search

pbreheny / biglasso

biglasso: Extending Lasso Model Fitting to Big Data in R
http://pbreheny.github.io/biglasso/
113 stars 29 forks source link

Running biglasso is much slower than before #15

Closed privefl closed 6 years ago

privefl commented 6 years ago

Before, with my CentOS 7.2, it was very fast. Now, I've upgraded to 7.4. I also tried with Linux Mint and with another computer.

Every run of biglasso (or my function that has a very similar code) is 1.5-10 times slower (sometimes more).

Have you any idea what could cause this huge drop in performance? A missing library? Some difference in compiler optimizations?

This prevents me from finishing my simulations for a paper, any insight would be MUCH appreciated.

YaohuiZeng commented 6 years ago

Hmm, that's strange. seems to have something to do with your compile environment. The quick thing top of my head is gcc compiler flag. Did you set any optimization before and not now? Or using a different compiler? Honestly, I don't have too much knowledge about compiler.

privefl commented 6 years ago

This is my best guest also. I wonder if this comes from a change in handling inlining by new compilers (I think inlining is important for the C++ matrix accessor of big.matrix objects). I'm making some tests and search and come back to you if I find a solution to this problem.

If anyone has another idea, please let me know.

privefl commented 6 years ago

Seems to be due to the sparse matrix accessor (beta) being much slower.

I reported this issue there.