language rule subsets for different people | 01:10 |
| blackburn | serialhex: the most difficult one will be to spell my last name properly ;) | 01:10 |
| alesis-novik | For example builders have to insert a cuss word ever 1 or 2 words. | 01:10 |
| blackburn | aha, builders often say something like тащи эту поебень сюда дурак блять | 01:11 |
| serialhex | Drop this stupid fucking poeben here | 01:11 |
| serialhex | i dont know what a 'poeben' is, but i get the idea :D | 01:12 |
| blackburn | serialhex: may be 'fuckery' | 01:12 |
| blackburn | often cuss words are used to describe some object | 01:12 |
| alesis-novik | I wonder what would people who would read this log think :D | 01:12 |
| serialhex | 'drop this stupid fucking fuckery here'? i've heard stuff like that before | 01:13 |
| blackburn | serialhex: aha, sth like that | 01:13 |
| blackburn | alesis-novik: will think something not good ;) | 01:13 |
| serialhex | idk alesis-novik :D | 01:13 |
| blackburn | hahah found some man with equal last name | 01:15 |
| blackburn | http://www.youtube.com/watch?v=BLGWnaYKrdE&feature=player_detailpage#t=22s | 01:16 |
| blackburn | almost right spelling! | 01:16 |
| blackburn | or pronouncing | 01:17 |
| blackburn | pronouncing is better word for what I mean, right? | 01:18 |
| alesis-novik | right, cleared up the changes sonney2k asked, time for a snack | 01:18 |
| blackburn | snack? | 01:18 |
| blackburn | thought you have 0.20AM in uk :) | 01:19 |
| alesis-novik | You're right, time for dinner | 01:19 |
| blackburn | тоже не любитель здорового режима сна?) | 01:19 |
| dvevre | hmm.. wonder what i'd call my meal now | 01:19 |
| alesis-novik | I'm fully nocturnal now | 01:20 |
| alesis-novik | I still sleep ~8hours, just not at night :) | 01:20 |
| serialhex | blackburn: pronunciation | 01:20 |
| blackburn | serialhex: ima dummkopf rusian ;) | 01:21 |
| serialhex | blackburn: it's ok, there are a lot of americans who wouldnt know how to say that either... fortunately you are much more intelligent than those people & they would proably never use a word like 'pronunciation' in a sentence anyway :D | 01:23 |
| blackburn | serialhex: do they use word 'shit' instead of it? :D | 01:23 |
| serialhex | alesis-novik: i'm on prettymuch the same schedule, i dont think i've gone to sleep before 1am in over a week | 01:23 |
| serialhex | blackburn: probably soething like that | 01:24 |
| alesis-novik | serialhex, 1am? Think 8 am :D Lectures are done so I spend my time on GSoC/revising for exams | 01:25 |
| serialhex | they'd probably say something to the effect of (transliterating): "it be kinda da rite spellin, or sum shit lik dat" | 01:25 |
| blackburn | ahahah | 01:25 |
| blackburn | kinda da rite spellin | 01:25 |
| blackburn | oh | 01:25 |
| serialhex | alesis-novik: i went to sleep at 6:30am on monday, 4:30 tuesday & so on... | 01:26 |
| serialhex | blackburn: yeah, and it's acceptable to speek like that quite often :( | 01:26 |
| alesis-novik | serialhex, trying to become normal again? :) | 01:26 |
| @bettyboo | HA | 01:26 |
| serialhex | alesis-novik: not really, but seeing as how most of you are in time-zones where it's stoopit-o-clock in the am when it's dinner time here i'm trying to be awake & on when other people are on | 01:27 |
| serialhex | btw, what time zone are you in alesis-novik? | 01:28 |
| serialhex | it's like 3:30am where blackburn is... | 01:28 |
| blackburn | it is | 01:28 |
| alesis-novik | 0 | 01:28 |
| alesis-novik | or rather +1 due to DST | 01:28 |
| serialhex | 0? ooh, spooky! | 01:28 |
| * serialhex adds to his spiffy clocks | 01:29 | |
| blackburn | ok, guys, will sleep a little :) | 01:30 |
| blackburn | have to wake up at 8-30 | 01:30 |
| serialhex | nite blackburn!! | 01:30 |
| blackburn | see you | 01:30 |
| alesis-novik | High chance I'll be here :) | 01:30 |
| alesis-novik | night | 01:30 |
| blackburn | alesis-novik: better go sleep ;) | 01:31 |
| @bettyboo | HA | 01:31 |
| -!- blackburn [~qdrgsm@188.168.3.124] has quit [Quit: Leaving.] | 01:31 | |
| serialhex | alesis-novik: so london would work? | 01:31 |
| alesis-novik | serialhex, same timezone, yes | 01:31 |
| serialhex | ok, cool | 01:31 |
| serialhex | 12:30am | 01:32 |
| alesis-novik | studying in Scotlan | 01:32 |
| serialhex | cool | 01:32 |
| serialhex | and you're from lithuania if i recall?? (i could be wrong, i saw it while i was skimming the chat log) | 01:32 |
| alesis-novik | yeah, did my undergrad there, now doing a masters | 01:35 |
| serialhex | cool cool | 01:35 |
| serialhex | where at & what in?? | 01:35 |
| alesis-novik | did or doing? :) | 01:36 |
| @bettyboo | ;D | 01:36 |
| serialhex | doing | 01:36 |
| alesis-novik | AI at University of Edinburgh | 01:36 |
| serialhex | cool | 01:37 |
| alesis-novik | What do you study? | 01:37 |
| serialhex | i'm getting an associates in Math & Computer Science (programming) right now (2 seperate degrees - 2 year degree) and will probably get a bachelors in game programming (2 more years) if i cant find anything cool in AI before then | 01:41 |
| alesis-novik | sounds cool, yet complicated | 01:44 |
| -!- ameerkat [HydraIRC@149-149-169-153-58.nat.asu.edu] has joined #shogun | 01:44 | |
| serialhex | it's the only way i could find to get the math i wanted & get a head start on AI programming | 01:45 |
| serialhex | there's almost no math taught in CS courses here | 01:46 |
| alesis-novik | My whole 1st year of undergrad was mostly math | 01:47 |
| serialhex | what did you take as an undergrad? | 01:48 |
| alesis-novik | Informatics | 01:49 |
| serialhex | cool | 01:49 |
| ameerkat | I go to Arizona State, for computer science they make you take cal 1-3 for engineers, discrete math, and linear algebra | 01:49 |
| serialhex | ameerkat: cool, yeah i only have to take calc 1 & 2 and maybe statistics | 01:50 |
| ameerkat | ah okay | 01:50 |
| serialhex | though since it's a 2 year degree i guess they're not expecting much from me :P | 01:50 |
| alesis-novik | stats and prob theory seems most important in AI (and knowing how to compute derivatives and integrals) | 01:51 |
| ameerkat | I see | 01:51 |
| serialhex | alesis-novik: yeah, thats why i'm also going for math, so i can take all that fun stuff :D | 01:51 |
| ameerkat | yeah, I decided I wanted more math so I added math as a 2nd major and I'm taking probability now because of AI | 01:51 |
| ameerkat | number theory is a lot of fun too but it has less direct applications than probability, unless you love cryptography | 01:52 |
| serialhex | cool, i'll have to remember that ameerkat | 01:53 |
| alesis-novik | cryptography is fun, but I'm more of a fan of coding theory | 01:54 |
| serialhex | coding theory?? | 01:55 |
| alesis-novik | data compression and and error correction. The second part is really interesting personally. | 01:57 |
