※ 引述《ChihJen ( )》之銘言：
: Very good.. This is closer to what we estimated.
: Compared to rafan's estimate of using C,
: your result clearly shows the power of optimized BLAS.
: I think for the next homework you are allowed to use any tool
: you like. This makes your analysis easier.
: However, I do hope some of you use tools other than matlab.
: For example, if you already have a simple C code
: for reading the data, you can try to link optimized BLAS
: directly or use ATLAS.
: Except intel's optimized BLAS, you can try "goto's BLAS,"
: which may be even faster:
: http://www.tacc.utexas.edu/~kgoto/
: It is also interesting to see the performance
: difference on 32/64 bit machines.
: I also hope someone could study how to let R use optimized
: BLAS.
There are some articles on Internet about R and optimized BLAS.
An assistant professor in UW said linking R with an optimized BLAS
won't help too much. One reason is that R don't spend much time in
BLAS. Another reason is that R use only level-one routines since
most optimization was done in level-three routines.
: ※ 引述《anguso (愛寶)》之銘言：
: : For cross-validation, Matlab may need about 3.3767 hours.
: : And for test the whole 61012 test data, it would take about
: : 1 hour.
: : Significant difference from R.......