| --- Log opened Wed Apr 06 00:00:36 2011 | ||
| -!- iLucian [~tc@145.116.231.246] has quit [Quit: Leaving] | 00:01 | |
| -!- brocolli [ylorb@unaffiliated/brocolli] has joined #shogun | 00:44 | |
| -!- user_ [533c6a96@gateway/web/freenode/ip.83.60.106.150] has joined #shogun | 00:48 | |
| user_ | hi everybody! | 00:49 |
|---|---|---|
| @bettyboo | yo | 00:49 |
| user_ | Hi, I'm almost a newbie in c++ and just have a little question | 00:50 |
| blackburn | to whom do you address it? | 01:01 |
| -!- user_ [533c6a96@gateway/web/freenode/ip.83.60.106.150] has quit [Quit: Page closed] | 01:17 | |
| lionelc | well, Shogun developers, I just found the problem I reported is ticket #182 http://trac.tuebingen.mpg.de/shogun/ticket/182 | 01:17 |
| lionelc | has it already been solved? | 01:18 |
| blackburn | lionelc: you could report with your problem to mailist | 01:23 |
| lionelc | ok thanks | 01:26 |
| blackburn | lionelc: I think Sören is away now, so you may not get answer here now, a bit late | 01:28 |
| lionelc | that's fine. I just realize that it is very late in Europe | 01:28 |
| lionelc | blackburn: am sending it to the mailing list shortly | 01:29 |
| blackburn | here where I am, it is even a very-very late :) | 01:29 |
| @bettyboo | HA! | 01:29 |
| blackburn | lionelc: there are some discussion of troubles with shogun on macos in mailing list | 01:30 |
| blackburn | btw, are you installing latest version? | 01:31 |
| lionelc | blackburn: oh I actually see the same problem reported yesterday. it seems to be a common issue | 01:32 |
| lionelc | blackburn: yes. 0.10.0 | 01:33 |
| blackburn | so may be you will find something in mailing list | 01:34 |
| blackburn | going sleep, see you later | 01:35 |
| -!- blackburn [~qdrgsm@188.168.5.125] has left #shogun [] | 01:35 | |
| lionelc | have a nice dream :-) ttyl | 01:35 |
| -!- siddharth_ [~siddharth@117.211.88.150] has quit [Ping timeout: 276 seconds] | 02:42 | |
| -!- sploving [~root@124.16.139.196] has left #shogun [] | 03:03 | |
| -!- Rickasaurus [~Rickasaur@pool-108-35-36-236.nwrknj.fios.verizon.net] has joined #shogun | 03:04 | |
| -!- Rickasaurus [~Rickasaur@pool-108-35-36-236.nwrknj.fios.verizon.net] has left #shogun [] | 03:08 | |
| -!- lionelc_ [48cb93da@gateway/web/freenode/ip.72.203.147.218] has joined #shogun | 03:39 | |
| -!- lionelc [48cb93da@gateway/web/freenode/ip.72.203.147.218] has quit [Ping timeout: 252 seconds] | 03:41 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 04:36 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has joined #shogun | 05:39 | |
| hiteshk | sonney2k , thr | 05:40 |
| -!- siddharth [~siddharth@117.211.88.150] has joined #shogun | 05:45 | |
| -!- aifargonos [~aifargono@46.18.27.35] has left #shogun [] | 05:59 | |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 06:03 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has left #shogun [] | 06:27 | |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 06:38 | |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 07:00 | |
| -!- Ryaether [~Ryaether@50-80-170-245.client.mchsi.com] has joined #shogun | 07:06 | |
| @sonney2k | lionelc_, please use the current git version | 07:10 |
| * sonney2k is hardly awake | 07:10 | |
| CIA-31 | shogun: Ameer Ayoub master * rea1505c / src/configure : Fixed test_ruby(), previously would error if ruby was installed but not ruby-dev. - http://bit.ly/dUMxHd | 07:19 |
| CIA-31 | shogun: Soeren Sonnenburg master * r159bbb4 / .gitignore : remove *.pyc - occurred twice - http://bit.ly/e5FnyJ | 07:19 |
| -!- Ziyuan [~Ziyuan@116.22.48.72] has joined #shogun | 07:20 | |
| CIA-31 | shogun: Soeren Sonnenburg master * rd7dab5c / src/configure : make configure executable - http://bit.ly/h8j7EO | 07:20 |
| -!- serialhex [~quassel@99-101-149-136.lightspeed.wepbfl.sbcglobal.net] has joined #shogun | 07:29 | |
| -!- lionelc_ is now known as lionelc | 07:39 | |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 07:46 | |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has joined #shogun | 08:14 | |
| warpy | hey guys, anyone here ? | 08:14 |
| warpy | could it possibly be that the shogun site is offline for some reason ? | 08:15 |
| ameerkat | it appears to be working for me, let me clear my caches and double check | 08:17 |
| ameerkat | still working | 08:17 |
| warpy | didnt work 10 mins ago | 08:25 |
| warpy | ouch | 08:25 |
| warpy | i was wondering about two things | 08:25 |
| warpy | are there any c# wrappers for it? the other, they report 10mil training samples,do you know how big was each sample (without going into those papers)? | 08:26 |
| ameerkat | I'm pretty new to the project but there's no c# interface from what I've seen in the code and documentation. I have no idea about the scale of the samples with regards to your second question so I'll let someone else more knowledgable answer that lol | 08:32 |
| warpy | heh sure | 08:38 |
| warpy | i'm hoping to get a student to do that for me | 08:38 |
| warpy | so if there is we'll contribute it back | 08:39 |
| @sonney2k | warpy, nope the site is online | 08:58 |
| @sonney2k | warpy, well the biggest training dataset I used was about 7GB in size, but I exploded it to have ~185 million dimensions (sparse features!) and 50million examples | 08:59 |
| @sonney2k | ameerkat, btw, thanks for the patch | 09:04 |
| ameerkat | no problem | 09:06 |
| @sonney2k | I've only just added the changes to get shogun's modular interfaces to compile with java, lua, ruby | 09:09 |
| @sonney2k | but the typemaps for these languages / examples are still missing | 09:10 |
| @sonney2k | warpy: I guess adding C# would be as easy but someone would need to do the typemaps... the most I could do is prepare configure for it | 09:11 |
| -!- Ryaether [~Ryaether@50-80-170-245.client.mchsi.com] has left #shogun [] | 09:20 | |
| warpy | 50mil examples and 185 features ? | 09:32 |
| warpy | i was thinking of compiling it to a c++ dll and using that as the base. but i'd need to consult someone who knows c++ better | 09:33 |
| warpy | sonney2k, did you just email me back ? | 09:33 |
| @sonney2k | 50mil examples and 185mil features | 09:34 |
| @sonney2k | yes | 09:34 |
| warpy | what was the train/test split, did you use 10cfv? how long did the training last, which kernel, what kind of feature data, binary, float, decimal, how long did the testing last ? | 09:36 |
| warpy | lots of qs :) | 09:36 |
| @bettyboo | :> | 09:36 |
| -!- blackburn [~qdrgsm@188.168.4.108] has joined #shogun | 09:39 | |
| -!- sploving [~sploving@124.16.139.196] has joined #shogun | 09:39 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has joined #shogun | 09:40 | |
| @sonney2k | sploving, btw the modular lua interface compiles and does not crash at least | 09:42 |
| @sonney2k | sploving, though I don't know how to call functions etc, after require of some module... | 09:42 |
| hiteshk | sonney2k , hi | 09:43 |
| blackburn | sonney2k: oh, hi, Sören | 09:43 |
| @sonney2k | warpy, it is from that paper http://sonnenburgs.de/soeren/coffin | 09:43 |
| blackburn | sonney2k: seen my patch? is it necessary, etc? | 09:43 |
| @sonney2k | hi all | 09:43 |
| @bettyboo | morning | 09:44 |
| @sonney2k | blackburn, yes I am just going through it... | 09:44 |
| @sonney2k | blackburn, so I can give you feedback directly now: | 09:44 |
| blackburn | sonney2k: I could send you a test, proving it's efficiency and correctness :) should I? | 09:44 |
| @bettyboo | ;D | 09:44 |
| sploving | hi sonney2k. I know that. | 09:45 |
| @sonney2k | blackburn, if you say you tested it I do even believe you that it works :) | 09:45 |
| @bettyboo | yo | 09:45 |
| @sonney2k | sploving, so how? | 09:45 |
| hiteshk | sonney2k , I mailed you my proposal ; did you get it | 09:46 |
| @sonney2k | blackburn, I think I will apply it as is but then make the classify_NN function protected - I mean this way one from the outside only ever calls classify() and internally it is using the most efficient routines | 09:46 |
