--- Log opened Tue May 03 00:00:42 2011 | ||
-!- alesis-novik [~alesis@188.74.87.84] has quit [Remote host closed the connection] | 00:34 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 01:43 | |
-!- alesis-novik [~alesis@188.74.87.84] has joined #shogun | 03:44 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 276 seconds] | 03:51 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 04:25 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 260 seconds] | 09:44 | |
-!- blackburn [~qdrgsm@188.168.5.65] has joined #shogun | 09:46 | |
-!- mikiobraun [~mikio@squid.ml.tu-berlin.de] has joined #shogun | 10:28 | |
-!- blackburn [~qdrgsm@188.168.5.65] has left #shogun [] | 10:38 | |
-!- mikiobraun [~mikio@squid.ml.tu-berlin.de] has quit [Quit: Leaving.] | 12:00 | |
alesis-novik | Good day | 14:56 |
---|---|---|
-!- vojta [~quassel@2001:718:2:1634:221:6aff:fe58:151a] has joined #shogun | 15:05 | |
-!- vojta is now known as VojtechFranc | 15:05 | |
VojtechFranc | hi alesis | 15:06 |
alesis-novik | Hello VojtechFranc | 15:06 |
VojtechFranc | I prepared Matlab implementation of the GMM-EM algorithm plus some examples. | 15:07 |
alesis-novik | can you send it over email? I'll have a look at it | 15:08 |
VojtechFranc | It is a small Matlab toolbox. you can install it from SVN: svn co http://ptak.felk.cvut.cz/datagrid/svn/stprtool/ | 15:09 |
VojtechFranc | username is stprtool, no pasword | 15:09 |
VojtechFranc | the algorithms relevant to your work are in the folder "probab" | 15:09 |
VojtechFranc | there are functions for deadling with the GMM and the EM algorithm itself | 15:10 |
VojtechFranc | there are also few examples of usage on systhetical 1d and 2d data | 15:10 |
VojtechFranc | in addition there is one example using the GMM-EM algorithm to train OCR classifier on USPS database | 15:11 |
alesis-novik | I'll have a look at it then. | 15:11 |
VojtechFranc | Great! Do not hasiate to ask questions. | 15:12 |
VojtechFranc | Curently there is a basic implementation of the EM which you shoul re-implement to C++. | 15:13 |
VojtechFranc | There a few differneces compared to textbook implementaion of GMM-EM. | 15:13 |
VojtechFranc | 1) the covariance matrix is represented in decomposed form which is uself fo several reasons (speed, numerical stability, ...) | 15:14 |
VojtechFranc | 2) Most of the computations are done in log domain which is more robust | 15:15 |
VojtechFranc | 3) there is a kind of regularization (namely, bound on the minimal variance) which makes the log-likelihood problem well posed | 15:15 |
VojtechFranc | In the time before you start implementing, you could try to get familiar with the code. | 15:17 |
alesis-novik | so log domain deals with high/low numbers and regularization deals with singular Gaussians then? | 15:18 |
VojtechFranc | Do you need to recomend some literature on EM ? I think it would be necessary only if you are not familiar with the EM | 15:18 |
VojtechFranc | yes, exactly | 15:19 |
VojtechFranc | if we have time, I'd like you to implement other imporvements: like merge/split rules etc. | 15:20 |
alesis-novik | I've studied EM from C. M. Bishops "Pattern Recognition and Machine Learning", so I think I know the general theory behind it | 15:20 |
VojtechFranc | yes, this should be enough | 15:21 |
VojtechFranc | you can ask me if you find something strange in the code | 15:22 |
VojtechFranc | but it should not be difficult | 15:22 |
alesis-novik | merge/split would be dynamically changing number of components? | 15:22 |
VojtechFranc | exactly | 15:22 |
VojtechFranc | it is useful to avoid some "bad" local maxima of the log-likelihood | 15:23 |
VojtechFranc | and to remove gaussian components not supported by enough data | 15:23 |
alesis-novik | So it would be a more robust way of dealing with singular Gaussians than minimal variance as well | 15:24 |
VojtechFranc | No, it does not improve robustness. | 15:25 |
VojtechFranc | It makes the model more meaningfull (by removing the singular components) and helps to escape some local maxima | 15:25 |
alesis-novik | I think I should have time to implement that | 15:27 |
VojtechFranc | it would be nice. However, let us proceed gradually: i) implementing the basic version, ii) testing/ debuggin and iii) implementing other improvements | 15:29 |
alesis-novik | well, I guess the *very* basic version is already there | 15:29 |
VojtechFranc | by basic I mean the one I sent you | 15:30 |
VojtechFranc | it would be also nice if you implement some demosntation scripts | 15:31 |
VojtechFranc | both toy examples and close to some real application | 15:31 |
VojtechFranc | this will require some graphical output | 15:32 |
VojtechFranc | do you prefer Matlab to Python or something else? | 15:32 |
alesis-novik | I think I would prefer Matlab, though I can't really use it from home. I'm not really that good at Python though. | 15:33 |
VojtechFranc | great, I also prefer Matlab | 15:34 |
alesis-novik | Like I said, the only problem is that I can't really do Matlab because I can only use it in the University | 15:35 |
VojtechFranc | Then decide what is better for you. I can use Python as well. | 15:36 |
alesis-novik | I assume I can do the demonstration scripts at the end, after doing the implementation? | 15:38 |
