--- Log opened Sat Mar 17 00:00:19 2012 | ||
-!- RagnarX [9d9f0a0e@gateway/web/freenode/ip.157.159.10.14] has joined #shogun | 00:05 | |
RagnarX | Hello, world ! | 00:06 |
---|---|---|
n4nd0 | RagnarX: hi there! | 00:12 |
RagnarX | n4nd0: how are you? :) | 00:13 |
n4nd0 | RagnarX: I am fine, what about you? | 00:13 |
RagnarX | I'm fine ! :D | 00:13 |
RagnarX | I'm a student and I have an idea which is not in the idea page ^^' | 00:14 |
n4nd0 | cool, I am a student too so I am not exactly the person you should propose it too | 00:14 |
n4nd0 | but anyway, what is your idea about? maybe I can suggest you something ;) | 00:15 |
RagnarX | It's about using OpenCL for Machine Learning :p | 00:15 |
n4nd0 | ok | 00:16 |
n4nd0 | I have no experience with OpenCL :( | 00:16 |
RagnarX | yes it's kind of special :p | 00:16 |
n4nd0 | I've heard of it of course, but nothing more than that | 00:17 |
RagnarX | :p | 00:17 |
RagnarX | where are you from? | 00:17 |
n4nd0 | what do you think it could be used for in ml? | 00:17 |
n4nd0 | I am from Spain, what about you? | 00:17 |
RagnarX | France ^^ | 00:18 |
RagnarX | to speed up machine learning :p | 00:18 |
RagnarX | I am a developer for a linear algebra library on OpenCL and started a MultiLayer Perceptron project using OpenCL, and from the test i did there was nice speed improvements | 00:18 |
n4nd0 | we use blas in shogun for linear algebra stuff | 00:19 |
n4nd0 | do you have any idea aobout comparisons between linear algebra done on OpenCL and blas? | 00:20 |
RagnarX | It mostly depends on the hardware :p but it is about 10 to 15 folds | 00:20 |
RagnarX | http://viennacl.sourceforge.net/viennacl-benchmarks.html | 00:21 |
RagnarX | I think the CPU benchmarks were done using blas | 00:21 |
n4nd0 | oh it looks like really good performance then | 00:22 |
RagnarX | yes :p | 00:23 |
RagnarX | are you a shogun developer? | 00:23 |
n4nd0 | I have done some contributions | 00:24 |
RagnarX | I see ^^ | 00:25 |
RagnarX | it looks like a cool project | 00:25 |
n4nd0 | :) yes I think it is indeed | 00:25 |
RagnarX | and gsoc always bring a lot of new developers | 00:27 |
RagnarX | :p | 00:27 |
n4nd0 | haha of course | 00:27 |
n4nd0 | so I think that you could try asking soeren about your idea | 00:32 |
n4nd0 | he is normally around here so he may answer you at any moment | 00:33 |
RagnarX | Haha okay :p | 00:33 |
RagnarX | isn't he sleeping right now? | 00:33 |
RagnarX | Well maybe he's not an European :p | 00:34 |
n4nd0 | oh sorry, I meant maybe not precisely now but if you come around tomorrow or sth like that | 00:35 |
RagnarX | Yes I probably will :) | 00:35 |
n4nd0 | in any case, you might want to use email | 00:35 |
RagnarX | Yes, I've just registered to the mailing list :p | 00:37 |
n4nd0 | good | 00:38 |
RagnarX | i've also cloned the repository, but it is kinda big :p | 00:40 |
n4nd0 | lot of work into there! | 00:41 |
n4nd0 | :) | 00:41 |
RagnarX | yep | 00:42 |
RagnarX | :p | 00:42 |
RagnarX | I think i'll send a mail on the list to describe my idea ^^ | 00:49 |
n4nd0 | that's probably a good idea | 00:53 |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun | 01:23 | |
n4nd0 | RagnarX: I have just made a quick search in the chat logs and found that at least they have commented the OpenCL topic a couple of times | 01:31 |
n4nd0 | RagnarX: maybe you want to check about it | 01:31 |
RagnarX | haha okay | 01:40 |
RagnarX | I think i'll firstly go to bed :p | 01:40 |
RagnarX | I have sent the mail today, and will probably read the logs tomorrow :p | 01:40 |
RagnarX | when was it? | 01:40 |
n4nd0 | google it | 01:51 |