| sploving | I have not write somecode now, as i am busying writing a research paper now | 09:46 |
| sploving | sonney2k, i saw you comment | 09:47 |
| warpy | thanks sonney2k | 09:47 |
| warpy | will be back later | 09:47 |
| @sonney2k | hiteshk, yes - I already had a look - I have some feedback. | 09:47 |
| @bettyboo | ho warpy | 09:47 |
| @sonney2k | (not now though) | 09:47 |
| warpy | hey bettyboo. bbs | 09:47 |
| @bettyboo | warpy: so the implementation of kfa does not requires any special kernel or feature representation | 09:47 |
| @bettyboo | warpy, yo | 09:47 |
| sploving | sonney2k, does anyother mentor ? all youself to lookup the proposal is tired | 09:48 |
| blackburn | sonney2k: should I send new patch with protectec classify_NN()? | 09:48 |
| @sonney2k | warpy, don't get trapped - betty is a bot | 09:48 |
| @bettyboo | sonney2k: I don't pretty sure, but there is an classifier/boosting | 09:48 |
| @sonney2k | blackburn, changes are only marginal so I can do that on my own | 09:48 |
| blackburn | sonney2k: okay, thank you. looking forward my first patch in shogun :) | 09:49 |
| @bettyboo | rahaha | 09:49 |
| blackburn | forward to. sorry | 09:49 |
| @sonney2k | sploving, well there are about 10 but I am the administrator so I have the highest load | 09:49 |
| hiteshk | sonney2k, will you mail me the feedback | 09:50 |
| @sonney2k | sploving, can you just tell me how I call a function in lua? I mean if I have a Kernel class in C++ what will it be in lua? Can I just write k=GaussianKernel() and then call k.fun() ? | 09:50 |
| @sonney2k | hiteshk, yes give me some more time please | 09:50 |
| hiteshk | Ok | 09:50 |
| warpy | heh okay | 09:52 |
| warpy | cya | 09:52 |
| sploving | K:fun() | 09:52 |
| sploving | the memer using k.a, the function using k:func | 09:53 |
| sploving | sonney2k, I mean k=GaussianKernel(), we use k:fun() to call the function, use k.member to get/set the member | 09:56 |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has quit [Ping timeout: 246 seconds] | 09:56 | |
| @sonney2k | sploving, hmmhh: $ lua | 09:57 |
| @sonney2k | Lua 5.1.4 Copyright (C) 1994-2008 Lua.org, PUC-Rio | 09:57 |
| @sonney2k | > require('Kernel') | 09:57 |
| @sonney2k | > k=GaussianKernel() | 09:57 |
| @sonney2k | stdin:1: attempt to call global 'GaussianKernel' (a nil value) | 09:57 |
| @sonney2k | stack traceback: | 09:57 |
| @sonney2k | stdin:1: in main chunk | 09:57 |
| @sonney2k | [C]: ? | 09:57 |
| @sonney2k | ahh so Kernel::GaussianKernel? | 09:57 |
| @sonney2k | let me try | 09:57 |
| -!- sonney2k was kicked from #shogun by bettyboo [flood] | 09:57 | |
| -!- sonney2k [~shogun@87.118.92.43] has joined #shogun | 09:58 | |
| -!- mode/#shogun [+o sonney2k] by ChanServ | 09:58 | |
| @sonney2k | bettyboo, nasty you!@ | 09:58 |
| @bettyboo | sonney2k: I noticed that all kernel-based methods derive from CKernelMachine | 09:58 |
| blackburn | kicked and bored you | 09:58 |
| blackburn | :) | 09:58 |
| @sonney2k | sploving, well closer: k=Kernel:GaussianKernel() | 10:00 |
| @sonney2k | Wrong arguments for overloaded function 'new_GaussianKernel' | 10:00 |
| @sonney2k | ..... | 10:00 |
| @sonney2k | but still doesn't work | 10:00 |
| sploving | maybe you set the arguments not right | 10:00 |
| sploving | i will have a try. Please wait for sometime | 10:02 |
| CIA-31 | shogun: Sergey Lisitsyn master * r3d8ccac / (2 files): Introduced new method classify_NN() for fast classification in case of k=1 - http://bit.ly/eJbdzu | 10:02 |
| CIA-31 | shogun: Sergey Lisitsyn master * r729a29e / src/libshogun/classifier/KNN.cpp : Added redirection to classify_NN(CFeatures* data) in general classify(CFeatures* data) in case of k==1 - http://bit.ly/dPynda | 10:02 |
| CIA-31 | shogun: Soeren Sonnenburg master * r6213351 / (2 files): | 10:02 |
| CIA-31 | shogun: make classify_NN protected; improve coding style | 10:02 |
| CIA-31 | shogun: (...by introducing a few spaces, brackets) - http://bit.ly/g3kdIK | 10:02 |
| CIA-31 | shogun: Soeren Sonnenburg master * ra8891a0 / (.gitignore src/configure): Merge branch 'master' of github.com:shogun-toolbox/shogun - http://bit.ly/i85SuH | 10:02 |
| @sonney2k | sploving, ok | 10:03 |
| @sonney2k | bettyboo, you dare killing our CIA bot! | 10:03 |
| sploving | I have to reboot the machines(now i am on windows) | 10:03 |
| @bettyboo | sonney2k: Have you seen the old school youtube or 3d xkcd? | 10:03 |
| sploving | see you | 10:03 |
| @sonney2k | blackburn, so your patch is in! | 10:03 |
| -!- sploving [~sploving@124.16.139.196] has quit [] | 10:03 | |
| blackburn | sonney2k: oh,yeah | 10:04 |
| blackburn | :) | 10:04 |
| @sonney2k | knrrrd, set up bettyboo with whatever cleverness but is now on vacations without internet... | 10:04 |
| @bettyboo | sonney2k: What packages do I need to enable HDF5 support? | 10:04 |
| blackburn | sonney2k: thank you. will make some other improvements in time :) | 10:05 |
| blackburn | sonney2k: and hope to get into GSoC.. | 10:05 |
| blackburn | oh, wonder I have git account. | 10:06 |
| -!- sploving [~root@124.16.139.196] has joined #shogun | 10:06 | |
| blackburn | sonney2k: thank you for your modification. there is no need of classify_NN(CFeatures* data) at all, I've done it sick | 10:14 |
| @sonney2k | blackburn, well if the classify_NN function is protected then not - and you can save the error checks | 10:15 |
| @sonney2k | but yes you could have re-used the code like it is now in classify(CFeatures* data) | 10:16 |
| blackburn | yeap | 10:16 |
| blackburn | I don't know is it a necessarily patch, but could be a start of me contributing :) | 10:17 |
| @sonney2k | blackburn, it is a good idea to start with small patches - so you get feedback and slowly can dive into deeper waters... | 10:18 |
| blackburn | at least doing this way I can get familiar with shogun development | 10:18 |
| blackburn | will send further patches later, thank you very much | 10:19 |
| blackburn | btw, sonney2k, you said there is about ten proposals? | 10:20 |
| sploving | Error: Unknown interface: libshogui | 10:21 |
| @sonney2k | blackburn, that we can potentially accept yes. However it is unclear how many slots we will get and I expect the total number of submitted proposals to be between 40-50 ... | 10:21 |
| sploving | ./configure --interfaces=libshogui,libshogunui, lua_modular, java_modular, r_modular, python_modular, octave_modular,ruby_modular | 10:22 |
| blackburn | ah, even 40-50.. | 10:22 |
| @sonney2k | sploving, just do ./configure --interfaces=libshogun,lua_modular for now | 10:22 |
| sploving | ok | 10:22 |
| @sonney2k | blackburn, well we had one student submitting 11 | 10:22 |
| blackburn | 11? seriuously? | 10:23 |
| blackburn | seriously* | 10:23 |
| @sonney2k | so 40-50 proposals != #students | 10:23 |
| blackburn | 11 proposals from one studend, different ones, right? | 10:23 |
| blackburn | student* | 10:23 |
| blackburn | what's the matter with me | 10:23 |
| blackburn | making mistakes | 10:23 |
| @sonney2k | blackburn, yes - so relax chances are not too bad I think | 10:23 |
| @sonney2k | well not really different | 10:24 |
| @sonney2k | (marginally at best) | 10:24 |
| blackburn | so, what he want with doing this way? | 10:24 |
| blackburn | seems to be a spam but not a proposal.. | 10:24 |
| blackburn | as for me, I even didn't sent it, discussing it with Chris and coming up with it gradually | 10:25 |
| blackburn | i will propose implentations of dimensionality reduction algorithms | 10:26 |
| sploving | sonney2k, a lot of undefined:"/root/gsoc_work/shogun/src/lua_modular/Classifier_wrap.cxx:13438: undefined reference to `lua_pushfstring'" | 10:27 |
| @sonney2k | blackburn, no idea... | 10:28 |
| @sonney2k | blackburn, yes sounds good to me... | 10:28 |
| blackburn | in particular, ISOMAP, MDS, LLE, SNE | 10:28 |
| @sonney2k | sploving, err this is with current git? | 10:28 |
| sploving | I git clone it after you send the email | 10:29 |
| sploving | how could i use the git to see the version | 10:29 |