| ameerkat | I've never heard of coding theory as a whole before | 01:57 |
| ameerkat | I've never seen any classes on it anyway I dont think they teach anything about it here | 01:58 |
| alesis-novik | It might be in a bundle with information theory | 01:58 |
| ameerkat | data compression sounds fun though, I've read some interesting things about it | 01:58 |
| ameerkat | ah yeah most likely | 01:58 |
| alesis-novik | modern data compression is all about probability distributions | 01:59 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Quit: Page closed] | 02:00 | |
| serialhex | very interesting! | 02:05 |
| alesis-novik | actually, all compression is about probability distributions, it's just new methods have interesting ways of using them | 02:07 |
| -!- ameerkat [HydraIRC@149-149-169-153-58.nat.asu.edu] has quit [Ping timeout: 240 seconds] | 02:10 | |
| -!- Guest79208 [d8a57e6e@gateway/web/freenode/ip.216.165.126.110] has quit [Quit: Page closed] | 02:15 | |
| alesis-novik | Going away for a while, I'll be Bach later | 02:25 |
| -!- alesis-novik [~alesis@188.74.87.84] has quit [Quit: I'll be Bach] | 02:25 | |
| -!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 03:47 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 06:07 | |
| alesis-novik | anyone around here? | 06:58 |
| serialhex | nope | 07:02 |
| serialhex | we've all gone to bed alesis-novik | 07:02 |
| serialhex | ffs dude... the sun's going to rise there soon!!! | 07:03 |
| alesis-novik | serialhex, us vampires don't like the sun. Also informatics students, so take your pick on what I am | 07:03 |
| serialhex | :P lol | 07:04 |
| serialhex | yeah i'm wrting a story, and so i'll probably be up for quite while longer | 07:04 |
| ameerkat | where do you guys live | 07:05 |
| ameerkat | its only 10pm where I live | 07:06 |
| serialhex | he's in scotland, and i'm in florida | 07:06 |
| serialhex | 1am here & 6am there | 07:06 |
| ameerkat | oh nice my uncle lives in glasgow, but I live all the way over in arizona | 07:07 |
| serialhex | cool cool | 07:07 |
| serialhex | i love this comic!! http://xkcd.com/ | 07:12 |
| alesis-novik | did you try the 3d version on april 1st? | 07:15 |
| serialhex | yeah, that was cool! | 07:17 |
| alesis-novik | 5-min drive to historic pripyiat made my day | 07:17 |
| serialhex | :P i had to look that one up, but once i saw i laughed! | 07:18 |
| alesis-novik | They do have tours there now though. I'm thinking of someday going on it :D | 07:19 |
| serialhex | interesting! | 07:22 |
| ameerkat | this is one of my favorite xkcd's http://xkcd.com/399/ | 07:22 |
| serialhex | i heard theres some cool fugus that actually feeds off of the radiation given off at the chernobyl site | 07:23 |
| serialhex | lol, i went through all of them when i found it in like 3 nights... :D | 07:23 |
| -!- siddharth [~siddharth@117.211.88.150] has joined #shogun | 07:23 | |
| serialhex | good morning siddharth! | 07:24 |
| @bettyboo | hiho | 07:24 |
| siddharth | morning | 07:24 |
| siddharth | wassup | 07:24 |
| serialhex | chillin like a villian, you? | 07:24 |
| siddharth | haha... | 07:24 |
| siddharth | u dont sleep ? | 07:24 |
| alesis-novik | Morning | 07:25 |
| serialhex | it's only ~1:30am, not time for sleep atill for hours!! | 07:25 |
| alesis-novik | Sleep? What is this sleep? | 07:25 |
| serialhex | :P lol | 07:25 |
| siddharth | ic | 07:25 |
| serialhex | what time is it there siddharth?? | 07:25 |
| siddharth | its around 11 am | 07:25 |
| serialhex | damn, what time zone? | 07:26 |
| siddharth | india | 07:26 |
| siddharth | + 5 30 hrs | 07:26 |
| alesis-novik | I think I butchered PCACut... | 07:27 |
| alesis-novik | It now has 3 conditions of different priorities of how many principal components to select | 07:28 |
| siddharth | like? | 07:29 |
| siddharth | max eigen value? | 07:29 |
| siddharth | what the lower limit on the eigen values? | 07:29 |
| alesis-novik | max eigenvalue? No, it has a threshold (min eigenvalue), static number of components to select or percentage of variance explained | 07:30 |
| alesis-novik | I wonder if I can trust lapack to return the eigenvalues sorted... | 07:31 |
| siddharth | earlier only min eigen value was there? | 07:32 |
| alesis-novik | yes | 07:32 |
| siddharth | well in normal PCA this is the sufficient condition | 07:33 |
| alesis-novik | siddharth, I think percentage of variance explained is also a very popular condition | 07:33 |
| alesis-novik | serialhex, the tour has some strict rules :D | 07:34 |
| alesis-novik | During the visit to the exclusion zone it is totally prohibited to: | 07:34 |
| alesis-novik | · Carry any kind of weapons; | 07:34 |
| alesis-novik | · Drink liquors or take drugs; | 07:34 |
| alesis-novik | · Have meal and smoke in the open air; | 07:34 |
| alesis-novik | · Touch any structures or vegetation; | 07:34 |
| alesis-novik | · Sit or place photo and video equipment on the ground; | 07:34 |
| alesis-novik | · Take any items outside the zone; | 07:34 |
| alesis-novik | · Violate the dress code (open-type shoes, shorts, trousers, skirts); | 07:34 |
| -!- alesis-novik was kicked from #shogun by bettyboo [flood] | 07:34 | |
| -!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 07:34 | |
| siddharth | haha got kicked | 07:34 |
| @bettyboo | ;D | 07:34 |
| alesis-novik | sorry bettyboo | 07:34 |
| @bettyboo | alesis-novik: bed can wait | 07:34 |
| alesis-novik | YES! | 07:34 |
| alesis-novik | high5 bettyboo | 07:35 |
| @bettyboo | alesis-novik: I thought it is equal to that initialization in classify(CFeatures* data) | 07:35 |
| siddharth | ur comment :P | 07:35 |
| serialhex | wtf?? thats insane! (maybe not, but still kind of crazy) | 07:35 |
| serialhex | and bettyboo is going to rule us all one day :D | 07:35 |
| @bettyboo | serialhex: *yes I will get to while he will not remember all the russian words* :D | 07:35 |
| alesis-novik | serialhex, it's still the zone of alienation. It's patrolled by the military and stuff. | 07:36 |
| alesis-novik | serialhex, also with a very high level of background radiation | 07:36 |
| serialhex | alesis-novik, hmm, interesting... so i guess that kind of makes sense then | 07:37 |
| serialhex | but what? you're expected to float over the ground and starve wheile you are there?? (and do they expect me to have a "radiation party" with vodka and extacy or something?? :P ) | 07:38 |
| alesis-novik | and some of the metals and vegetation have characteristics of being very adsorbent. | 07:38 |
| serialhex | hmm, interesting | 07:38 |
| alesis-novik | That's why the fire-trucks used during the disaster are still there | 07:38 |
| serialhex | wow | 07:38 |
| serialhex | siddharth: is kolkata in your time zone? | 07:39 |