n4nd0 | opencl site:shogun-toolbox.org/irclogs will tell you about it ;) | 01:52 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 01:58 | |
-!- RagnarX [9d9f0a0e@gateway/web/freenode/ip.157.159.10.14] has quit [Quit: Page closed] | 02:17 | |
-!- vikram360 [~vikram360@117.192.176.170] has joined #shogun | 03:45 | |
-!- vikram360 [~vikram360@117.192.176.170] has quit [Ping timeout: 260 seconds] | 03:52 | |
-!- vikram360 [~vikram360@117.192.179.129] has joined #shogun | 03:53 | |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 05:07 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Quit: This computer (or maybe me) has gone to sleep] | 05:43 | |
-!- harshit_ [~harshit@182.68.31.206] has quit [Ping timeout: 245 seconds] | 05:50 | |
-!- kkundu10__ [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has joined #shogun | 06:02 | |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 06:06 | |
-!- kkundu10__ [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has quit [Quit: Page closed] | 06:14 | |
-!- Miggy [~piggy@14.139.82.6] has joined #shogun | 06:30 | |
-!- harshit_ [~harshit@182.68.31.206] has quit [Remote host closed the connection] | 06:34 | |
-!- gsomix [~gsomix@178.45.67.23] has joined #shogun | 06:53 | |
-!- Miggy [~piggy@14.139.82.6] has quit [] | 06:56 | |
-!- in3xes [in3xes@49.249.16.237] has joined #shogun | 06:58 | |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 07:55 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 08:19 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 08:27 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit] | 08:29 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 09:01 | |
-!- blackburn [~qdrgsm@188.122.253.192] has joined #shogun | 09:21 | |
-!- in3xes_ [in3xes@49.249.16.237] has joined #shogun | 09:32 | |
-!- in3xes [in3xes@49.249.16.237] has quit [Ping timeout: 250 seconds] | 09:35 | |
-!- harshit__ [~harshit@182.68.31.206] has joined #shogun | 09:36 | |
harshit__ | hello everybody | 09:38 |
harshit__ | does anyone knw a function for minimum residual method in C++ | 09:39 |
blackburn | no idea | 09:39 |
harshit__ | i have been using TAUCS | 09:39 |
harshit__ | but its not working when i integrate it with shogun | 09:40 |
blackburn | I made use of superlu in some of my code here | 09:40 |
harshit__ | thanks i'll have a look on it | 09:41 |
-!- in3xes_ [in3xes@49.249.16.237] has quit [Quit: Leaving] | 09:58 | |
-!- harshit__ [~harshit@182.68.31.206] has quit [Ping timeout: 246 seconds] | 10:02 | |
-!- harshit_ [~harshit@182.68.31.206] has quit [Ping timeout: 248 seconds] | 10:02 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 252 seconds] | 10:22 | |
-!- GenX__ [~Sambhav@14.139.82.6] has joined #shogun | 10:43 | |
-!- GenX [~Sambhav@14.139.82.6] has quit [Ping timeout: 246 seconds] | 10:46 | |
-!- GenX__ is now known as GenX | 10:51 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 11:01 | |
-!- blackburn [~qdrgsm@188.122.253.192] has quit [Ping timeout: 246 seconds] | 11:02 | |
-!- aang [0e8be4d2@gateway/web/freenode/ip.14.139.228.210] has joined #shogun | 11:05 | |
aang | Hi | 11:08 |
n4nd0 | aang: hey! | 11:13 |
aang | Hi.. I am interested in contributing to shogun this year in GSoC | 11:15 |
aang | Is there any qualification task? | 11:15 |
n4nd0 | great, do you have any idea in mind? | 11:15 |
aang | I am familiar with the machine learning algorithms | 11:16 |
aang | and i wish to contribute in your idea of Integrating Existing and Implementing New Machine Learning Algorithms in shogun | 11:16 |
n4nd0 | aang: with qualification you mean something like a contribution to do before applying to GSoC? | 11:17 |
aang | Yes. | 11:17 |