| @sonney2k | sploving, git log ... but just do git pull and try again | 10:29 |
| -!- hiteshk [~hitesh_na@117.211.88.150] has left #shogun [] | 10:33 | |
| sploving | sonney2k, i git pull, and shows: | 10:36 |
| sploving | cannot find -llua5.1 | 10:37 |
| @sonney2k | sploving, so you don't have the lua developer files installed? | 10:37 |
| @sonney2k | sploving, you are on ubuntu? | 10:37 |
| sploving | fedora, i installed lua | 10:38 |
| @sonney2k | liblua5.1-0, liblua5.1-0-dev are the packages here | 10:38 |
| sploving | lLua 5.1.3 Copyright (C) 1994-2008 Lua.org, PUC-Rio | 10:38 |
| sploving | > | 10:38 |
| @sonney2k | ls /usr/lib/liblua* | 10:38 |
| @sonney2k | err please run ls /usr/lib/liblua*.so | 10:38 |
| sploving | I know that my liblua.so in /usr/local/lib | 10:39 |
| sploving | i will use "ldconfig" | 10:39 |
| sploving | a lot of warning now | 10:41 |
| sploving | likes :"Features.i:303: Warning 453: Can't apply (shogun::TSparse< uint64_t > **ARGOUT_SPARSE,int32_t *DIM1,int32_t *DIM2,int64_t *NNZ). No typemaps are defined. | 10:41 |
| sploving | " | 10:41 |
| sploving | this typemap is what i need to do | 10:42 |
| sploving | in gsoc | 10:42 |
| @sonney2k | hiteshk, I've written you an email with comments... | 10:43 |
| @sonney2k | hmmhh just see that he left :( | 10:44 |
| @sonney2k | sploving, yes | 10:44 |
| @sonney2k | well then please do ls /usr/local/lib/liblua*.so | 10:44 |
| sploving | /usr/local/lib/liblua.so /usr/local/lib/libluaT.so | 10:46 |
| sploving | this is the ls result | 10:46 |
| @sonney2k | sploving, OK so it is not suffixed with the lua version | 10:47 |
| @sonney2k | then for now manually edit .config in the src dir and change -llua5.1 to -llua | 10:48 |
| sploving | i will try | 10:48 |
| blackburn | sonney2k: when you sent a patch, github automatically found my account on github, right? | 10:48 |
| @sonney2k | ? | 10:49 |
| blackburn | I mean when I commit it I only write my name and email | 10:49 |
| blackburn | when using github.com there is a link to my account (those even I forgot) | 10:49 |
| @sonney2k | blackburn, you mean that one https://github.com/SergeyLisitsyn ? | 10:50 |
| blackburn | yeap | 10:50 |
| @sonney2k | blackburn, the best way to work with github is to fork shogun internally | 10:51 |
| @sonney2k | then you can mess around in your git and push to your repository when you are ready | 10:51 |
| blackburn | I've done it this way | 10:52 |
| sploving | sonney2k, you mean :"LINKFLAGS_LUA="-llua${LV} --shared" change to LINKFLAGS_LUA="-llua --shared" | 10:52 |
| @sonney2k | then you can even issue a pull request and I get an email like ameerkat has done in https://github.com/shogun-toolbox/shogun/pull/1 | 10:52 |
| blackburn | I just wonder how github put a link to me :) | 10:52 |
| @sonney2k | sploving, no just open the file 'src/.config' and change it there | 10:53 |
| @sonney2k | blackburn, I guess based on your committer id | 10:53 |
| sploving | ok | 10:53 |
| blackburn | but I haven't set up anything related to this account in my eclipse. I think it's because e-mail | 10:54 |
| blackburn | oh, anyway thank you and sorry for disturbing | 10:54 |
| @sonney2k | well git config -l | 10:54 |
| blackburn | that's that I mean | 10:56 |
| blackburn | there is another user.name, not SergeyLisitsyn | 10:57 |
| blackburn | but user.email is right, so may be github linked it because it | 10:57 |
| sploving | sonney2k, my ls /usr/local/include/lua* is as the following: | 10:57 |
| sploving | /usr/local/include/luaconf.h /usr/local/include/lualib.h | 10:57 |
| sploving | /usr/local/include/lua.h /usr/local/include/luaT.h | 10:57 |
| sploving | then i also need change the INCLUDES_LUA = | 10:57 |
| @sonney2k | sploving, that should be ok then no? | 10:58 |
| @sonney2k | I mean /usr/local/include will also be in the include path then | 10:58 |
| blackburn | sonney2k: one more question, I'm lecturing an optional machine learning classes at my university and have beamer presentations for my lections, should I put a link to it in my proposal? (the problem - it is in russian, so may be only you can understand something) | 11:01 |
| @sonney2k | blackburn, 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 :) | 11:02 |
| @bettyboo | lol ;> | 11:02 |
| blackburn | привет, как дела? :) | 11:03 |
| -!- sploving [~root@124.16.139.196] has quit [Ping timeout: 246 seconds] | 11:03 | |
| blackburn | makes sense? :) | 11:04 |
| blackburn | you could know russian a bit more than I know german | 11:04 |
| @sonney2k | I don't remember what privet (no idea how you manage to get these characters typed) means | 11:04 |
| blackburn | it is 'hello' | 11:04 |
| @sonney2k | so it is hello, how are you? | 11:05 |
| blackburn | yeap | 11:05 |
| @sonney2k | I only know things like, kak tebja sowut (looks weird in these latin characters though) | 11:05 |
| siddharth | hi sonney2k | 11:06 |
| siddharth | I have just mailed you my proposal | 11:06 |
| blackburn | kak tebya zovut :) | 11:06 |
| @bettyboo | yep | 11:06 |
| siddharth | like to know your views/comments | 11:06 |
| blackburn | it is like Hallo! Wie geht es Ihnen? as well (am I spell it right?) | 11:06 |
| @sonney2k | blackburn, yes that is correct German :) | 11:07 |
| @bettyboo | rahaha? :> | 11:07 |
| @sonney2k | siddharth, I've just seen the email | 11:07 |
| blackburn | i even know 'Einen wunderschónen guten Tag' | 11:07 |
| blackburn | oh, guys will not distub you :) | 11:07 |
| @sonney2k | heh | 11:08 |
| blackburn | oh, not ó, but ö | 11:08 |
| @sonney2k | siddharth, I guess you could mention the matlab dim-red toolbox by its name / with link | 11:09 |
| siddharth | sonney2k, ok sure | 11:09 |
| blackburn | another one want to work with dim reduction? | 11:10 |
| blackburn | sad | 11:10 |
| siddharth | yeap | 11:10 |
| siddharth | there are many algorithms,maybe we can divide | 11:11 |
| blackburn | what algorithms do you propose to implement? | 11:11 |
| blackburn | yeap, in case of both accepted :0 | 11:11 |
| blackburn | :) | 11:11 |
| siddharth | :D | 11:11 |
| @bettyboo | ;D | 11:11 |
| @sonney2k | what shall I say there are many algorithms and siddharth also has another proposal not only dim red methods | 11:11 |
| @sonney2k | s/many algorithms/many dim red. algos/ | 11:11 |
| blackburn | sonney2k: is it necessary to put a link to recommendation letter by my associate professor? | 11:12 |
| siddharth | thats not needed I think | 11:12 |
| @sonney2k | siddharth, my guess is that you won't need all the time to do the newtonsvm if you take that one | 11:13 |
| @sonney2k | blackburn, not required but can strengthen you proposal of course | 11:14 |
| blackburn | sonney2k: okay, thank you | 11:14 |
| @sonney2k | siddharth, also writing interfaces is not really an issue with shogun - it more or less automatically provides interfaces | 11:14 |
| siddharth | sonney2k, yes thats why I was thinking to work on integrating other SVM solver | 11:14 |
| siddharth | ohh k | 11:15 |
| lionelc | sonney2k: hi, I am a student who also applied for GSoC. just sent an email to the mailing list about building shogun on a cluster. Is there any chance to discuss with you about the "framework improvement" project? | 11:15 |
| @sonney2k | siddharth, ahh yes only looked at your proposal not emal text | 11:15 |
| @sonney2k | 11:15 | |
| @sonney2k | I guess simplesvm would make sense too | 11:15 |
| @sonney2k | ^^ siddharth | 11:16 |
| @sonney2k | and the task is similar: code is in matlab - translate to C++ | 11:16 |
| @sonney2k | there is also a paper about simplesvm somewhere | 11:16 |
| siddharth | sonney2k, yes I will include that in the proposal along with primal SVM solver | 11:16 |
| siddharth | ok i will search tha paper | 11:17 |
| @sonney2k | siddharth, I mean newton svm is much easier I suspect (few lines of code) simple svm could be a bigger job | 11:17 |
| siddharth | I have read the paper on Olivier Chapelles | 11:17 |