VojtechFranc | Yes, but it is useful to have at least some simple example which you can used during development | 15:39 |
VojtechFranc | like some synthetic data where you can visualized the found model to see whether the estimate makes a sense ... | 15:40 |
VojtechFranc | the code I sent you contains such examples | 15:40 |
alesis-novik | I wonder if R could be used for this? | 15:41 |
VojtechFranc | no idea, I know nothing about R. | 15:41 |
VojtechFranc | As I said I would prefer i) Matlab or ii) Python or iii) Octave | 15:42 |
VojtechFranc | So, that's it from my side. I'll have to disconnect soon. | 15:43 |
VojtechFranc | Please, go though the code and don't hasitate to send me an email if you meet any problem. | 15:44 |
alesis-novik | Hmm, Octave looks quite similar to Matlab, and given that it's not going to be used for computation, just visualization, it might port to Matlab easily | 15:44 |
alesis-novik | So the implementation should also have an input parameter where you can specify if the cov is "full" "diag" or "spherical"? | 15:45 |
VojtechFranc | I thing the same holds for Matlab and python - they will be used just for visualization | 15:45 |
VojtechFranc | the code will be in C++ | 15:45 |
VojtechFranc | in the shogun library linked to the corresponding environment | 15:45 |
alesis-novik | But I can't really use Matlab from where I'll be developing this and I'm not too comfortable with Python. I think I'll deal with this issue when time comes | 15:46 |
VojtechFranc | ok, think about other alternatives | 15:47 |
alesis-novik | so about the covariance types, I should use the 3 used in the Matlab code, correct? | 15:48 |
VojtechFranc | for now yes, in the future I'd like to ask other options lake the Factor Analysis Decomposition etc | 15:48 |
VojtechFranc | it sould not be problem to add other representations if the code is well implemented | 15:49 |
alesis-novik | I'll think about separating different parts so they can be expanded on later | 15:50 |
VojtechFranc | great, the code I sent you might give you an inspiration beause it allows simple extensions | 15:51 |
VojtechFranc | unfortunatelly, I need to go now. Shell we agree on another chat (besides emails) ? | 15:52 |
alesis-novik | I think we can agree on it over the emails. I'll probably only have time to analyse the Matlab code during the next 2 weeks because of my exams | 15:53 |
VojtechFranc | yes, of course. Let's communicate via emails. | 15:54 |
alesis-novik | When the actual coding starts we could have weekly talks or so | 15:54 |
VojtechFranc | yes, sure. | 15:54 |
VojtechFranc | So, I wish you succesful examination period. | 15:55 |
alesis-novik | Thanks. 4 exams to go. | 15:55 |
VojtechFranc | then good luck! | 15:56 |
VojtechFranc | bye | 15:56 |
@bettyboo | VojtechFranc, back to real-life? | 15:56 |
alesis-novik | See you | 15:56 |
VojtechFranc | bettyboo, sorry don't know who is hiding behind this nice nickname | 15:57 |
@bettyboo | VojtechFranc: most however try to improve kpca in terms of robustness and not run-time performance | 15:57 |
VojtechFranc | bettyboo, I concentrate on the EM becasue KPCA doesn't seem to be very practical algorithm | 15:59 |
@bettyboo | VojtechFranc: Going to discuss details of EM with Vojtech tomorrow | 15:59 |
alesis-novik | A bot that is now saying scary relevant things | 15:59 |
VojtechFranc | battyboo, sorry have to disconnect now. See you | 16:00 |
-!- VojtechFranc [~quassel@2001:718:2:1634:221:6aff:fe58:151a] has quit [Remote host closed the connection] | 16:00 | |
alesis-novik | bettyboo, you scare me sometimes | 16:00 |
@bettyboo | alesis-novik: now I am already scared of having to wade through all of them | 16:00 |
-!- alesis-novik [~alesis@188.74.87.84] has quit [Quit: I'll be Bach] | 16:02 | |
-!- blackburn [~qdrgsm@188.168.2.173] has joined #shogun | 17:44 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has joined #shogun | 18:12 | |
-!- ameerkat [~ameerkat@184-98-140-155.phnx.qwest.net] has quit [Ping timeout: 260 seconds] | 18:33 | |
-!- blackburn [~qdrgsm@188.168.2.173] has left #shogun [] | 20:33 | |
-!- blackburn [~qdrgsm@188.168.2.173] has joined #shogun | 21:49 | |
@sonney2k | So Blackburn you are now the one to take over my job | 22:57 |
blackburn | sonney2k: oh.. | 22:57 |
blackburn | sonney2k: your baby is born? | 22:57 |
@sonney2k | Yes exactly | 22:58 |
blackburn | sonney2k: congrats! | 22:58 |
@sonney2k | Sleeping next to me currently. | 22:58 |
@sonney2k | Thanks. | 22:59 |
blackburn | sonney2k: is it a boy or girl? | 22:59 |
@sonney2k | Boy | 23:00 |
blackburn | nice, did you chose a name? | 23:00 |
@sonney2k | Yes Jan | 23:00 |
blackburn | nice! congrats to you again and pass congrats to your wife from cold russia :) | 23:01 |
@bettyboo | :Q blackburn | 23:01 |
@sonney2k | I am still full of adrenaline so cannot sleep. | 23:01 |
blackburn | sonney2k: I'll try to take some your work | 23:02 |
@sonney2k | Thanks ... I hope we can leave the hospital this week then I also have my computer plus kbd back | 23:05 |
@sonney2k | Anyways good night | 23:06 |
blackburn | sonney2k: no problem, I will just react on some things there will be | 23:07 |
blackburn | sonney2k: good night! let you have some sleep | 23:07 |
--- Log closed Wed May 04 00:00:43 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!