Passion is like genius; a miracle.

http://en.wikipedia.org/wiki/Partial_least_squares_regression PCA 와 유사하지만 데이터의 분산만을 잡으려는 PCA와 달리 Y값까지 함께 고려해 orthogonal한 새로운 feature와 response들을 X, Y로 부터 만들고, 이로부터 수행하는 linear regression. 데이터에 비해 변수가 많을때 OLS(Ordinary Least Square)에 비해 유용함. Tutorial: http://en.wikipedia.org/wiki/Partial_least_squares_regression 책: http://www.maths.bath.ac.uk/~jjf23/LMR/

Training time becomes an issue, esp., if someone is using complicated model like svm with huge data. When that happens, it’s critical to estimate when the training will end. I’ve written a simple R code to do that efficiently (in terms of time but not in terms of accuracy). Here’s the graph:

http://www.johndcook.com/distribution_chart.html 확률분포간의 관계. 정말 잘 정리되어있네요.

Here’s output.

If you have trouble because of RGtk2 (e.g., it’s saying you don’t have gtk2.8.0 or higher while you have one already). If so, follow this procedure. (Note: Don’t use macports! It won’t work with RGtk2 properly.) Instead, install R from r-project.org. This is pretty much important. Don’t install R from macports. Install R from http://www.r-project.org/.…

Random projection in dimensionality reduction: Applications to image and text data This is really easy way of dimensionality reduction. Simply, multiply data with random matrix where is a random number from . If is a dxN dimension where d is very high dimension and N is the number of data and is a kxd dimension…

It’s the Effect Size, Stupid http://en.wikipedia.org/wiki/Effect_size Statistical significance를 사용하여 두 그룹간의 차이를 검증할때에는 두 그룹간의 차이뿐만 아니라 샘플의 크기에도 영향을 받습니다. 왜냐하면 샘플의 크기가 커질수록 confidence interval 이 줄어들기 때문입니다. 또 어찌되었던지간에 statistical significance라는건 두 그룹간에 차이가 실제로는 없는데 우연히 차이가 발생할 확률로부터 계산하는데 이 값은 두 그룹간의 차이가 얼마나 큰가를 직접적으로 평가하는 것이 아닙니다.…

Pretty much readable document on SVM: A User’s Guide to Support Vector Machines. It’s a doc posted on python based machine learning tools: PyML.