| @sonney2k | siddharth, ok | 11:17 |
| siddharth | well then I will start working on simple svm first | 11:18 |
| siddharth | sonney2k, what about integrating SGD-QN | 11:19 |
| @sonney2k | siddharth, no its better to start with the simple task to get used to it (and finishing something increases motivation) | 11:19 |
| @sonney2k | is also a valid option - much easier too | 11:20 |
| @sonney2k | (probably) | 11:20 |
| @sonney2k | but have a look at the simple svm code - maybe it is not that much - I mean it must be called simplesvm for a reason | 11:20 |
| siddharth | so should I include all the three :) | 11:20 |
| @sonney2k | anyway got to go now | 11:20 |
| @bettyboo | ho | 11:20 |
| siddharth | yeah | 11:21 |
| @sonney2k | lionelc, please if you have specific questions email them - I have no time right now. | 11:21 |
| lionelc | sonney2k: sure. thanks. ttyl | 11:21 |
| @sonney2k | siddharth, well check that it is realistic. I mean I guess you could do newton in 1-2 weeks but hey YMMV | 11:21 |
| siddharth | sonney2k, I will let you know later | 11:21 |
| @bettyboo | hey | 11:21 |
| @sonney2k | l8r | 11:22 |
| -!- sploving [~root@124.16.139.196] has joined #shogun | 11:23 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 248 seconds] | 11:42 | |
| Tanmoy | @sonney2k for structured learning you have prototype of Semi Markov SVM or HMM-SVM | 11:54 |
| Tanmoy | ? | 11:54 |
| -!- sploving [~root@124.16.139.196] has quit [Ping timeout: 260 seconds] | 11:58 | |
| -!- aifargonos [~aifargono@46.18.27.35] has joined #shogun | 12:00 | |
| @sonney2k | Tanmoy, we don't have prototypes for SO learnign, we only have the argmax for SHM-SVM | 12:07 |
| @sonney2k | Tanmoy, doing the prototype would be the work to do | 12:07 |
| Tanmoy | http://mloss.org/software/view/250/ | 12:12 |
| @sonney2k | Tanmoy, that is outside of shogun - written in matlab mostly | 12:13 |
| @sonney2k | not really nice code :( | 12:13 |
| @sonney2k | but both nico and jonas know much more about it (I don't) | 12:13 |
| -!- sploving [~root@124.16.139.196] has joined #shogun | 12:13 | |
| Tanmoy | oh okie | 12:13 |
| @sonney2k | sploving, so did it compile? | 12:14 |
| Tanmoy | @sonney2k i tht u implemented tht :) | 12:14 |
| @bettyboo | rahaha | 12:14 |
| @sonney2k | Tanmoy, I implemented a lot including HMMs / viterbi decoding that gunnar evolved into semi-hiddem viterbi decoding that is now the base for the SO stuff | 12:14 |
| Tanmoy | yup | 12:15 |
| Tanmoy | had a word with Nico yday ...thnks for the word wth him | 12:15 |
| @sonney2k | Tanmoy, I think emailing nico / jonas is the better way to communicate with them | 12:16 |
| sploving | sonney2k, I went to supper just now | 12:16 |
| sploving | it compile well | 12:16 |
| Tanmoy | oh i did that....i think they were unwell | 12:17 |
| @sonney2k | sploving, I will do so soon too. Very good. So could you run anything? | 12:17 |
| Tanmoy | had a word abt what is to be done ...... | 12:17 |
| Tanmoy | i think i was more interested in CRF's aka Undirected Models while i think the emphasis wld be on SO-SVM | 12:17 |
| sploving | sonney2k, are there examples that we do not need the typemap? | 12:19 |
| sploving | very simple exmaple | 12:19 |
| @sonney2k | sploving, well to run simple things we don't even need typemaps | 12:19 |
| sploving | sonney2k, I found that the exmples in the python_modular/ or others are complex | 12:20 |
| @sonney2k | e.g. in python I can do >>> from shogun.Features import Labels | 12:20 |
| @sonney2k | >>> x=Labels() | 12:20 |
| @sonney2k | sonney2k, well there are simple and more complex ones | 12:20 |
| @sonney2k | sploving, I just want to create one object and call one function ... | 12:23 |
| @sonney2k | sploving, no complex stuff yet | 12:24 |
| sploving | oh i know | 12:24 |
| @sonney2k | hmmhh, have to leave now | 12:24 |
| @sonney2k | l8r | 12:24 |
| sploving | sonney2k, in r, it uses library(shogun); then in octave, it used init_shogun | 12:26 |
| sploving | why are they not same? | 12:27 |
| sploving | then in lua, we need require('shogun') | 12:29 |
| sploving | could i know where is shogun.so | 12:30 |
| sploving | ? | 12:30 |
| -!- blackburn [~qdrgsm@188.168.4.108] has quit [Quit: Leaving.] | 12:31 | |
| @sonney2k | sploving, well it is modular | 12:39 |
| @sonney2k | we decided to have everything separately, i.e. Features.so Kernel.so Classifier.so | 12:40 |
| @sonney2k | etc | 12:40 |
| @sonney2k | I don't know if it is reasonably possible to have everything in one .so but then have the structure as before | 12:41 |
| -!- sploving [~root@124.16.139.196] has quit [Read error: Operation timed out] | 12:43 | |
| siddharth | sonney2k, I just went through the documentation of simple svm | 13:05 |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 13:05 | |
| * siddharth brb | 13:06 | |
| -!- sploving [~sploving@124.16.139.198] has joined #shogun | 13:34 | |
| sploving | hello my network had some problem just now | 13:34 |
| @bettyboo | morning | 13:34 |
| sploving | hello bettyboo. | 13:34 |
| @bettyboo | sploving: yes that is correct German :) | 13:34 |
| @bettyboo | yo | 13:34 |
| sploving | which projetc are you mentor? | 13:35 |
| sploving | hello, sonney2k, i know where is the problem. when call GaussianKernel, | 13:36 |
| @bettyboo | hey | 13:36 |
| sploving | the wrapper code is "shogun::CGaussianKernel" | 13:36 |
| sploving | so we should use it in lua like these: | 13:36 |
| sploving | require("shogun") | 13:37 |
| sploving | x = shogun.CGaussianKernel() | 13:37 |
| sploving | x.getwidth() | 13:38 |
| sploving | no, x:getwidth() | 13:38 |
| sploving | hello bettyboo | 13:42 |
| sploving | are you a mentor? | 13:42 |
| @bettyboo | sploving: *rofl* | 13:42 |
| @bettyboo | hiho | 13:42 |
| @sonney2k | sploving, betty boo is a bot | 13:51 |
| @bettyboo | sonney2k: did you look at the book? | 13:51 |
| @sonney2k | grrgh | 13:51 |
| sploving | hello sonney2k, when i tried require("shogun") | 13:52 |
| @bettyboo | hiho | 13:52 |
| @sonney2k | sploving, it really works :) | 13:52 |
| @bettyboo | he ;> | 13:52 |
| @sonney2k | $ lua | 13:52 |
| @sonney2k | Lua 5.1.4 Copyright (C) 1994-2008 Lua.org, PUC-Rio | 13:52 |
| @sonney2k | > require('Kernel') | 13:52 |
| @sonney2k | > x=Kernel.GaussianKernel() | 13:52 |
| sploving | what is the reason? | 13:53 |
| @sonney2k | sploving, how can I print out values, e.g. when I do x:get_width() (that should return a double) it doesn't show anything | 13:54 |
| sploving | print(x:get_width()) | 13:54 |
| @sonney2k | sploving, well you showed me the correct syntax that is why | 13:54 |
| @sonney2k | print(x:get_width()) | 13:54 |
| @sonney2k | 1 | 13:54 |
| @sonney2k | :-) | 13:54 |
| sploving | is it OK? | 13:54 |
| sploving | very well | 13:54 |
| @sonney2k | yes - works :) | 13:55 |
| @bettyboo | ;> sonney2k | 13:55 |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Quit: Page closed] | 13:55 | |
| sploving | it works well in my machine now | 13:55 |
| @sonney2k | sploving, that means for lua it is really only the typemaps that are left to do -> good news | 13:56 |
| sploving | yeap | 13:56 |
| * sonney2k checks ruby | 13:56 | |
| CIA-31 | shogun: Soeren Sonnenburg master * rce7a0f6 / (2 files): | 14:02 |
| CIA-31 | shogun: add a function to set rbf kernel width | 14:02 |
| CIA-31 | shogun: such that it can now be set not only when constructing the object - http://bit.ly/hn9Pwu | 14:02 |
| CIA-31 | shogun: Soeren Sonnenburg master * r4a00202 / .gitignore : ignore the cmdline binary - http://bit.ly/hxsZFv | 14:02 |
| siddharth | sonney2k, I went through the documentation of simple svm | 14:16 |
| siddharth | they are basically solving the dual form | 14:16 |
| siddharth | i.e alpha | 14:16 |
| @sonney2k | siddharth, yes it is a kernel svm | 14:17 |
| @sonney2k | but they have a different way of getting the solution | 14:17 |
| siddharth | They have used Hard margin and L1 soft margin as objective function | 14:17 |
| siddharth | yes they have 3 constraints on alpha | 14:17 |