| serialhex | (theres like 8 time zones in india :P ) | 07:39 |
| siddharth | serialhex, yeah | 07:39 |
| serialhex | ok, cool | 07:39 |
| * serialhex has a new time zone in his collection | 07:40 | |
| siddharth | How to add a vector(which is not a feature vector) to a dense feature vector? | 07:43 |
| siddharth | I dont have the index of that vector | 07:44 |
| siddharth | so how can i do it?...anyone | 07:44 |
| * serialhex wishes he could answer siddharths Q | 07:46 | |
| alesis-novik | So what do you want to add to where? :D | 07:47 |
| serialhex | i know how to do normal vectors and stuff... but i get lost when you start talking about 'feature' vectors :-/ | 07:47 |
| siddharth | W is my dense vector....and I want to add another vector to it...there is a add_to_dense_vec() function which take input as the index of that vector and then adds to it.....but here I dont have the index | 07:49 |
| alesis-novik | So how did you get the vector? | 07:49 |
| siddharth | I have a feature vector...I did some computation and now I have a new vector | 07:50 |
| siddharth | basically its matrix multiplication...(m*m) and (m*1) dimension...so we will get a m*1 vector | 07:51 |
| siddharth | this m*1 vector is my new vector | 07:51 |
| -!- alesis-novik [~alesis@188.74.87.84] has left #shogun ["Leaving"] | 07:51 | |
| -!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 07:51 | |
| alesis-novik | And what exactly do you want to do with this new vector? | 07:52 |
| siddharth | add to another dense vector | 07:52 |
| alesis-novik | which is where? | 07:52 |
| siddharth | http://www.shogun-toolbox.org/doc/classshogun_1_1CDotFeatures.html | 07:53 |
| siddharth | see here | 07:53 |
| siddharth | multiplication with a scalar and addition to a dense vector | 07:54 |
| siddharth | 3rd point | 07:54 |
| serialhex | dihydrogen monoxide: the awesomest stuff in the universe! | 07:54 |
| alesis-novik | siddharth, so you have two vectors as float64_t*? | 07:55 |
| siddharth | yeah | 07:55 |
| serialhex | http://www.shogun-toolbox.org/doc/classshogun_1_1CDotFeatures.html#ab87956faaaffa041a1a60b0f01dcb6e6 maybe?? | 07:55 |
| alesis-novik | So you can just add them? What do you want to do with the result though? | 07:55 |
| serialhex | oh, no :P | 07:56 |
| siddharth | I have to add them not take dot product | 07:56 |
| siddharth | http://www.shogun-toolbox.org/doc/classshogun_1_1CDotFeatures.html#a72581fa30d58a83dcc2f10b5e1f51a7a | 07:57 |
| alesis-novik | so just add them? I don't really understand what you need, sorry... | 07:57 |
| siddharth | the function is this...but it needs index of the vector | 07:58 |
| siddharth | just see the function once | 07:58 |
| alesis-novik | siddharth, but you said you have both vectors in basic(?) form (float64_t * and int32_t for length) | 07:58 |
| alesis-novik | The function takes one of the vectors from the matrix and adds it to the other (with additional constant multiplication and stuff) | 07:59 |
| siddharth | yes...this function does that only | 07:59 |
| siddharth | yes both have float64_t form | 08:00 |
| alesis-novik | so why not just add? | 08:00 |
| siddharth | I want to make use of this function | 08:01 |
| siddharth | somehow | 08:01 |
| alesis-novik | why? | 08:01 |
| alesis-novik | From what you're saying I *think* you need CMath::add | 08:02 |
| siddharth | basically it does adding...so i was thinking if we can do something | 08:02 |
| * serialhex wonders why it computers sometimes make simple things exceedingly difficult | 08:02 | |
| siddharth | yeah CMath I know...but this function is cool :P | 08:03 |
| * serialhex rolls his eyes | 08:03 | |
| alesis-novik | but it specifically uses vectors inside the matrix. It wouldn't make sense to have a method that has nothing to do with the object | 08:04 |
| serialhex | because it's cool: the doom of men ages 16-24 :P | 08:04 |
| siddharth | http://www.shogun-toolbox.org/doc/classshogun_1_1CMath.html | 08:04 |
| siddharth | hmm may be I have to use CMath :) | 08:05 |
| alesis-novik | Yeah, just add them :) | 08:05 |
| alesis-novik | Right, Sergej said he will work on KPCA, I'll need something new to do tomorrow (after I correct any problems with the current pull request) | 08:07 |
| alesis-novik | Man, I wish C++ would have a standard of code formatting like Java. Now it takes a while getting used to every new project you work on... | 08:08 |
| alesis-novik | Well, since sonney2k isn't here, I'll go actually get some sleep | 08:10 |
| serialhex | alesis-novik, i dont think many languages have an 'enforced' way to code | 08:10 |
| serialhex | i know python does, but i'm pretty sure most of it is simply a 'reccomended way' to do things | 08:11 |
| serialhex | (i mean most other languages, not python specifically) | 08:11 |
| alesis-novik | serialhex, c++ doesn't even have that | 08:11 |
| serialhex | yeah, cause c++ is older than god! | 08:12 |
| alesis-novik | Java isn't enforced, but it's more or less a standard | 08:12 |
| serialhex | (i know, i asked him once) | 08:12 |
| serialhex | the same way with ruby, i can code almost any way i want, but theres an 'accepted' way to do things that the community kind of enforces | 08:13 |
| alesis-novik | c/c++ is just awesome (I especially like inline assembly options), but I wish everyone would start using one way of formatting. | 08:13 |
| alesis-novik | Personally, I like opening the brackets for "if" "while" "for" on the same line | 08:13 |
| serialhex | personally i like it on the same line too... but i cant STAND having to make sure i have a semicolon at the end of each line! :P | 08:14 |
| siddharth | is there any matrix multiplication function? | 08:15 |
| siddharth | was not able to find in the library | 08:15 |
| serialhex | also i think it's that they dont really teach any particular formatting in school, at least not from what i've seen. my brothers code is all left-aligned, and NO whitespace!!! | 08:15 |
| serialhex | i'm sure there is... | 08:16 |
| alesis-novik | serialhex, that's scary | 08:16 |
| alesis-novik | siddharth, if not it CMath, then blas probably has something | 08:16 |
| serialhex | scary isnt the half of it! i have to re-format his stuff every time i look at it! | 08:16 |
| serialhex | he claimes he like to "be able to view his whole program on the screen at once, and whitespace would destroy that" but WTF is he going to do when his CODE is 100+ lines?? | 08:17 |
| serialhex | have a 16,000x12,000 monitor resolution?? :P | 08:18 |
| @bettyboo | ;> | 08:18 |
| alesis-novik | Personally, I think the only way to learn this is from experience | 08:18 |
| alesis-novik | I didin't properly format code in school either. Then a bigger program that I can't write in one sitting came along and I couldn't find where stuff was the next day | 08:19 |
