shotgun

shotgun copied to clipboard

REFERENCE

Joseph K. Bradley, Aapo Kyrola, Danny Bickson, and Carlos Guestrin (2011). "Parallel Coordinate Descent for L1-Regularized Loss Minimization." International Conference on Machine Learning (ICML 2011). http://arxiv.org/abs/1105.5379

INSTALLATION

1. For running as a mex code called from Matlab Just run:

     make

2. For running as a C application, using MatrixMarket input format:

     make cversion_debug

OR

     make cversion_release

Current build is only tested on Linux, so you might need to modify the Makefile to suit your system.

COST FUNCTION

We use the following cost function formulation. For Lasso: argmin_x sum_i [(A_ix - y_i)^2 + lambda * |x|_1] For sparse logistic regression: argmin_x sum_i [-log(1 + exp(-y_i * x A_i) ) + lambda * |x|_1]

where |x|_1 is the first norm (sum of absolute value of the vector x).

USAGE

Do not call the mex-library directly. Instead use the provided Matlab-scripts shotgun_logreg.m and shotgun_lasso.m.

Both have same signature:

     shotgun_logreg(A,y,lambda)
     shotgun_lasso(A,y,lambda)

They return the optimized feature/weight-vector. For tuning the parameters, please modify the scripts. A more user-friendly options-passing will be provided later.

For an example, see example/ directory.

2. RUNNING AS A STANDALONE C PROGRAM: Matrix and vector files are mandaroty inputs

Usage: ./mm_lasso
	-m matrix A in sparse matrix market format
	 -v vector y in sparse matrix market format
	 -o output file name (will contain solution vector x, default is
x.mtx)
	 -a algorithm (1=lasso, 2=logitic regresion, 3 = find min lambda for
all zero solution)
	 -t convergence threshold (default 1e-5)
	 -k solution path length (for lasso)
	 -i  max_iter (default 100)
	 -n num_threads (default 2)
	 -l lammbda - positive weight constant (default 1)
	 -V verbose: 1=verbose, 0=quiet (default 0) 

REMARKS

Provided code is not exactly same as the one we used for running our experiments for the ICML 2011 paper. Particularly this code runs slower sequentially, because special code for running with only one cpu has been removed for clarity. Parallel code needs to do some extra work compared to sequential algorithm, and therefore for fairness we had special versions for the sequential tests.

This version uses OpenMP for parallel execution. For the paper, we used CILK++. Since CILK is not as widely available as OpenMP, we decided to switch for the source code release.

MORE INFO

This code is is not well tested and you should not rely on it on any mission-critical tasks.

For bug-fixes or other questions, please contact Aapo Kyrola : [email protected] or Danny Bickson: [email protected]