| siddharth | alpha=0,C and between 0 and C | 14:18 |
| * sonney2k checks http://www.hpl.hp.com/conferences/icml2003/papers/352.pdf | 14:18 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has joined #shogun | 14:20 | |
| siddharth | I think this is somewhat similar to the SMO algorithm | 14:21 |
| @sonney2k | but it is not as simple ... you have a whole working set to consider | 14:33 |
| siddharth | sonney2k, hmm...I will read the paper in detail | 14:33 |
| @sonney2k | siddharth, in libsvm / SMO one chooses 2 variables at a time | 14:34 |
| @sonney2k | and then solves the problem for just these 2 variables | 14:34 |
| siddharth | actually I am currently busy in writing research paper for the BMVC conference | 14:34 |
| siddharth | yes | 14:34 |
| @sonney2k | siddharth, here it seems you have 3 sets and one set is the working set that can have x (=any number of) variables | 14:35 |
| @sonney2k | well, try to give some rough idea in the proposal - should be ok | 14:35 |
| siddharth | yes i will do that...Is any patch necessary? | 14:36 |
| siddharth | my deadline for conference is 15th march so... | 14:36 |
| siddharth | sry April | 14:36 |
| @sonney2k | siddharth, Re patches - it certainly adds weight to your proposal | 14:37 |
| siddharth | ok will try to work something out | 14:38 |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has joined #shogun | 14:38 | |
| @sonney2k | sploving, ruby works too:LD_LIBRARY_PATH=../libshogun irb1.8 | 14:52 |
| @sonney2k | irb(main):001:0> require('Kernel') | 14:52 |
| @sonney2k | ./Kernel.so: warning: rb_define_const: invalid name `__STDC_FORMAT_MACROS' for constant | 14:52 |
| @sonney2k | => true | 14:52 |
| @sonney2k | irb(main):002:0> x = GaussianKernel.new() | 14:52 |
| @sonney2k | => #<Kernel::GaussianKernel:0x7f0a976f5e98> | 14:52 |
| @sonney2k | irb(main):003:0> x.get_width() | 14:52 |
| @sonney2k | => 1.0 | 14:52 |
| sploving | sonney2k, good news | 14:53 |
| @sonney2k | heh | 14:53 |
| @sonney2k | in the java case I am not sure how to test it: in the end ton's of .java files are lying around | 14:54 |
| sploving | you mean in ruby, just add LD_LIBRARY_PATH, nothing anything do we need change | 14:55 |
| @sonney2k | sploving, well if you don't want to issue make install then you have to set the path to libshogun | 14:58 |
| sploving | another thing about lua is that we need many require | 14:59 |
| sploving | require("Kernel"), require("Classifier") ... | 15:00 |
| sploving | I think we shoud just require("shogun") | 15:00 |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 15:32 | |
| -!- sploving [~sploving@124.16.139.198] has quit [] | 15:37 | |
| -!- makokal [~nyumbanil@2001:638:709:43:daa2:5eff:fe91:d5ca] has joined #shogun | 15:46 | |
| -!- siddharth_ [~siddharth@117.211.88.150] has joined #shogun | 15:57 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has quit [Ping timeout: 246 seconds] | 15:59 | |
| -!- siddharth [~siddharth@117.211.88.150] has quit [Ping timeout: 276 seconds] | 16:00 | |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 16:23 | |
| Tanmoy | is nicco or jonas arnd ? | 16:24 |
| @sonney2k | no, I think they prefer email | 16:24 |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has joined #shogun | 16:31 | |
| CIA-31 | shogun: François-David Collin multiboost * rc2ffd34 / (197 files in 13 dirs): Merge with last MultiBoost version + hack of class registering scheme taking MultiBoost peculiarities into account - http://bit.ly/gcvBa9 | 16:37 |
| CIA-31 | shogun: François-David Collin multiboost * r8e28e51 / : Merge remote-tracking branch 'origin/multiboost' - http://bit.ly/dL5CE0 | 16:37 |
| CIA-31 | shogun: François-David Collin multiboost * r516549c / (10 files in 3 dirs): New MultiBoost Learners - http://bit.ly/i7GyIl | 16:37 |
| -!- hiteshk [~hitesh_na@117.211.88.150] has joined #shogun | 16:47 | |
| -!- makokal [~nyumbanil@2001:638:709:43:daa2:5eff:fe91:d5ca] has quit [Quit: makokal] | 16:59 | |
| -!- akhil_ [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Quit: Page closed] | 17:00 | |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has quit [Ping timeout: 246 seconds] | 17:01 | |
| -!- Tanmoy [75d35896@gateway/web/freenode/ip.117.211.88.150] has quit [Ping timeout: 252 seconds] | 17:12 | |
| -!- blackburn [~qdrgsm@188.168.5.128] has joined #shogun | 17:14 | |
| -!- makokal [~nyumbanil@2001:638:709:43:daa2:5eff:fe91:d5ca] has joined #shogun | 17:19 | |
| Ziyuan | In "get_column_sum" of Mathematics.h, the casting is dubious. | 17:48 |
| Ziyuan | (int)0.99999=0, where we prefer 1. | 17:49 |
| Ziyuan | So a rounding is better. | 17:50 |
| Ziyuan | And a typo...T* rowsums=get_column_sum(para) | 17:53 |
| Ziyuan | (Fixing | 17:54 |
| CIA-31 | shogun: Soeren Sonnenburg master * rc0c5990 / src/libshogun/lib/HDF5File.cpp : | 18:14 |
| CIA-31 | shogun: drop using hdf5 1.6 based functions | 18:14 |
| CIA-31 | shogun: Now use *2 functions like H5Dopen2 etc - http://bit.ly/ieuT9Q | 18:14 |
| @sonney2k | Ziyuan, looking | 18:14 |
| -!- makokal [~nyumbanil@2001:638:709:43:daa2:5eff:fe91:d5ca] has quit [Quit: makokal] | 18:16 | |
| @sonney2k | Ziyuan, it should be T* colsums=new T[m]; indeed | 18:16 |
| @sonney2k | Ziyuan, the other row sum thingy is not really an issue in the standard use case (matrix is squared) but you are right | 18:18 |
| CIA-31 | shogun: Soeren Sonnenburg master * r827a449 / src/libshogun/lib/Mathematics.h : Fix type issues - thanks Lin Ziyuan for the report. - http://bit.ly/eYHlMP | 18:24 |
| Ziyuan | But it will cause problem when calculating the mean if all types are int | 18:25 |
| Ziyuan | problems | 18:25 |
| @sonney2k | Ziyuan, which line are you referring to please? | 18:32 |
| Ziyuan | Oh sorry, no problems. | 18:33 |
| Ziyuan | num_data is double, in center_matrix | 18:34 |
| brocolli | man | 18:36 |
| brocolli | I hope I get selected, I want to do the Olivier Chapelle PSVM implementation. | 18:36 |
| Ziyuan | Yes? | 18:36 |
| brocolli | That thing is epic. | 18:36 |
| -!- makokal [~nyumbanil@212.201.44.245] has joined #shogun | 18:38 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 19:35 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 19:36 | |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has joined #shogun | 19:41 | |
| -!- Bautista [533c6a96@gateway/web/freenode/ip.83.60.106.150] has joined #shogun | 19:43 | |
| Bautista | Hi everybody, to whom may direct questions about running the octave examples? | 19:44 |
| Bautista | My problem is that when I try to run the examples an the data is loaded by the function load_matrix.m | 19:53 |
| Bautista | it searches for data in the ../data folder | 19:53 |
| Bautista | which is empty | 19:53 |
| Bautista | do i assume that the data in this folder has to be the one in git? | 19:54 |
| Bautista | yes it was a correct assumption | 20:01 |
| Bautista | thanks anyways :) | 20:01 |
| @bettyboo | ;> | 20:01 |
| -!- Bautista [533c6a96@gateway/web/freenode/ip.83.60.106.150] has quit [Quit: Page closed] | 20:06 | |
| brocolli | Just curious, how many of the applicants are undergraduates, and how many are graduates? | 20:25 |
| makokal | @brocolli I am a finishing undergrad | 20:27 |
| brocolli | Oh. I see. | 20:27 |
| serialhex | *undergrad... though as far as statistics go, i dont know the ratio... | 20:28 |
| brocolli | I am just curious, hehe. | 20:28 |
| serialhex | np | 20:28 |
| brocolli | I hope I"m not the tallest midget, that would be unfortunate ;( I don't think graduates should even be allowed but heh I figured why not | 20:29 |
| serialhex | i think i saw at least one person on the mailing list talking about his PhD & working on this... | 20:30 |
| brocolli | Oh I see. | 20:30 |
| serialhex | so you're probably not the tallest midget :P | 20:30 |
| brocolli | Phew! | 20:30 |
| @bettyboo | yep ;> | 20:30 |
| brocolli | Some of those "difficult" ones in red are truly difficult and i've been doing ML for nearly 2 years | 20:30 |