n4nd0 | aang: I think that the page will be updated soon with info about that | 11:17 |
aang | Ok. | 11:18 |
n4nd0 | have to go now, bye | 11:18 |
aang | Which all projects are in your priority list this year? | 11:18 |
aang | Ok, catch you later !! | 11:18 |
n4nd0 | all the ones that appear in the ideas page are stuff that would be interesting to have | 11:19 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 11:19 | |
-!- pokoko222 [~chatzilla@77.28.121.98] has joined #shogun | 11:32 | |
pokoko222 | hello | 11:32 |
pokoko222 | Are there students involved in your project, for gsoc? | 11:33 |
pokoko222 | I am looking for projects and I am curious if there are already serious candidates for your projects | 11:33 |
-!- l0nr4n [~l0nr4n@g225161164.adsl.alicedsl.de] has joined #shogun | 11:40 | |
-!- l0nr4n [~l0nr4n@g225161164.adsl.alicedsl.de] has quit [Changing host] | 11:40 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun | 11:40 | |
-!- vikram360 [~vikram360@117.192.179.129] has quit [Ping timeout: 255 seconds] | 11:41 | |
-!- GenX [~Sambhav@14.139.82.6] has quit [Quit: Leaving] | 11:48 | |
-!- aang [0e8be4d2@gateway/web/freenode/ip.14.139.228.210] has quit [Ping timeout: 245 seconds] | 11:49 | |
-!- gsomix [~gsomix@178.45.67.23] has quit [Ping timeout: 246 seconds] | 11:58 | |
-!- pokoko222 [~chatzilla@77.28.121.98] has quit [Quit: ChatZilla 0.9.88.1 [Firefox 10.0.2/20120215223356]] | 11:59 | |
-!- gsomix [~gsomix@95.67.141.113] has joined #shogun | 12:10 | |
-!- GenX [~Sambhav@14.139.82.6] has joined #shogun | 12:14 | |
-!- GenX [~Sambhav@14.139.82.6] has quit [Read error: Connection reset by peer] | 12:31 | |
-!- GenX [~Sambhav@14.139.82.6] has joined #shogun | 12:31 | |
wiking | :> | 12:37 |
wiking | amazing how popular suddenly #shogun got with gsoc ;) | 12:38 |
-!- GenX__ [~Sambhav@14.139.82.6] has joined #shogun | 12:42 | |
gsomix | everything was becoming popular with gsoc :) | 12:42 |
-!- GenX [~Sambhav@14.139.82.6] has quit [Ping timeout: 276 seconds] | 12:46 | |
gsomix | is there anybody who is waiting Stanford's online courses about ML? | 12:47 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 12:51 | |
-!- varagrawal [7bc955a2@gateway/web/freenode/ip.123.201.85.162] has joined #shogun | 13:08 | |
varagrawal | Hey! | 13:08 |
varagrawal | I am a student of Computer Science and Engineering from National Institute of Tech, Surat, INdia | 13:09 |
-!- vikram360 [~vikram360@117.192.179.129] has joined #shogun | 13:09 | |
varagrawal | I wanted to join in the Shogun toolkit Open source project | 13:09 |
varagrawal | I have read your list of ideas page for GSoC 2012 | 13:10 |
n4nd0 | hi | 13:10 |
varagrawal | and being an undergrad with some basic knowledge of Machine Learning | 13:10 |
n4nd0 | varagrawal: good, what idea did you find appealing? | 13:11 |
varagrawal | I was thinking I could help with either Olivier Chapelles fast newton based SVM solver | 13:11 |
n4nd0 | ok | 13:13 |
n4nd0 | do you have experience with shogun? I mean if you have used it before | 13:13 |
wiking | n4nd0: :> you should be a mentor ;) | 13:14 |
n4nd0 | wiking: haha I don't think so :) | 13:15 |
wiking | well u are all the time guiding possible gsocers here, so why not ;P | 13:15 |
-!- varagrawal [7bc955a2@gateway/web/freenode/ip.123.201.85.162] has quit [Ping timeout: 245 seconds] | 13:15 | |
n4nd0 | I have still stuff to learn as student :) I actually like being student | 13:16 |
-!- varagrawal [744a6982@gateway/web/freenode/ip.116.74.105.130] has joined #shogun | 13:16 | |
varagrawal | Hey | 13:16 |
wiking | hihi | 13:16 |
varagrawal | sorry | 13:16 |
n4nd0 | but it is good to help people | 13:16 |
