Drowning

Today, I was almost drowned during swim class. We moved out from a practice pool which is sallow to deep one (~ 2.5m) to try 'floating' (i.e. stay still in one spot and breath. Is it vertical floating? I don't what is a correct word for that).

Since this is my first time to learn how to swim, I have never tried to float before as well.
At the first time, I grabbed the wall and tried to float. After a few minutes, I released the wall in order to see if I could float.. But, you know, of course I was sinking into the water. Actually, I expected that so I was not scared.

But... I expected that I could reach the bottom and kick it to come back to the surface.. But... no.. I was failed to reach the bottom and running out of my breath just inside the water..

After about 10(?) seconds of struggling to reach the bottom, I almost ran out of my breath and actually drank water through my nose and mouse.. I think I was about to be drowned :)

Anyway, fortunately, just at that time (after drank some water), I reached the bottom (thanks god...) and could come back to the surface and grab the wall again.

Interestingly, I was not in panic at all even when I drank a lot of water due to lack of oxygen in my lung. I was almost calm.. hm...., which means... what?? Maybe I'm that much dull? haha..


Lesson : Stick to the stairs at the wall if you want to try float. At least you can grab the stairs and climb.

카디건 산 산행, 보스톤 (Mt. Cardigan, Boston)

몸도 영 찌뿌둥하고 한 차에..보스톤 산악회에서 산행을 한다기에 냅다 쫓아갔다 왔다.

뭔 산행 중에 과제가... 살짝 귀찮긴 했으나 ㅡ.ㅡ;;
처음 나간거니 '되도록' 열심히 ㅋㅋ

같은 조 소속이셨던 '캔두' 님. 하산하면서 말벗이 되주셔서 심심하지 않았다.
집으로 놀러오라시던데.. 언제 불러주시려나 :)

마지막 정리하는 사진. 산에서는 날씨가 너무 안좋아서 시야가 전혀 없더니..
하산하니 해가 쨍쨍... 모래알은 반짝...

여기부터는 내 카메라 사진 몇장

영.. 전부 안개

그나마 산 아랫자락에서는 앞은 보였음.




산 정상 부근.. 아무것도 안보인다 0_0..


정년퇴임하시고 올해 일흔이 되신다는 캔두님. 한발 한발 쉬지않고 올라가시는 모습.


산행만을 기대하고 갔었지만, 좋은 사람들도 만나 더 값졌던 토요일 하루.

My Macbook Pro and iPod touch.

I really like them :)

What a amazing.

Different colors and symbols mean different variability types. The majority of the biggest cluster above (blue one) might be non-variable stars, even though the cluster is overlapped with tons of variable stars.

With these kind of indice (I have multiple indice derived from light-curves), I used a machine learning technique to classify each of the variability. And, a preliminary result is pretty amazing.. It can classify each of class with very high accuracy.

I also have totally different kind of indices I didn't implement yet, so if I use those indice, I guess our algorithm could beat any of previous algorithms with a factor of (at least) ~2. For a specific variability type, maybe the factor would be bigger than 5 or 10.


Nevertheless, I'm exciting now hehe :)

Swim class

I just starts to learn. Well yeah, I don't know how to swim :)
Today was the first class, and... wow.. I'm full now... drank too much water..

update #1. now I can swim from one side to other side without breathing and with only kicking.. well.. because I do not know how to stroke or how to breath T_T..
Anyway, it's still hard for me to finish one-way trip :)

Support Vector Machine

There are several python packages which provide SVM.

LIBSVM, MLPY, PyML, SVM^python

LIBSVM has been included in a lot of other packages to provide SVM. It looks like very powerful library for SVM. However, it is not 'perfectly' and 'clearly' wrapped with python and doesn't provide other classification method except SVM.

PyML seems simple and easy to use, but looks like it is in its beginning stage of development.

So far, MLPY looks most 'useful' package for python users. It provides multiple kernels for SVM and also provides KNN or other classification function. Moreover, it has a function which can calculate accuracy of prediction. Not bad at all. (Actually, LIBSVM has almost identical SVM modules too.)

But, I didn't compare their performance (especially for LIBSVM and MLPY) or didn't do real test yet. I'll update the results soon.

You can download all the packages above from internet. Do Google :)

P.S The above image is from Google image.

update #1: Shit... MLPY looks like not supporting multiclass-classification for SVM..

update #2. I was wrong. PyML has almost all function for SVM. It uses LIBSVM for SVM and also provide KNN function. It just doesn't have a good tutorial document :)

update #3. well.. I'm going to use LIBSVM even though it doesn't provide good python wrapper. I'm implementing a python routine for parameter set search and modified version of cross_validation of LIBSVM.

Python ctypes and Cython

http://cython.org/

Looks promising, especially Cython. I'm going to test both of them.

Update #1. After I tested ctypes with some C shared object which calculates periods of time series, I found ctypes is about 5% slower than pure C. Well, I need to test more, however, it looks very good so far. Only 5% :)

TAOS Delta Scuti Stars

Detected from 2-year TAOS data. Complete list is here :

http://timemachine.iic.harvard.edu/coati/delta/list.html

The paper is published in AJ.

Dae-Won Kim, Pavlos Protopapas, Charles Alcock, Yong-Ik Byun, and et al.

http://www.iop.org/EJ/abstract/1538-3881/139/2/757/

TAOS Flare Stars

Dae-Won Kim, Pavlos Protopapas, Yong-Ik Byun, Charles Alcock, and et al.

http://adsabs.harvard.edu/abs/2009ATel.2035....1K

(This is not a paper.)

Pain in the ass

OK OK.. It's all my fault to upgrade to Snow Leopard, and to be persistent to use 64bit python2.6..

BTW.. Even though how hard I try, I cannot compile matplotlib.. How much hours I have wasted for that.. Numpy, Scipy works just fine after installing them using easy_install.

but.. bloody matplotlib.. shit...

Do I have to use python 2.5 or.. just wait until they correct problems..?

update #1 : There is one solution to install matplotlib.

http://blog.hyperjeff.net/?p=160

Though..

update #2 : OK, fink (new version for Snow Leopard) looks like working well. I'm installing 64 bit python2.6 and all other packages. So far, so good. Nevertheless, Iit's a little bit strange because my experience of fink at Leopard (OS X 10.5) was not that good.

update #3 : R is not ready yet because gcc43 is not available on Snow Leopard. R should change its dependency from gcc43 to gcc44.. hm..