| serialhex | damn... yeah i'm just starting on alot of this stuff, and so i'm going for the ruby typemaps thing... interested in doing the cooler stuff, but not yet skilled enough | 20:31 |
| brocolli | Heh I was fortunate to have a Machine Learning class as an undergrad. IT was cross listed with the grads, but man that helped so much in my last year of undergrad. | 20:32 |
| brocolli | From there I just started doing so much stuff related to it, the more experience you get the more comfortable you will get with it. I will say this: Ididn't feel like I learned anything in my ML class even though I finished in the top 2 :/ | 20:33 |
| brocolli | The theory is so heavy, but it's so beautiful when you see the origins of mathematics/probability theory come in to the fold and give us these nice properties to exploit ;) | 20:33 |
| @bettyboo | he | 20:33 |
| serialhex | cool cool | 20:33 |
| brocolli | Make sure you know you're linear algebra like the back of your hand ! ;) | 20:33 |
| serialhex | there arent many courses like that listed at my college (or any around here) so i was going to sneak it in while doing game programming :D | 20:34 |
| * serialhex makes mental note to get a textbook | 20:34 | |
| brocolli | hehe, the elitist in me comes out (even though I'm not) when I hear "games" heheh | 20:34 |
| brocolli | Game theory | 20:34 |
| brocolli | Too many textbooks my friend,all good at different things. | 20:34 |
| brocolli | For example, a great book that's "readable" for ML is called Pattern REcognition by bishop, but it lacks rigor. | 20:34 |
| serialhex | yeah, game theory is cool but != games | 20:35 |
| serialhex | ...i think i've heard of it | 20:35 |
| brocolli | Then there's Learning Kernel Classifiers by the great Ralf Herbrich, but it's too rigorous for some. | 20:35 |
| serialhex | cool | 20:35 |
| brocolli | yeah I didn't mind Game Theory but it was too limited for me when I took the course. Solving games == eh. | 20:35 |
| brocolli | My favorite is Pattern Classification by Duta Hart and Stork. | 20:35 |
| brocolli | It's a bridge between rigorous and readable. | 20:35 |
| * serialhex thinks it's odd hes taking advice from brocolli... some of his friends might think he flipped bits :P | 20:36 | |
| serialhex | cool | 20:36 |
| brocolli | Well you're a statistician, right? | 20:36 |
| serialhex | i'm more like a guy who wants to make computers think like humans | 20:37 |
| brocolli | These shouldn't be too bad for you. Not as bad as it is when some CSers start out on their foray into ML, the mathematical nature can sometimes overwhelm. That's why sometimes going from bishop to dhs's book is good | 20:37 |
| brocolli | oooh boy ;P Don't say that! :P | 20:37 |
| serialhex | ok, maybe not like humans... but think on their own | 20:37 |
| serialhex | but i'm not afraid of math... i've been teaching myself calculus, and while it's ugly & not rigorous, it works (mostly) | 20:38 |
| brocolli | that's good enough in most cases. It seems like you're interested in *applications* of ML moreso. Don't get me wrong, people in robotics have good ML research but it's often focused on their Area. | 20:38 |
| brocolli | There are always general ML people, and more focused. For example Andrew Ng, Koller, Chappelle and those types do more general stuff that appleis to a larger scope | 20:39 |
| serialhex | interesting | 20:39 |
| brocolli | Then there are legends like Valiant, Vapnik, Bartlett (to be legend), Jordan (To be) who work on "learning theory" which is the theoretical aspect of ML. It essentially lays the "track" for many algorithms we use and appreciate today | 20:40 |
| serialhex | cool, i'll have to remember those names | 20:41 |
| serialhex | have you read Information Theory, Inference & Learning Algorithms by David MacKay?? i've started it and it's pretty interesting so far | 20:43 |
| brocolli | You'll have a fun time, just brave the initial investment and you will be very thankful for it. | 20:43 |
| serialhex | (and in my price range right now -- free!!) | 20:43 |
| brocolli | No sir, I have not. | 20:43 |
| serialhex | sweet! thanks | 20:43 |
| brocolli | Some books are available online if you google. NO I AM NOT ENCOURAGING BOOK PIRACY. | 20:44 |
| brocolli | coughcough google preview | 20:44 |
| serialhex | :P | 20:44 |
| @bettyboo | :> | 20:44 |
| serialhex | i know all about that | 20:44 |
| brocolli | "google preview" | 20:44 |
| * serialhex copies names dow | 20:44 | |
| brocolli | wink wink nudge nudge | 20:44 |
| -!- lionelc_ [48cb93da@gateway/web/freenode/ip.72.203.147.218] has joined #shogun | 20:44 | |
| brocolli | i'm itchy | 20:44 |
| brocolli | damn. | 20:44 |
| brocolli | no wonder i'm nudging and winking myself | 20:45 |
| serialhex | heh heh lol... | 20:45 |
| brocolli | less coffee | 20:45 |
| serialhex | ooh, redbull + 5 hour energy + ice == awesomeness incarnate!!! :P | 20:46 |
| @bettyboo | ;> | 20:46 |
| brocolli | hahaha | 20:46 |
| brocolli | nice. | 20:46 |
| @bettyboo | funny ;D | 20:46 |
| brocolli | which part :P | 20:46 |
| ameerkat | you didn't tell us if it evaluates to true or false though | 20:46 |
| -!- lionelc [48cb93da@gateway/web/freenode/ip.72.203.147.218] has quit [Ping timeout: 252 seconds] | 20:47 | |
| serialhex | very very true | 20:47 |
| brocolli | run a classifier and threshold on awesomeness > 0.5 | 20:47 |
| brocolli | in the decision space of [0,1] | 20:47 |
| serialhex | lol | 20:47 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Ping timeout: 252 seconds] | 20:47 | |
| serialhex | maybe i should have started: | 20:49 |
| serialhex | awesomeness_incarnate = redbull + 5_hour_energy + frozen_dihydrogenmonoxide | 20:49 |
| serialhex | err, stated | 20:49 |
| brocolli | weights of each of those values? | 20:50 |
| brocolli | uniform? | 20:50 |
| serialhex | well the three on the right are objects... and when they are summed you get the object on the left | 20:52 |
| brocolli | but what are the weights of each | 20:52 |
| brocolli | i.e. on [0,1] what values do you assign them such that \sum = 1 | 20:52 |
| serialhex | ahh... | 20:52 |
| serialhex | wow... tough Q | 20:52 |
| serialhex | :P | 20:52 |
| brocolli | It's like if I said I like rebull the most, 5 hour second, then: 0.5 0.3, 02 would satisfy the weights ;P | 20:53 |
| @bettyboo | HA :> | 20:53 |
| -!- hiteshk [~hitesh_na@117.211.88.150] has left #shogun [] | 20:53 | |
| serialhex | well, it'd have to be something like 0.35, 0.35, 0.30 | 20:53 |
| serialhex | because the ice just makes it that much better! | 20:54 |
| serialhex | though your weights would work just fine | 20:54 |
| * serialhex is glad to be back using linux | 20:55 | |
| brocolli | lol mine were an illustration. | 20:55 |
| @bettyboo | ;D | 20:55 |
| brocolli | I have ADHD so I don't drink energy drinks or coffee,giving away my coughcough but still | 20:56 |
| serialhex | i dont know if thats good or bad :P i know a number of people who have ADHD and yet still like coffee & energy drinks | 20:57 |
| @bettyboo | serialhex, hihi | 20:57 |
| serialhex | ...things get crazy when they start up on alot of it at once! | 20:57 |
| serialhex | bettyboo: hi, how are you? | 20:57 |
| @bettyboo | serialhex: however, there are some clever things one can do in chunking based methods like svmlight. | 20:57 |
| serialhex | bettyboo: really? tell me more about that | 20:58 |
| @bettyboo | serialhex: my guess is that you won't need all the time to do the newtonsvm if you take that one | 20:58 |
| brocolli | I take like 90 mg of concerta because I have combined type, hardcore ADHD. :P | 20:58 |
| @bettyboo | don't care | 20:58 |
| brocolli | I was "that kid" as a youngin' | 20:58 |
| brocolli | still got this far ;) | 20:58 |
| serialhex | ...this ought to be interesting | 20:58 |
| serialhex | wow.... | 20:58 |
| brocolli | bettyboo: I don't know why, but when I used svmlight for matlab, I felt it was limited. | 20:58 |
| @bettyboo | brocolli: though I don't know how to call functions etc, after require of some module... | 20:58 |