n4nd0 | varagrawal: np | 13:16 |
varagrawal | net got disconnected | 13:16 |
varagrawal | the other idea was the implementation of Gaussian processes and regression techniques | 13:17 |
n4nd0 | varagrawal: cool, those ideas look nice | 13:17 |
n4nd0 | varagrawal: so have you used shogun before for sth? | 13:17 |
varagrawal | I haven't used shogun, but I have used octave | 13:18 |
wiking | varagrawal: it'd be nice that u start something with shogun before trying to apply for a gsoc project ... | 13:18 |
varagrawal | I am starting to port those octave codes to shogun and try and achieve similar functionality | 13:18 |
wiking | so at least check out the latest code | 13:18 |
n4nd0 | exactly | 13:18 |
wiking | and try to implement something in it | 13:18 |
n4nd0 | varagrawal: have you already forked the repo? | 13:19 |
wiking | since you'll have do most of the work in c/c++ | 13:19 |
varagrawal | Yeah, I have a DBMS project, where I try to guess a good branch of study for a person | 13:19 |
varagrawal | depending on their rank in the entrance exam | 13:19 |
varagrawal | I was thinking of using Shogun for that with the KNN algo most probably | 13:19 |
n4nd0 | ok | 13:20 |
varagrawal | I will check out the code now | 13:20 |
n4nd0 | but you might want you check out shogun code from the repo | 13:20 |
varagrawal | Yes | 13:21 |
varagrawal | Am cloning it now.. | 13:21 |
n4nd0 | cool | 13:21 |
n4nd0 | so maybe it is a good idea to try out some of the examples and get used to the structure of the project | 13:22 |
varagrawal | aww net went again.... | 13:22 |
varagrawal | I'll do that | 13:22 |
n4nd0 | :) | 13:22 |
varagrawal | Ok, thanks a lot for the suggestions :D | 13:22 |
n4nd0 | varagrawal: you're welcome | 13:23 |
varagrawal | I'll sit now and get back to you guys tomorrow morning fine? | 13:23 |
n4nd0 | sure | 13:23 |
varagrawal | Thank you :) | 13:23 |
-!- varagrawal [744a6982@gateway/web/freenode/ip.116.74.105.130] has quit [Quit: Page closed] | 13:23 | |
n4nd0 | wiking: how is it going with your latent variables stuff? | 13:24 |
-!- vikram360 [~vikram360@117.192.179.129] has quit [Ping timeout: 260 seconds] | 13:35 | |
-!- vikram360 [~vikram360@117.192.179.129] has joined #shogun | 13:40 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Operation timed out] | 14:02 | |
-!- vikram360 [~vikram360@117.192.179.129] has quit [Ping timeout: 252 seconds] | 14:15 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 14:35 | |
-!- GenX__ is now known as GenX | 14:42 | |
-!- vikram360 [~vikram360@117.192.168.236] has joined #shogun | 15:01 | |
-!- GenX__ [~Sambhav@14.139.82.6] has joined #shogun | 15:47 | |
-!- GenX [~Sambhav@14.139.82.6] has quit [Ping timeout: 245 seconds] | 15:48 | |
-!- blackburn [~qdrgsm@188.122.253.192] has joined #shogun | 16:06 | |
wiking | n4nd0: hehehe just starting to sketch up the classes | 16:06 |
wiking | blackburn: yoyo man | 16:06 |
wiking | blackburn: your fucking email made a mess here in the channel :))) | 16:06 |
n4nd0 | haha :) | 16:07 |
blackburn | wiking: which one? | 16:07 |
wiking | blackburn: gsoc...now there's heaps of people coming to talk about various gsoc projects ;) | 16:08 |
blackburn | oh well we can't keep it in secret | 16:08 |
wiking | :P | 16:08 |
n4nd0 | so what do you think guys about the number of slots? | 16:11 |
n4nd0 | is it likely to get those 9 you talked about yesterday? | 16:11 |
wiking | it's amazing that VLC did not get a spot in gsoc this year :( | 16:12 |
n4nd0 | I didn't know about that | 16:12 |
n4nd0 | is there also a list of rejected orgs? | 16:13 |
wiking | i don't think so | 16:13 |
wiking | i've just seen it in twitter | 16:13 |