| alesis-novik | Then I thought 1) smart variable/function names 2) smart formatting | 08:19 |
| serialhex | yeah, i've been reading about coding a LOT longer than i've actually been doing anything resembling a program :P so i've got all these ideas of how things should look & flow in my head while i'm learning everything | 08:21 |
| serialhex | shogun dosnt have a matrix class?? wtf??? | 08:22 |
| serialhex | the only thing that even references 'matrix': http://www.shogun-toolbox.org/doc/classshogun_1_1CPlifMatrix.html | 08:23 |
| alesis-novik | Isn't really needed | 08:24 |
| alesis-novik | lapack and blas use pointer row_size col_size | 08:24 |
| alesis-novik | Other than that, you have SimpleFeatures<class T> | 08:25 |
| serialhex | hmm.... | 08:25 |
| alesis-novik | as far as I know, shogun doesn't even have a vector class | 08:25 |
| * serialhex thinks it's going to make his job of making ruby bindings MUCH harder... | 08:26 | |
| alesis-novik | enough distractions. sleep now. see anyone who is here in ~8 hours later | 08:28 |
| serialhex | alright, 'nite! | 08:29 |
| alesis-novik | here's hoping for comments on the pull when I wake up :D | 08:29 |
| alesis-novik | Night | 08:29 |
| serialhex | *the sun should be rising where alesis-novik is right now :D * | 08:29 |
| @sonney2k | Re | 08:39 |
| serialhex | re?? | 08:41 |
| @sonney2k | ==back | 08:41 |
| @mlsec | Hey sonney2k | 08:41 |
| @sonney2k | I saw you all partying already | 08:41 |
| serialhex | ahh, i see. welcome back oh fearless leader :D | 08:41 |
| * sonney2k notices mlsec is awake too | 08:41 | |
| @mlsec | Gerarde mal so | 08:42 |
| serialhex | yeah... we try and keep it going in here... we're teaching bettyboo well :P | 08:42 |
| @bettyboo | serialhex: optimize for auROC not accuracy and adjust the bias manually to the fpr-threshold you want | 08:42 |
| @sonney2k | my words! | 08:42 |
| serialhex | yeah bettyboo, i dont know how TF to do that, sorry, no can do :P | 08:43 |
| @bettyboo | serialhex: doesn't sound right - maybe only for certain parameter settings? | 08:43 |
| serialhex | definitely! | 08:43 |
| serialhex | nite all! | 09:00 |
| * serialhex goes to sleep | 09:00 | |
| CIA-110 | shogun: Soeren Sonnenburg master * r53e8143 / (7 files in 6 dirs): fix the data related symlinks such that examples and tests run through - http://bit.ly/fpTP2P | 09:28 |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Quit: HydraIRC -> http://www.hydrairc.com <- Nine out of ten l33t h4x0rz prefer it] | 09:32 | |
| @sonney2k | serialhex, sleep well :) | 09:47 |
| @mlsec | time to get to office | 10:19 |
| @mlsec | what do you think, bettyboo? | 10:19 |
| @bettyboo | mlsec: so what data is in your examples/undocumented/data dir? mine is empty | 10:19 |
| @mlsec | mine says: "go to office now" ;) | 10:20 |
| -!- mlsec [~mlsec@bane.ml.tu-berlin.de] has left #shogun [] | 10:24 | |
| -!- mlsec [~mlsec@bane.ml.tu-berlin.de] has joined #shogun | 11:02 | |
| -!- mode/#shogun [+o mlsec] by ChanServ | 11:02 | |
| @mlsec | whois shogun-toolbox? | 11:06 |
| @sonney2k | mlsec, the logger | 11:06 |
| @mlsec | keymachine.de, that's a provider then | 11:07 |
| @sonney2k | well running on shogun-toolbox.org | 11:07 |
| @mlsec | ok | 11:07 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 11:47 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [] | 12:01 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 12:02 | |
| -!- blackburn [~qdrgsm@188.168.4.217] has joined #shogun | 12:12 | |
| -!- blackburn [~qdrgsm@188.168.4.217] has quit [Client Quit] | 12:16 | |
| @sonney2k | blackburn, you are up late ;-) | 12:16 |
| * sonney2k hmmhh cannot take the slightest criticism ;-) | 12:17 | |
| siddharth | sonney2k, have you decided the 5 slots? | 12:18 |
| * sonney2k counts the number of students in this channel | 12:18 | |
| siddharth | 5 :) | 12:18 |
| * sonney2k hmmhhhh 4 only | 12:18 | |
| @sonney2k | siddharth, nope - we will have a meeting on monday | 12:18 |
| @sonney2k | I hope then we can decide. | 12:19 |
| siddharth | I will have to code SGD-QN before that? | 12:19 |
| @sonney2k | Might be that some student(s) applied for multiple projects though and in fact don't want this one. | 12:20 |
| @mlsec | siddharth: what's your first name? | 12:20 |
| siddharth | mlsec, siddharth | 12:20 |
| @mlsec | ah okay ;) | 12:20 |
| @mlsec | that was simple | 12:20 |
| siddharth | full name:siddharth kherada :) | 12:20 |
| dvevre | sonney2k: when you told me to integrate shogun with the parser yesterday, what things about shogun did you have in mind? | 12:22 |
| @sonney2k | dvevre, just a class derived from CFeatures - CStreamingFeatures - that provides the get_next_feature_vector and does i/o in parallel underneath | 12:23 |
| @sonney2k | steal err borrow from vw whenever necessary :D | 12:23 |
| @bettyboo | sonney2k, ;> | 12:23 |
| dvevre | sonney2k: and the underlying features class? | 12:24 |
| dvevre | i mean like we have simplefeatures<float64_t> and <int32_t> and so on | 12:24 |
| @sonney2k | dvevre, CFeatures | 12:24 |
| @sonney2k | ahh I see | 12:24 |
| @sonney2k | like VW - I think sparse features | 12:24 |
| dvevre | so i shouldn't think about reusing classes like simplefeatures, stringfeatures etc? | 12:25 |
| @sonney2k | dvevre well use the same data representation - but no more I would say | 12:26 |
| @sonney2k | hmmhh | 12:26 |
| @sonney2k | difficult question | 12:26 |
| @sonney2k | gtg | 12:26 |
| @sonney2k | l8r | 12:26 |
| dvevre | see you | 12:26 |
| @sonney2k | dvevre, ok was thinking about it | 12:40 |
| @sonney2k | I see 2 optieons currently | 12:40 |
| @sonney2k | 1) implement CStreamingFeatures (derived from CFeatures) and then CSimpleStreamingFeatures / CSparseStreamingFeatures / CStringStreamingFeatures deriving from them | 12:41 |
| @sonney2k | they should use similar things like CSimpleFeatures etc | 12:42 |
| @sonney2k | but otherwise everyone does his own stuff | 12:43 |
| @sonney2k | option 2) | 12:43 |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 12:44 | |
| @sonney2k | you do these things in option 1) and later on derive CSimpleFeatures etc from them | 12:44 |
| @sonney2k | dvevre, other ideas? | 12:45 |
| dvevre | sonney2k: was away.. i like the second option. if an online algorithm works on SimpleFeatures, it should be able to take both CSimpleStreamingFeatures and CSimpleFeatures as input, I think | 13:38 |
| dvevre | and use the get_next_feature_vector method from both classes | 13:38 |
| dvevre | so i think i would derive CSimpleFeatures from CSimpleStreamingFeatures | 13:38 |
| dvevre | and the algorithm would take CSimpleStreamingFeatures as an input argument | 13:39 |
| * mlsec back from lunch with sonne | 13:59 | |
| @sonney2k | dvevre, for now it does not really matter though | 13:59 |