| @bettyboo | zzz | 20:59 |
| brocolli | iirc the toolbox can't really do real valued classification properly, maybe I read it wrong. | 20:59 |
| brocolli | Most of them deal with discrete classes, and I don't recall SVR being in it. SVR == win. | 20:59 |
| serialhex | SVR? scaable vector rockstars?? | 21:00 |
| brocolli | support vector regression | 21:00 |
| serialhex | ahh... cool | 21:00 |
| brocolli | It's the regression analog of SVMs. | 21:00 |
| * serialhex thinks it makes much more sense than his definition | 21:00 | |
| brocolli | It'll come in time ;) fear not padawan. | 21:01 |
| serialhex | yeah... this is the main reason i'm going back to college: for rigor | 21:01 |
| serialhex | ATM i know BUNCHES of stuff on different machine learning things, but not how they interact and work. or even how to implement them! | 21:02 |
| brocolli | Not many do though, people often just go "HAW YEAH GOOD RESULTS, GIMME" | 21:03 |
| serialhex | yeah... see while i like good results, it's not science unless it's repeatable & you can at least kind of explain why | 21:04 |
| brocolli | Ehhh even then it becomes tricky, because only fewer can do that | 21:04 |
| serialhex | (your explanation might be wrong, but thats for your grandkids to point out) | 21:04 |
| brocolli | See Michael Kearns. | 21:04 |
| brocolli | Legend in his own right. I'd say he's a person who does that, applies Machine Learning Theory to real world problems and gets results with rigor. | 21:05 |
| brocolli | But I only know of him at that level, where he can do exactly what you said. | 21:05 |
| serialhex | hmm... :-/ | 21:06 |
| brocolli | You can be in Machine Learning and map real world problems to the theory, but the mapping is difficult | 21:06 |
| brocolli | It requires a crazy amount of well roundedness | 21:06 |
| serialhex | something to aim for... :D | 21:07 |
| brocolli | Absolutely | 21:08 |
| brocolli | That's the only way. | 21:08 |
| brocolli | you achieve success | 21:08 |
| serialhex | ooh! doctoral thesis for free!! yet another thing i can read and only understand 20% of :P | 21:09 |
| @bettyboo | serialhex, ;D | 21:09 |
| serialhex | bettyboo: yes :P | 21:09 |
| @bettyboo | serialhex: yes of course - now I understand the reason why you are smiling. | 21:09 |
| serialhex | bettyboo: and are you smiling?? | 21:09 |
| @bettyboo | serialhex: no. you are not. something seems wrong. | 21:09 |
| serialhex | bettyboo: maybe i am! | 21:09 |
| @bettyboo | serialhex: do you know any further code for pca and covariance? | 21:09 |
| serialhex | no, but i do know: def fact(n); (n<=0) ? 1 : n* fact(n-1); end | 21:11 |
| serialhex | bettyboo: no, but i do know: def fact(n); (n<=0) ? 1 : n* fact(n-1); end | 21:11 |
| @bettyboo | serialhex: yeah .: I understood the code, but where are they used ?? Are they used inside shogun, or are they just for users ?? | 21:11 |
| serialhex | bettyboo: users are so over rated!! and i dont think that is used inside shogun | 21:11 |
| @bettyboo | serialhex: I think we shoud just require("shogun") | 21:11 |
| serialhex | bettyboo: yes! you read my mind!!! | 21:12 |
| @bettyboo | serialhex: the wrapper code is shogun::CGaussianKernel | 21:12 |
| CIA-31 | shogun: Soeren Sonnenburg master * re437a5f / src/libshogun/lib/HDF5File.cpp : replace H5Gopen/create also with the *2 functions - http://bit.ly/i0DaTS | 21:24 |
| -!- Ziyuan [~Ziyuan@116.22.48.72] has quit [] | 21:27 | |
| -!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 240 seconds] | 21:29 | |
| blackburn | serialhex, brocolli, guys, you really discussed more than there was all time before :) | 21:33 |
| serialhex | lol... yeah there dosnt seem to be much going on in this channel ever | 21:33 |
| @sonney2k | brocolli, there is SVR in shogun | 21:34 |
| @sonney2k | and also real-valued features | 21:34 |
| brocolli | Darn! The powers that be caught me talking >< | 21:34 |
| brocolli | noooo! | 21:34 |
| @sonney2k | look at CSimpleFeatures / CSparseFeatures | 21:35 |
| @sonney2k | these are templated classes though | 21:35 |
| @sonney2k | so they work with anything from bytes to 128bit floats | 21:35 |
| @sonney2k | (or more) | 21:35 |
| brocolli | I don't do enough ML in C++, largely due to the frustration of no matrices. I know you can use blas, but blas has ugly prototypes ;P | 21:35 |
| -!- chlin [8161e1b0@gateway/web/freenode/ip.129.97.225.176] has joined #shogun | 21:35 | |
| -!- sonney2k is now known as big_brother | 21:35 | |
| brocolli | I hope I didn't speak too much ;) | 21:35 |
| blackburn | big brother is watchin' you, hahha | 21:36 |
| brocolli | lol | 21:36 |
| blackburn | oh, I have to feed my bear and drink vodka | 21:36 |
| brocolli | ML books are pricey for many people :) | 21:36 |
| blackburn | brocolli: do you really buy all the books you read? | 21:37 |
| -!- big_brother is now known as sonney2k | 21:37 | |
| brocolli | Usually yes, because going to class doesn't help me learn. Often the materials online are weak unless they're from berkeley.edu, stanford.edu, etc ;P | 21:37 |
| brocolli | And sometimes I like more explanation and less brevity | 21:38 |
| brocolli | and I can't learn off an online pdf :p | 21:38 |
| blackburn | :) may be I am a 'thief', but even doesn't have any bought textbook about ML :) | 21:39 |
| @bettyboo | blackburn, ;D | 21:39 |
| brocolli | http://img12.imageshack.us/img12/8234/img2011040600012.jpg | 21:39 |
| brocolli | this is my shelf | 21:39 |
| blackburn | frankly, there some shared books, like ESL or book of McKay | 21:39 |
| blackburn | spendin' all the money for books? :) | 21:40 |
| brocolli | The Wiley series books are unparalleled | 21:40 |
| brocolli | For statistical theory, there is nothing better. period. | 21:40 |
| brocolli | I have PDFs for almost all of them, and then I see how well written they and I buy them because I know how hard it is to write a good textbook while maintaining rigor | 21:41 |
| chlin | A question: has there been much interest in the implementation of kernel feature analysis? | 21:41 |
| * serialhex is sad that his bookshelf is not nearly as cool | 21:41 | |
| blackburn | chlin: yeap | 21:41 |
| chlin | I've only gotten around to reading the paper on it and trying to understand the maths | 21:41 |
| brocolli | You'll get there. | 21:41 |
| brocolli | Getting comfortable with ML concepts most definitely does not happen overnight. | 21:41 |
| brocolli | For me at least. | 21:41 |
| serialhex | yeah... it's taken me a while to get to this point | 21:42 |
| brocolli | In my experience, to be "great", you will always feel that way | 21:42 |
| brocolli | That's my motivation, to keep learning and familiarizing myself with it | 21:42 |
| blackburn | btw, guys, that ideas you come to? | 21:42 |
| serialhex | erm... i'm having touble parsing your last statement blackburn :P | 21:44 |
| @bettyboo | ;> | 21:44 |
| blackburn | sorry | 21:44 |
| blackburn | not native in english :) | 21:45 |
| blackburn | I mean what ideas you have choosen to send a proposal? | 21:45 |
| brocolli | I think he is asking what items we picked | 21:45 |
| brocolli | I'm not telling, I don't wanna be identified :P My writing style already gives me away! | 21:45 |
| brocolli | ;) | 21:45 |
| blackburn | I feel ashamed now | 21:45 |
| @bettyboo | brocolli, he? :> | 21:45 |
| brocolli | she | 21:45 |
| brocolli | sorry | 21:45 |
| brocolli | I didn't know | 21:45 |
| blackburn | she? | 21:46 |
| serialhex | it's alright blackburn... | 21:46 |
| brocolli | I don't know, she said "he?" | 21:46 |
| brocolli | and I said she because she asked "he?" | 21:46 |
| brocolli | Was my original assumption correct? | 21:46 |
| blackburn | I don't know what that bot is telling all the time.. :) | 21:46 |
| brocolli | Why do you feel ashamed? | 21:46 |
| serialhex | yeah, that bot is kinda confusing | 21:47 |
| blackburn | brocolli: of my bad english :) | 21:47 |
| brocolli | pfft | 21:47 |