n4nd0 | ok | 16:13 |
n4nd0 | think of it this way ... we are contributing to sth more important than VLC :P | 16:14 |
wiking | comeoooon | 16:15 |
wiking | vlc is essential we can all agree | 16:15 |
n4nd0 | haha just kidding | 16:15 |
n4nd0 | different projects indeed | 16:16 |
n4nd0 | wiking: about this deep learning, is it very good? | 16:19 |
wiking | yep | 16:20 |
n4nd0 | what is the main idea with it? | 16:20 |
wiking | since it's unsupervised | 16:20 |
wiking | and get's very good results | 16:20 |
wiking | well it's a neural net | 16:20 |
n4nd0 | ok | 16:20 |
n4nd0 | I've used neural nets in supervised learning but nothing further than that | 16:21 |
n4nd0 | it is always cool to hear from new stuff | 16:21 |
wiking | well the mentioned theano toolbox is good enough | 16:21 |
wiking | it even has cuda backend... | 16:21 |
blackburn | I think we can get 7-8 at least | 16:24 |
blackburn | cause we have reached success with 5 slots | 16:25 |
blackburn | but 9 would be wonderful | 16:25 |
n4nd0 | let's keep fingers crossed | 16:27 |
n4nd0 | is that also announced as well? | 16:27 |
n4nd0 | or just when applications are submitted and the students are notified? | 16:28 |
blackburn | we will be notified how many slots we have got a little bit later | 16:54 |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 17:00 | |
-!- in3xes [in3xes@49.248.244.21] has joined #shogun | 17:13 | |
-!- harshit_ [~harshit@182.68.31.206] has quit [Ping timeout: 272 seconds] | 17:17 | |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 17:29 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 245 seconds] | 17:33 | |
-!- harshit__ [~harshit@182.68.31.206] has joined #shogun | 17:33 | |
-!- Ram108 [~amma@117.231.112.53] has joined #shogun | 17:43 | |
-!- Ram108 [~amma@117.231.112.53] has quit [Read error: Connection reset by peer] | 17:44 | |
-!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has joined #shogun | 17:57 | |
puffin444 | Hi | 17:57 |
-!- surtani [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has joined #shogun | 18:02 | |
puffin444 | hello | 18:02 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 18:02 | |
gsomix | puffin444, hi | 18:03 |
puffin444 | Hey, I'm a student interested in the Gaussian regression project for the GSoC. | 18:04 |
puffin444 | Do you know if the mentor (Stegle) hangs out on IRC, or should I try contacting him via email? | 18:05 |
-!- harshit_ [~harshit@182.68.31.206] has quit [Ping timeout: 252 seconds] | 18:05 | |
-!- harshit__ [~harshit@182.68.31.206] has quit [Ping timeout: 246 seconds] | 18:05 | |
n4nd0 | puffin444: I may be wrong but I think that I have not seen him around here | 18:10 |
gsomix | puffin444, i think, in this case, email is better than chat. | 18:12 |
blackburn | puffin444: these guys are right, please contact him by email | 18:18 |
puffin444 | Okay. I'll try to contact him via email. Would it help my case if I tried to fix a minor bug in Shogun? | 18:19 |
blackburn | puffin444: any contributions are encouraged and would be an advantage | 18:22 |
-!- RagnarX [52e82691@gateway/web/freenode/ip.82.232.38.145] has joined #shogun | 18:22 | |
RagnarX | Hello ^^ | 18:23 |
blackburn | hi | 18:23 |
n4nd0 | RagnarX: hey! | 18:24 |
RagnarX | How are you all? :p | 18:26 |
n4nd0 | I am good :) what about you? | 18:26 |
RagnarX | I'm fine :) | 18:27 |
-!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has left #shogun [] | 18:44 | |
-!- in3xes [in3xes@49.248.244.21] has quit [Read error: Connection reset by peer] | 18:45 | |
-!- in3xes [in3xes@49.248.244.21] has joined #shogun | 18:46 | |
n4nd0 | have to go now guys | 18:49 |