| dvevre | sonney2k: yeah.. i'll just make that class right now, assuming a feature vector of float64_ts | 14:00 |
| @sonney2k | just do it in a compatible way, e.g. vector == pointer to it and length | 14:00 |
| @sonney2k | ok | 14:00 |
| -!- blackburn [~qdrgsm@188.168.5.180] has joined #shogun | 14:02 | |
| -!- yayo3 [~jakub@195.113.242.138] has joined #shogun | 14:23 | |
| blackburn | sonney2k: examples/python_modular/features_string_char_compressed_modular.py fails with [ERROR] failed to read compression number of vectors | 14:25 |
| @sonney2k | blackburn, I am currently debugging heikos changes | 14:26 |
| blackburn | sonney2k: okay, just reporting :) | 14:26 |
| @sonney2k | thx | 14:27 |
| blackburn | oh, seems you have some collision with GMMs :) | 14:28 |
| -!- yayo3 [~jakub@195.113.242.138] has quit [Quit: Lost terminal] | 14:31 | |
| blackburn | sonney2k: is it ok I have no work on shogun today? just came from university and going home (~100km :) today | 14:31 |
| blackburn | anyway, just informing ;) | 14:32 |
| CIA-110 | shogun: Soeren Sonnenburg master * r11a53be / src/libshogun/features/StringFeatures.h : | 14:32 |
| CIA-110 | shogun: fix a number of issues related to the split of num_vectors / | 14:32 |
| CIA-110 | shogun: num_vectors_total / max_string_length / max_string_length_total - http://bit.ly/gqkO5g | 14:32 |
| @sonney2k | blackburn, you absolutely must have finished your 3 month gsoc project starting in may until the end of next week (april 22!). no lame excuses please work all night and day ;) | 14:33 |
| blackburn | sonney2k: oh, I really like your sense of humour :D | 14:34 |
| blackburn | see you | 15:02 |
| -!- blackburn [~qdrgsm@188.168.5.180] has quit [Quit: Leaving.] | 15:02 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Ping timeout: 252 seconds] | 15:24 | |
| * mlsec is getting extremely tired | 15:24 | |
| -!- yayo3 [~jakub@195.113.242.138] has joined #shogun | 15:31 | |
| yayo3 | sonney2k are you here? I need an architectural advice :) | 15:37 |
| @sonney2k | yayo3, I am here | 15:38 |
| yayo3 | so I'm about to start working on the kernel logistic regression, but I'm not sure where should I put it | 15:39 |
| yayo3 | it's a Kernel Machine, basically, but I'm not sure the KernelMachine class is the right place | 15:40 |
| yayo3 | as some of the methods don't make much sense, work with support points and stuff | 15:40 |
| yayo3 | or maybe I should ask Christian? it was his idea? | 15:40 |
| yayo3 | also, there are no classes that provide general optimization methods, right? or did I miss something again? :) | 15:42 |
| @sonney2k | kernel machine is the right place probably | 15:43 |
| @sonney2k | no there is no optimization framwork if you mean that (was a GSoC idea) | 15:44 |
| yayo3 | alright | 15:56 |
| CIA-110 | shogun: Soeren Sonnenburg master * rd5998c5 / src/libshogun/features/StringFeatures.h : fix the fread size checks (that where causing breakage of various i/o tests) - http://bit.ly/dKd8Fo | 15:56 |
| yayo3 | there's no iteratively re-weighted least squares to reuse, right? :) | 15:56 |
| yayo3 | (that's just me daydreaming) | 15:57 |
| @sonney2k | heh no | 16:02 |
| yayo3 | ok and what should I do with methods like "get_support_vector" when there is no support vector? call SG_ERROR? switch them to being private? | 16:04 |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 16:06 | |
| yayo3 | anyway I have to run. see ya later | 16:07 |
| -!- yayo3 [~jakub@195.113.242.138] has quit [Quit: leaving] | 16:08 | |
| @mlsec | bettyboo, it is getting quiet here? | 16:23 |
| @bettyboo | mlsec: Man, I wish C++ would have a standard of code formatting like Java. Now it takes a while getting used to every new project you work on... | 16:23 |
| @bettyboo | mlsec, sure!?! | 16:23 |
| @mlsec | bettyboo: you are funny! | 16:24 |
| @bettyboo | mlsec: where are you from? | 16:24 |
| @mlsec | bettyboo: i am from germany and you? | 16:24 |
| @bettyboo | mlsec: yep... just added a widget for germany-time and it's 3:33am there :P ok, maybe i'll just e-mail my Q | 16:24 |
| @mlsec | it's not 3:33am in germany btw | 16:25 |
| @mlsec | bettyboo: maybe you should check your clock. you are off by 11 hours | 16:25 |
| @bettyboo | mlsec: so is your question sufficiently answered (you were already offline when I saw your question) | 16:25 |
| @mlsec | bettyboo: not really, but never mind. | 16:26 |
| @bettyboo | mlsec: not really, but seeing as how most of you are in time-zones where it's stoopit-o-clock in the am when it's dinner time here i'm trying to be awake & on when other people are on | 16:27 |
| @mlsec | bettyboo: stoopid-o-clock! that's funny- | 16:27 |
| @bettyboo | mlsec: and that's it | 16:27 |
| CIA-110 | shogun: Soeren Sonnenburg master * r688b11c / examples/undocumented/python_modular/features_string_char_compressed_modular.py : Add example for snappy compression - http://bit.ly/eYHAko | 16:28 |
| CIA-110 | shogun: Soeren Sonnenburg master * rfff70f5 / (11 files): fix the get_name() functions in preprocs causing serialization to fail - http://bit.ly/g3tVJO | 16:28 |
| * mlsec osx (nx) -> linux (virtualbox) -> windows | 16:41 | |
| @mlsec | bettyboo: have you worked with pthreads? i have a problem | 16:52 |
| @bettyboo | mlsec: Hi, I'm almost a newbie in c++ and just have a little question | 16:52 |
| @mlsec | bettyboo: and what about pthreads? i am stuck with pthread_create | 16:52 |
| @bettyboo | mlsec: she convinced the committee with her profound knowledge of c++ coding and machine learning algorithms | 16:52 |
| @mlsec | bettyboo: the problem is that no thread is created, yet i get a positive feedback from the call | 16:52 |
| @bettyboo | mlsec: well do but warn us that it is in Russian. I admittedly had Russian at school (for 11 years or so) but forgot everything - lets see if I can make sense of some of the words :) | 16:52 |
| @mlsec | bettyboo: okay thanks. i will try Russian then | 16:53 |
| @bettyboo | mlsec: That or Russian (majority rules) :D | 16:53 |
| @mlsec | bettyboo: that's what i said, you nasenbaer | 16:53 |
| @bettyboo | mlsec: what will you do in case of 2 slots? :) | 16:53 |
| @mlsec | bettyboo: okay. | 16:55 |
| @bettyboo | mlsec: okay :) | 16:55 |
| @mlsec | bettyboo: i am confused. still stuck with pthread_create | 17:08 |
| @bettyboo | mlsec: I'm still confused about what you suggest doing with init(mean, cov, dim) | 17:08 |
| @mlsec | bettyboo: so we are both confused now | 17:08 |
| @bettyboo | mlsec: I only know things like, kak tebja sowut (looks weird in these latin characters though) | 17:08 |
| -!- heiko [~heiko@infole-06.uni-duisburg.de] has joined #shogun | 17:19 | |
| heiko | hi there | 17:20 |
| @mlsec | welcome back heiko | 17:20 |
| @bettyboo | ho | 17:20 |