| @bettyboo | lol | 21:47 |
| brocolli | My friend, in the end, all that matters is what you want to proveand how you show it | 21:47 |
| brocolli | And that requires one language: MATH | 21:47 |
| serialhex | ...i'm looking to do the modular ruby typemaps & stuff... nothing too sporty | 21:47 |
| * serialhex agrees with brocolli | 21:47 | |
| blackburn | so, anyway, what's up with ideas? | 21:47 |
| blackburn | as for me, I have only one proposal: dimensionality reduction algos | 21:48 |
| serialhex | sounds interesting | 21:48 |
| blackburn | particulary isomap, mds, lle and sne | 21:48 |
| * serialhex has no clue what a dimensionality reduction algorithm is :P will learn sometime | 21:49 | |
| blackburn | serialhex: it's all about finding a suitable low-dimensional data | 21:49 |
| blackburn | with high-dimensional available | 21:49 |
| blackburn | it is important in case of low performance, could serve with cleaning data, etc | 21:50 |
| blackburn | at least there is always a curse of dimensionality | 21:50 |
| serialhex | hmm... sounds interesting | 21:51 |
| brocolli | curse of dimensionality is a problem that often plagues MLers. | 21:52 |
| brocolli | We try to find ways to fix it | 21:52 |
| blackburn | I don't know why you aren't saying anything about your ideas :) | 21:53 |
| serialhex | i would imagine so... if you have 100 dimensions & not all of them are needed, then you'll have to sift through a bunch of 100 dimensional arrays all over the place... way slow!! | 21:54 |
| blackburn | serialhex: not only sift it, but have to find a new features in another feature space | 21:55 |
| blackburn | like in MDS, there is an optimization problem | 21:55 |
| serialhex | blackburn: yeah... thats nuts!!! | 21:55 |
| blackburn | so, yes, it is a bit slowly, but there is always a gain | 21:55 |
| blackburn | sometimes at performance, sometimes at quality | 21:55 |
| serialhex | cool, | 21:56 |
| brocolli | because the ideas I picked aren't necessarily the coolest, and I dunno if I get a GSoC position if I will accept it just yet. There is a huge time commitment (40 hours) which I am going to think about. | 21:56 |
| blackburn | my idea isn't the coolest too :) | 21:57 |
| blackburn | no problem, anyway | 21:57 |
| blackburn | one guy already sent 11 proposals, sounds very funny at all | 21:57 |
| serialhex | wow!! | 21:57 |
| blackburn | I asked if they are very different and they aren't | 21:58 |
| blackburn | I don't know what implies this way | 21:59 |
| blackburn | seems to be a spam ) | 21:59 |
| brocolli | 11 | 21:59 |
| brocolli | that sounds unrealistic. | 21:59 |
| brocolli | That's so many | 21:59 |
| brocolli | How is gonna finish all of em | 21:59 |
| blackburn | I said, they aren't very different at all | 21:59 |
| serialhex | well you can only be accepted for 1 for the summer... | 22:00 |
| @sonney2k | yes indeed | 22:00 |
| @sonney2k | and I guess we will get <10 slots | 22:00 |
| blackburn | big brother is watching again | 22:00 |
| brocolli | How many applicants have you got so far? | 22:00 |
| blackburn | sonney2k: I hope more than 5? | 22:00 |
| @sonney2k | blackburn, you can almost add a zero to that nr. | 22:01 |
| blackburn | is it only depends on google? | 22:01 |
| @sonney2k | well we have 13 mentors, lets say 10 of them can take a student so we could max do 10 | 22:02 |
| @sonney2k | then we are in for the first time | 22:02 |
| @sonney2k | so google will not give us all the slots | 22:02 |
| @sonney2k | (though we will be trying) | 22:03 |
| blackburn | I mean will you request some or google will choose itself? | 22:03 |
| serialhex | so each mentor is only going to have 1 student?? | 22:03 |
| @sonney2k | usually yes | 22:03 |
| serialhex | hmm :-/ | 22:03 |
| blackburn | what's the problem with it | 22:03 |
| blackburn | ? | 22:03 |
| -!- warpy [~warpy@bzq-79-182-10-204.red.bezeqint.net] has quit [Ping timeout: 248 seconds] | 22:04 | |
| @sonney2k | blackburn, ? we have to read and rank the students | 22:04 |
| @sonney2k | err proposals | 22:04 |
| blackburn | sonney2k: I said it to serialhex | 22:04 |
| * sonney2k is tired | 22:04 | |
| blackburn | I mean there is no problem with 1 mentor-1 student | 22:04 |
| blackburn | is there is 13 mentors | 22:04 |
| blackburn | sonney2k: it is almost equal if all of proposals are similar :) | 22:05 |
| @bettyboo | lol ;> | 22:05 |
| @sonney2k | the google gsoc faq says that the more students apply for an organization the more student-slots they get | 22:05 |
| serialhex | hmm... but that means tha ~1/5 of the students are going to be accepted | 22:05 |
| @sonney2k | serialhex, that is the google average btw | 22:05 |
| serialhex | wow | 22:05 |
| * serialhex really hopes his application rocks enough!! | 22:06 | |
| -!- lionelc_ [48cb93da@gateway/web/freenode/ip.72.203.147.218] has quit [Ping timeout: 251 seconds] | 22:06 | |
| blackburn | sonney2k: is there an opportunity to change application when already sent? | 22:06 |
| @sonney2k | IIRC they had 5500 proposals last year and about 1100 that got accepted | 22:07 |
| blackburn | especially after friday | 22:07 |
| @sonney2k | blackburn, not after friday I think | 22:07 |
| blackburn | could it be 'locked'? | 22:07 |
| @sonney2k | it will be hard for us to do a fair ranking... believe me | 22:08 |
| blackburn | oh, already April, 7th | 22:08 |
| blackburn | not in your countries, I think :) | 22:09 |
| blackburn | sonney2k: it is a challenge for you, I see | 22:09 |
| @sonney2k | at some point you have to make a decision though... but surely if we think we have 2 or more students applying for the same project (that both are really good) we will ask them if they could imagine to do alternative projects... | 22:09 |
| serialhex | that sounds cool | 22:10 |
| blackburn | sonney2k: could you say how many students already applied to Chris' idea? | 22:10 |
| blackburn | just for information :) | 22:10 |
| @bettyboo | hihi | 22:10 |
| @sonney2k | I don't know | 22:11 |
| blackburn | ok, thought all the proposals going to you | 22:12 |
| @sonney2k | they are all different anyways... | 22:12 |
| @sonney2k | blackburn, well the interface is not perfect | 22:13 |
| blackburn | sonney2k: is it necessary to introduce a method in CKNN for classify using rank weight function? for example q^(k-i), 0<q<1 | 22:14 |
| @sonney2k | blackburn, sorry but my brain is no longer working | 22:15 |
| blackburn | no problem :) | 22:15 |
| * sonney2k goes brushing his teeth and then takes a nap | 22:15 | |
| serialhex | mmm... nap | 22:16 |
| @sonney2k | l8r guys, big brother turns on the surveillance cameras awaiting good news | 22:16 |
| blackburn | sonney2k: I'll send you a patch with description to email, okay? | 22:16 |
| -!- lionelc [48cb93da@gateway/web/freenode/ip.72.203.147.218] has joined #shogun | 22:16 | |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has joined #shogun | 22:17 | |
| @sonney2k | blackburn, to the ML? | 22:17 |
| blackburn | may be straightly to you? | 22:17 |
| blackburn | or better to ML? | 22:17 |
| -!- dvevre [b49531e3@gateway/web/freenode/ip.180.149.49.227] has quit [Ping timeout: 252 seconds] | 22:39 | |
| brocolli | sonney2k: I can see it being difficult, but your questions essentially ask us about how much theory we know combined with the practical | 22:39 |
| brocolli | I mean, there are only so many good ML researchers with that kinda background who could do all of those problems, and then the number of students would be like, maybe 20 times that? I can't imagine there being more than 400 of us | 22:40 |
| brocolli | maybe 5 | 22:40 |
| -!- hiteshk [~hitesh_na@117.211.88.150] has joined #shogun | 22:41 | |
| -!- josip [~josip@unaffiliated/josip] has joined #shogun | 23:00 | |
| -!- hiteshk [~hitesh_na@117.211.88.150] has left #shogun [] | 23:13 | |
| -!- makokal [~nyumbanil@212.201.44.245] has left #shogun [] | 23:55 | |
| --- Log closed Thu Apr 07 00:00:36 2011 | ||
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!