n4nd0 | bye | 18:49 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 18:49 | |
-!- in3xes [in3xes@49.248.244.21] has quit [Read error: Connection reset by peer] | 18:50 | |
-!- in3xes [in3xes@2002:31f8:f415::31f8:f415] has joined #shogun | 18:50 | |
-!- harshit_ [~harshit@182.68.31.206] has joined #shogun | 19:01 | |
-!- vikram360 [~vikram360@117.192.168.236] has quit [Ping timeout: 244 seconds] | 19:13 | |
-!- harshit__ [~harshit@182.68.31.206] has joined #shogun | 19:18 | |
-!- RagnarX [52e82691@gateway/web/freenode/ip.82.232.38.145] has quit [Ping timeout: 245 seconds] | 19:26 | |
-!- harshit__ [~harshit@182.68.31.206] has quit [Remote host closed the connection] | 19:29 | |
-!- harshit_ [~harshit@182.68.31.206] has quit [Remote host closed the connection] | 19:29 | |
-!- surtani [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has quit [Ping timeout: 245 seconds] | 19:54 | |
-!- PhilTillet [52e82691@gateway/web/freenode/ip.82.232.38.145] has joined #shogun | 20:52 | |
-!- NexT [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has joined #shogun | 21:18 | |
-!- NexT [0e8b5206@gateway/web/freenode/ip.14.139.82.6] has left #shogun [] | 21:19 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has quit [Ping timeout: 265 seconds] | 21:48 | |
-!- l0nr4n [~l0nr4n@g231201166.adsl.alicedsl.de] has joined #shogun | 21:49 | |
-!- l0nr4n [~l0nr4n@g231201166.adsl.alicedsl.de] has quit [Changing host] | 21:49 | |
-!- l0nr4n [~l0nr4n@unaffiliated/l0nr4n] has joined #shogun | 21:49 | |
-!- gsomix [~gsomix@95.67.141.113] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)] | 21:57 | |
-!- in3xes [in3xes@2002:31f8:f415::31f8:f415] has quit [Quit: Leaving] | 22:27 | |
PhilTillet | Hello hello | 22:59 |
blackburn | hi | 23:03 |
PhilTillet | I have seen that it was necessary to submit a little pull request for Shogun ... :p How "small" does it have to be? | 23:12 |
blackburn | PhilTillet: well.. | 23:17 |
blackburn | it is not about size I think | 23:17 |
PhilTillet | The other problem is that I don't really understand the "Issues" section on the github, to be honest | 23:18 |
blackburn | it is not highly relevant, it is ok | 23:18 |
blackburn | I've got not much time to keep it updated | 23:18 |
PhilTillet | oh ok :p | 23:19 |
PhilTillet | So i should rather browse code and see "what i can do" ? :p | 23:19 |
blackburn | yeah rather do that | 23:19 |
blackburn | or ask me/soeren | 23:19 |
PhilTillet | okay :p | 23:19 |
PhilTillet | Yes, the source code is rather big. I think i wouldn't be against having some pointers | 23:20 |
blackburn | do you know any kernel that is not implemented in shogun? | 23:21 |
PhilTillet | Well, my field is more Neural Networks / Optimization than SVM | 23:22 |
PhilTillet | According to the website, conjugate gradient is not implemented, if I remember well | 23:22 |
PhilTillet | And BFGS either | 23:23 |
blackburn | well it could turn as a lot of work? | 23:25 |
PhilTillet | Hmmm, well. It is something I have already done in C++ (even though I think my L-BFGS was buggy :p) ; it mostly depends on, well, how to interface it | 23:26 |
PhilTillet | concretely, what should be minimized | 23:27 |
PhilTillet | if it is just a functor which can return the value of a function and a gradient, it shouldn't be long | 23:28 |
blackburn | can't you adapt it for some svm? | 23:28 |
PhilTillet | Hmmm | 23:30 |
PhilTillet | I think i would need to put myself up to date with SVM, but I could try | 23:31 |
blackburn | I still do not know even my opinion on implementing something NN here | 23:31 |
PhilTillet | well, it's true that SVM are very efficient, and most of the time better than NN | 23:32 |