| @sonney2k | hi heiko | 17:34 |
| @mlsec | hi sonney2k | 17:34 |
| @mlsec | hi heiko | 17:34 |
| @sonney2k | I hope I fixed the issues... | 17:35 |
| @sonney2k | but well hard to find without test suite available | 17:35 |
| @sonney2k | heiko, do git submodular update --init to get the data git repo | 17:35 |
| @sonney2k | (from shogun git root) | 17:35 |
| @sonney2k | then you can now run make check-examples | 17:36 |
| @sonney2k | make tests | 17:36 |
| heiko | ok | 17:36 |
| heiko | i will check it out | 17:36 |
| @sonney2k | heiko, have a look at the git log / you missed a couple of cases (fread checks were broken, num_vectors_total was unset ...) | 17:40 |
| heiko | sonney2k, oh yes sorry for the fread stuff, guess i was a bit too tired yesterday evening | 17:45 |
| @sonney2k | heiko, yeah next time better not doing it / respectively run the tests before and after | 17:46 |
| heiko | good idea ;) | 17:46 |
| @bettyboo | heiko, ;> | 17:46 |
| @sonney2k | heiko, you could give dotfeatures ago (then that's it about features) | 17:57 |
| @sonney2k | or try more in the framework direction | 17:57 |
| heiko | i think i will go for dotfeatures, because now i am a bit into it | 17:58 |
| heiko | with the framework, we have to discuss a bit more i think | 17:59 |
| @sonney2k | heiko, for framwork first parameter registration for e.g. gaussian kernels :) | 18:00 |
| @sonney2k | then more | 18:00 |
| heiko | so this worked like this: every SGObject class may register parameters to an extra list for cross-val | 18:01 |
| @sonney2k | heiko, exactly | 18:01 |
| heiko | and the gaussian kernel then registers its width and etc | 18:02 |
| @sonney2k | we probably need extra info like valid range or so | 18:02 |
| @sonney2k | yes | 18:02 |
| heiko | yes and some kind of type | 18:02 |
| @sonney2k | type is already there | 18:02 |
| heiko | i mean how fine refinement steps are possible etc | 18:02 |
| @sonney2k | features are no model selection parameters - as we defined | 18:02 |
| heiko | yes indeed | 18:03 |
| heiko | and min/max would be nice? | 18:03 |
| @sonney2k | min /max/default | 18:03 |
| heiko | default=start ? | 18:04 |
| @sonney2k | some reasonable value | 18:05 |
| heiko | so to add the parameters, something like add(param, min, max, default) is needed in the CParameter class, right? | 18:09 |
| heiko | or would you do this with vectors? | 18:09 |
| @sonney2k | heiko, I would for now just do use the parameter class default + range extrensions later | 18:12 |
| heiko | When I understand this right: the paramter class does only store a pointer to its parameter data | 18:14 |
| heiko | then it is as simple as m_cross_val_parameters->add(&width, "width", "Kernel width."); | 18:14 |
| heiko | in the constructor | 18:14 |
| @sonney2k | hopefully yes | 18:15 |
| heiko | and to add this new Parameter to SGObject is also straight. only ojne question: serialization | 18:16 |
| heiko | if (!m_parameters->save(file, prefix)) | 18:16 |
| heiko | in SGObject | 18:16 |
| heiko | is it safe to just add a similar call for the other Parameter object? | 18:16 |
| heiko | seems to me | 18:17 |
| @sonney2k | but we don't need save / load right? just for setting / getting | 18:19 |
| @sonney2k | (though we could load from memory or so) | 18:19 |
| heiko | yes, true | 18:20 |
| heiko | how do you think should the getter/setter look like? how are parameter identified? | 18:21 |
| heiko | i thought of something like get_all_parameters | 18:22 |
| heiko | get_parameter(index) | 18:22 |
| heiko | set_parameter(index) | 18:22 |
| heiko | (, value) | 18:22 |
| @sonney2k | heiko, btw have a look at the xvalidateion framework that scikits.learn has | 18:24 |
| @sonney2k | we should get some inspiration from there | 18:24 |
| heiko | nice idea | 18:24 |
| @sonney2k | heiko, yes just iterate over parameters | 18:25 |
| @sonney2k | and parameters are either SGObjects's or scalars (or matrices,vectors) | 18:26 |
| @mlsec | bye guys | 18:56 |
| @mlsec | bye bettyboo | 18:57 |
| @bettyboo | mlsec: iirc the toolbox can't really do real valued classification properly, maybe I read it wrong. | 18:57 |
| heiko | byebye | 18:59 |
| * sonney2k leaves home and switches on his mobile | 19:16 | |
| @sonney2k | Yeah | 19:17 |
| -!- blackburn [~qdrgsm@188.122.237.148] has joined #shogun | 19:19 | |
| heiko | leaving | 19:20 |
| heiko | bye | 19:20 |
| alesis-novik | Good timing blackburn, I just came on myself | 19:20 |
| -!- heiko [~heiko@infole-06.uni-duisburg.de] has left #shogun [] | 19:20 | |
| @sonney2k | Blackburn huh missed the train? | 19:20 |
| alesis-novik | Good morning (?) | 19:20 |
| blackburn | hehe | 19:20 |
| @sonney2k | Bye | 19:21 |
| blackburn | alesis-novik: it's me :D | 19:21 |
| blackburn | sonney2k: why? and why train? | 19:21 |
| @sonney2k | Thought you are going home | 19:21 |
| blackburn | aha, now at home | 19:21 |
| alesis-novik | sonney2k, did you have a chance to review the changes I made? | 19:21 |
| @sonney2k | Hi alesis-novik any news on your patch? I didn't see Another pull request | 19:22 |
| blackburn | have I missed something? ;) | 19:22 |
| alesis-novik | Hmm, don't you get notified if a pull request get's changed? every push I make goes to the same pull request | 19:23 |
| @sonney2k | You are typing faster than me on this touch thing phone | 19:23 |
| @sonney2k | Haven't received anything | 19:23 |
| alesis-novik | weird. Well, the changes are on the same pull request | 19:24 |
| @sonney2k | Am I daydreaming | 19:24 |
| alesis-novik | alesis added some commits » | 19:24 |
| @sonney2k | Cannot check now will try | 19:24 |
| alesis-novik | I also made a big mess in PCACut | 19:25 |
| alesis-novik | But didn't push it on to this pull request | 19:26 |
| alesis-novik | basically now I made 3 different priority cut-off conditions: the threshold, user defined number or percentage of variance explained | 19:27 |
| alesis-novik | after we finish changing these two things (Gaussian and PCACut changes) up to standard, I have a new "project" | 19:28 |
| alesis-novik | Vojtech asked me to implement EM for 1D Gaussians | 19:28 |
| blackburn | why did you start working so early? | 19:29 |
| @sonney2k | Can you do that? | 19:29 |
| alesis-novik | This would probably be nowhere near the complexity of the proper EM | 19:29 |
| alesis-novik | Since I won't use any fancy things like SVD or removing singular Gaussians | 19:31 |
| @sonney2k | Yeah but can you duo that till Monday? | 19:31 |
| @sonney2k | Which is when we will have a meeting and decide | 19:32 |
| alesis-novik | I can probably code it until Monday. Question is if it will not have some style issues I'm having with the current pull :) | 19:32 |
| blackburn | sonney2k: do you mean we should work hardly till monday? | 19:33 |