PhilTillet | well at least MLP | 23:32 |
blackburn | it is not about efficiency - we've got a pretty long list of ideas | 23:34 |
PhilTillet | And i think it would be indeed hard to implement CG ... I mean i'd have to first review all the architecture of SVMs in Shogun ... hmm ... | 23:34 |
blackburn | then no need to do that | 23:34 |
blackburn | our req about small patch is just a kind of a 'filter' | 23:35 |
blackburn | cause we are probably going to receive >50 (~70 last year) proposals | 23:35 |
PhilTillet | Got it :) | 23:36 |
blackburn | at least it makes things more fair | 23:36 |
PhilTillet | Well as far as I know Shogun implements linear, gaussian, sigmoid kernel | 23:36 |
blackburn | i.e. chances we pick some student who just wrote good proposal and too lame with any development are pretty low then | 23:37 |
blackburn | uh | 23:37 |
PhilTillet | (and others) | 23:37 |
blackburn | we have pretty many kernels here | 23:37 |
PhilTillet | I know :p | 23:37 |
PhilTillet | Wanted to tell that these 3 are the only kernels I know | 23:37 |
blackburn | ah | 23:37 |
blackburn | btw why do you want to work only on OpenCL? | 23:38 |
PhilTillet | well, I don't want to work only on OpenCL | 23:38 |
blackburn | did you found any other ideas interesting? | 23:39 |
blackburn | as I said before we don't mind anything related to such things but we are pretty tight with slots | 23:39 |
PhilTillet | the point is that I have "a lot" of experience using OpenCL because I have already done some projects on it and quite a bit of Linear Algebra on it too, and as of now no machine learning library use OpenCL so ... :) but the "Built generic multiclass learning framework" idea seems cool too | 23:41 |
blackburn | aha I see | 23:41 |
blackburn | ok I just got how can we solve it | 23:42 |
blackburn | while you have a lot of experience with OpenCL you may help us with some basic operations here | 23:42 |
blackburn | like dot product etc | 23:42 |
PhilTillet | yep | 23:42 |
blackburn | but mainly you may work on something different | 23:42 |
PhilTillet | I see :p | 23:42 |
blackburn | this is the case if we won't receive much slots | 23:43 |
PhilTillet | Well, the point is that i'm also a contributor to an OpenCL linear algebra library | 23:43 |
PhilTillet | which already implements dot product ... | 23:43 |
blackburn | then probably it won't take much time for you | 23:43 |
PhilTillet | yes, hope so :) | 23:44 |
PhilTillet | but why not using already existing code? | 23:44 |
blackburn | I did not say you can't use that code ;) | 23:44 |
PhilTillet | haha :p | 23:44 |
blackburn | depends on license though | 23:44 |
PhilTillet | is MIT license okay? | 23:45 |
blackburn | probably | 23:46 |
blackburn | iirc we can integrate mit code in gpl project | 23:46 |
PhilTillet | yes I think too, just have to be careful about how to integrate the license | 23:48 |
PhilTillet | I'll ask to the project leader of that library :) | 23:50 |
blackburn | sure | 23:50 |
PhilTillet | well, to get back to that generic multiclass learning framework | 23:51 |
PhilTillet | would it be about making some generic code that would map any binary classifier to multiclass ? | 23:52 |
blackburn | there is such code already done by me recently | 23:52 |
blackburn | probably main thing here would be to implement ECOC and some tree learning stuff | 23:53 |
PhilTillet | I see | 23:55 |
PhilTillet | there seems to be already a paper and some matlab code | 23:55 |
blackburn | yes | 23:56 |
PhilTillet | This will involve some generic programming right? | 23:56 |
--- Log closed Sun Mar 18 00:00:19 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!