| @sonney2k | Not everyone I guess | 19:34 |
| siddharth | hi all | 19:34 |
| alesis-novik | work hardly or hardly work? :D | 19:34 |
| blackburn | alesis-novik: just say how to say it properly ;) have bad english | 19:34 |
| * sonney2k hardly works hard | 19:35 | |
| siddharth | so all have started with their respective projects? | 19:35 |
| @sonney2k | Siddhart not everyone | 19:36 |
| blackburn | I don't, just 'crawling' to it | 19:36 |
| @sonney2k | We still don't know about 2 slots | 19:37 |
| siddharth | sonney2k, btw should I finish SGDQN by monday? | 19:37 |
| @sonney2k | If you can ... | 19:37 |
| * blackburn thinks about getting some task to work on :D | 19:37 | |
| blackburn | alesis-novik: nu tak kak pravilno? ya tupoy! :) | 19:38 |
| siddharth | sonney2k, I have understood the code...now I am getting to know shogun classes and function...will start coding tomorrow :) | 19:39 |
| @bettyboo | :> | 19:39 |
| alesis-novik | blackburn, work hard - mnogo rabotat', hardly work - pochti nerabotat' | 19:43 |
| blackburn | alesis-novik: thank you, really didn't feel the difference :) | 19:44 |
| * blackburn should be punished for his english | 19:44 | |
| alesis-novik | I think I have everything needed for a simple EM for 1d GMM implementation now | 19:48 |
| @sonney2k | Ok | 19:50 |
| alesis-novik | blackburn, can you believe vodka here in UK is 37.5 %? :D | 19:51 |
| @bettyboo | ;> | 19:51 |
| @sonney2k | Bah typing without keyboard sux | 19:52 |
| blackburn | alesis-novik: :D why it is? | 19:53 |
| alesis-novik | no idea. I told my dad about it and he was "Well, that's not real vodka then, is it?" :D | 20:08 |
| blackburn | don't sure there will be a serious 'taste' differences :) | 20:11 |
| alesis-novik | People don't drink it here without mixers anyway | 20:11 |
| -!- siddharth [~siddharth@117.211.88.150] has quit [Remote host closed the connection] | 20:20 | |
| * serialhex wakes up and goes over the chat log... | 20:29 | |
| serialhex | mmm... cold pizza | 20:29 |
| serialhex | you guys are always havong so much fun when i'm not around | 20:30 |
| * serialhex feels left out :P | 20:30 | |
| alesis-novik | I just woke up and hour ago :D | 20:31 |
| serialhex | niiice!! ~6 then? | 20:32 |
| * serialhex feels he woke up too early :P | 20:32 | |
| alesis-novik | fully nocturnal | 20:33 |
| serialhex | yeah, i dont think i can be at this point | 20:33 |
| -!- siddharth [~siddharth@117.211.88.150] has joined #shogun | 20:45 | |
| -!- serialhex [~quassel@99-101-149-136.lightspeed.wepbfl.sbcglobal.net] has quit [Remote host closed the connection] | 20:48 | |
| -!- serialhex [~quassel@99-101-149-136.lightspeed.wepbfl.sbcglobal.net] has joined #shogun | 20:49 | |
| alesis-novik | Right, it's Friday and it's beer'o'clock. I'll see some of you later :) I hope I'll see some comments on my pull as well :) | 21:50 |
| blackburn | alesis-novik: haha, have a nice beer | 21:51 |
| * sonney2k leaves the water | 21:56 | |
| @sonney2k | Ah that was a good swim | 21:57 |
| @sonney2k | Now if only this f... train arrives... | 21:58 |
| @sonney2k | Me checks the gsoc schedule. | 22:00 |
| @sonney2k | ...hmmhh deduplication meeting on the 22nd | 22:00 |
| serialhex | alesis-novik: beer'o'clock?? :P | 22:00 |
| @sonney2k | Did anyone here apply for other organizations swell | 22:01 |
| blackburn | sonney2k: is deduplication mean that you will have a list of accepted? | 22:01 |
| serialhex | i didnt :-/ everything else i was looking at that interested me was uncomfortably above my level | 22:02 |
| @sonney2k | Students can apply multiple times at different orgs | 22:02 |
| @sonney2k | Hmmhh noone? | 22:04 |
| blackburn | sonney2k: I don't | 22:04 |
| @sonney2k | That would be good then we might not have to attend that meeting | 22:04 |
| @sonney2k | ha | 22:05 |
| * sonney2k is finally on a real keyboard again | 22:05 | |
| @sonney2k | typing speeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeed | 22:05 |
| blackburn | hehehe | 22:05 |
| @sonney2k | uncompared | 22:05 |
| @sonney2k | I hope this struggle with not knowing who gets a slot is over soon | 22:06 |
| serialhex | i *hated* the ifone keyboard when i had one... so tiny and kind of wonky | 22:06 |
| * serialhex agrees with sonney2k | 22:06 | |
| @sonney2k | there is just no fairplay in there - ohh well there is you work more you get it if you have the skills though | 22:07 |
| @sonney2k | serialhex, android is slightly better | 22:07 |
| serialhex | in where?? the gsoc thing?? | 22:07 |
| @sonney2k | but hmmhh still I'd prefer a phone with a keyboard | 22:07 |
| @sonney2k | for irc at least | 22:07 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 22:07 | |
| serialhex | yeah, i have a 7" tablet with android 1.6 and it rocks! | 22:07 |
| serialhex | i especially lik the <- && -> keys, SOOO much easier to make edits with those instead of trying to touch between an i & an l :P | 22:08 |
| @bettyboo | HA! | 22:08 |
| @sonney2k | serialhex, yes | 22:08 |
| serialhex | ahh, so it's cutthroat politics in gsoc-land huh? | 22:09 |
| @sonney2k | *sigh* I am down to GPRS speed this month | 22:09 |
| * sonney2k dataplan is not sufficient for these irc / github sessions | 22:09 | |
| @sonney2k | 5 more days on that speed *sigh* | 22:09 |
| * serialhex feels sonney2k's pain | 22:10 | |
| serialhex | though i'm stuck with open wi-fi spots, so it's kinda worse | 22:10 |
| @sonney2k | github still didn't load... | 22:10 |
| @sonney2k | not to mention any email arrived just yet | 22:11 |
| serialhex | wow... something must be wrong with your connection :-/ | 22:12 |
| @sonney2k | btw, everyone git submodule update --init in the git reposoitory root gives you the data files | 22:13 |
| @sonney2k | I fixed the symlinks today | 22:13 |
| @sonney2k | so make check-examples | 22:13 |
| @sonney2k | and make tests should work now again | 22:13 |
| blackburn | oh! nice | 22:13 |
| @sonney2k | (though some are still broken) | 22:13 |
| serialhex | sweet!! | 22:14 |
| @sonney2k | blackburn, for example regression_svrlight_modular.py - it crashes when we train with >1 thread | 22:14 |
| blackburn | sonney2k: I didn't it :D | 22:14 |
| * sonney2k tryied to debug this for a few hours without success | 22:14 | |
| blackburn | hmm | 22:14 |
| @sonney2k | also kernel_salzberg_word_string_modular.py seems broken | 22:14 |
| blackburn | sonney2k: yeap! | 22:15 |
| @sonney2k | could be that it was always broken before though ?!?! | 22:15 |
| blackburn | forgot to report it | 22:15 |
| @sonney2k | blackburn, you caused the preprocs to fail :) | 22:15 |
| @sonney2k | but that was an easy fix | 22:15 |
| serialhex | shame on you blackburn!!! | 22:15 |
| blackburn | sonney2k: ehhh | 22:15 |
| blackburn | sonney2k: what's up with preprocs? | 22:15 |
| @sonney2k | blackburn, check the git log | 22:15 |
| blackburn | sonney2k: oh.. sorry | 22:16 |
| @sonney2k | heiko killed the stringfeatures and I cleaned up after you all :) | 22:16 |
| @sonney2k | no worries | 22:16 |
| @bettyboo | rahaha | 22:16 |
| serialhex | the 2 make commands, i run them ni the shogun/ dir right?? | 22:17 |
| @sonney2k | blackburn, you forgot just one const | 22:17 |
| @sonney2k | serialhex, in src dir | 22:17 |
| blackburn | sonney2k: I should to test it.. | 22:17 |
| serialhex | ok, just checking | 22:17 |
| @sonney2k | now tests don't run through for all stringfeatures | 22:17 |
| @sonney2k | due to new features heiko implemented | 22:17 |
| @sonney2k | so that cannot be avoided | 22:17 |
| @sonney2k | alright | 22:17 |
| @sonney2k | I am levaing the train | 22:17 |
| * serialhex LOVES being able to string multiple commands on the command line :D | 22:18 | |
| serialhex | i sooo don't miss windoze!!! | 22:19 |
| serialhex | i'm running those 2 commands sonney2k, and i have a handful of errors. i'm going to the grocery store for some supplies but i'll have a list of what isnt working right when i get back | 22:23 |
| serialhex | maybe it's stupid little things, but we'll see | 22:23 |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 22:33 | |
| blackburn | if (result<0) | 22:37 |
| blackburn | SG_ERROR("Negative area - x1=%f x2=%f y1=%f y2=%f\n", x1,x2, y1,y2); | 22:37 |
| blackburn | :DDDDDDD | 22:37 |
| @sonney2k | blackburn, great :/ | 22:37 |
| blackburn | better "Negative area, seems you are in black hole now" | 22:37 |
| blackburn | sonney2k: PerformanceMeasures.cpp | 22:38 |
| @sonney2k | blackburn, yes I knew when you showed that... | 22:38 |
| blackburn | but why?! | 22:38 |
| blackburn | why it is :) | 22:38 |
| blackburn | for me it is just like if (2+2!=4) SG_ERROR("Math isn't working in this universe"); | 22:39 |
| @sonney2k | blackburn, I guess there was an error at some point | 22:40 |
| @sonney2k | or even is | 22:40 |
| * sonney2k doesn't trust this module too much | 22:40 | |
| @sonney2k | I still have a big python script for all these functions | 22:40 |
| @sonney2k | definitely worth to polish this | 22:40 |
| blackburn | I could port it ;) | 22:40 |
| @sonney2k | do I have your email addr? | 22:41 |
| @sonney2k | ahh yes you posted on the ml | 22:41 |
| blackburn | blackburn91@gmail.com | 22:41 |
| @sonney2k | blackburn, actually it is in here http://mloss.org/software/view/262/ | 22:42 |
| @sonney2k | in mldata utils | 22:42 |
| @sonney2k | in directoy mleval | 22:43 |
| @sonney2k | then files multiclass.py classification.py and regression.py | 22:43 |
| @sonney2k | for some perfmeasures for the appr. kind | 22:43 |
| blackburn | sonney2k: ok, will take a look on it at this weekend :) | 22:46 |
| @sonney2k | blackburn, I wonder whether this thing shoudl be redesigned | 22:46 |
| blackburn | which thing? | 22:47 |
| @sonney2k | I mean just have 20 small classes or so and one baseclass | 22:47 |
| @sonney2k | perfmeasure | 22:47 |
| blackburn | we should, if you use not internal utils for that | 22:48 |
| blackburn | sonney2k: give me an advice, where should I inspire doing my CGraph class? | 22:49 |
| @sonney2k | blackburn, lib/ | 22:49 |
| blackburn | eh.. mean where should I look for some proper code | 22:49 |
| @sonney2k | blackburn, I mean heiko will need some option to select the perfmeasure | 22:49 |
| @sonney2k | blackburn, ahh there is none | 22:50 |
| blackburn | sonney2k: in some C++ libs | 22:50 |
| blackburn | like boost or etc | 22:50 |
| @sonney2k | we don't have graphs in shogun - I don't know which C++ lib there is good one | 22:50 |
| blackburn | ah | 22:50 |
| @sonney2k | google? | 22:50 |
| blackburn | okay :) I see | 22:50 |
| @sonney2k | http://stackoverflow.com/questions/2751826/which-c-graph-library-should-i-use | 22:51 |
| blackburn | sonney2k: google is a good thing, just asked if you had some experience :) | 22:51 |
| blackburn | sonney2k: btw, will I be an only user of this class? | 22:52 |
| @sonney2k | blackburn, so far yes | 22:52 |
| blackburn | * it needed by LLE and etc | 22:52 |
| @sonney2k | but you never know | 22:52 |
| @sonney2k | I mean graph kernels ... | 22:52 |
| @sonney2k | though they work usually with adjacency matrices | 22:52 |
| Tanmoy | @balckburn | 22:53 |
| @sonney2k | blackburn, so how can heiko select a perfmeasure? | 22:53 |
| Tanmoy | @blackburn if u get the graph stuff gr8 | 22:53 |
| Tanmoy | i need it too | 22:53 |
| @sonney2k | blackburn, I mean it will certainly need an interface | 22:53 |
| blackburn | sonney2k: I don't know! just say what you will need and I will do it | 22:53 |
| blackburn | sonney2k: how soon will heiko need it? | 22:54 |
| blackburn | Tanmoy: great, just let me know if you will need some special methods | 22:54 |
| @sonney2k | blackburn, I think it makes sense to have a abstract base class right with some evaluate() function that gets 2 Label objects as input | 22:55 |
| @sonney2k | and then does the comparison between the true one and the other one. | 22:55 |
| blackburn | sonney2k: what are derivates of this class? | 22:55 |
| Tanmoy | wld need it for graph kernels | 22:55 |
| Tanmoy | tht of submitting then realised dnt have graph | 22:56 |
| @sonney2k | blackburn, wow! http://www.ogdf.net/doc-ogdf/ has lots of graph business | 22:57 |
| blackburn | sonney2k: yeap, thanks. the only problem - I want to implement small and fast graph class | 22:57 |
| blackburn | will mind with time | 22:58 |
| @sonney2k | blackburn, hmmh http://lemon.cs.elte.hu/pub/tutorial/ looks then better maybe | 22:59 |
| @sonney2k | but I don't know how much you relaly need | 23:00 |
| blackburn | sonney2k: for LLE not much, just fast neighbors search and may be BFS, etc | 23:00 |
| blackburn | sonney2k: what's up with abstract class for evaluation? | 23:01 |
| @sonney2k | blackburn, an evalutate function that gets two label objects as args | 23:02 |
| @sonney2k | sigh, baby crying ... brb | 23:02 |
| blackburn | oh :) | 23:03 |
| blackburn | sonney2k: will try to represent it in UML, will be more clearly | 23:08 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Quit: Page closed] | 23:27 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 23:28 | |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Quit: Page closed] | 23:28 | |
| blackburn | sonney2k: will discuss some things with perf. measures tomorrow, now i'm yawning :) | 23:40 |
| blackburn | moreover you are away now :) | 23:41 |
| blackburn | sonney2k: let your baby not cry, good night | 23:47 |
| blackburn | see you | 23:47 |
| -!- blackburn [~qdrgsm@188.122.237.148] has quit [Quit: Leaving.] | 23:47 | |
| --- Log closed Sat Apr 16 00:00:36 2011 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!