PASSCoDe: Parallel ASynchronous Stochastic dual Co-ordinate Descent

The Program

PASSCoDe implements the multi-core parallel DCD algorithms, which aims to solve large-scale linear Support Vector Machines.

Download

PASSCoDe is developed under the code base of LIBLINEAR with multi-core parallelization using OpenMP.

Download the zip-file and extract the files. On a UNIX system with GCC 4.7.2 or above, compile the program using the provided Makefile

> make

Data set Preparation

[Usage]: convert2binary training_set_file [training_binary]
	

> ./convert2binary dataset dataset.cbin

Usage

[Usage]: train-shrinking [options] training_set_file test_set_file
options:
-s type : set type of solver (default 31)
        31 -- L2-regularized L2-loss support vector classification PASSCoDe-Wild (dual)
        33 -- L2-regularized L1-loss support vector classification PASSCoDe-Wild (dual)
        41 -- L2-regularized L2-loss support vector classification PASSCoDe-LOCK (dual)
        43 -- L2-regularized L1-loss support vector classification PASSCoDe-LOCK (dual)
        51 -- L2-regularized L2-loss support vector classification PASSCoDe-ATOMIC (dual)
        53 -- L2-regularized L1-loss support vector classification PASSCoDe-ATOMIC (dual)
-c cost : set the parameter C (default 1)
-n nr_threads : the number of threads
-t max_iterations: the max number of iterations (default 100)
-b binary_mode : if binary_mode = 1, read binary format (default 1)
	

Citation

Please acknowledge the use of the code with a citation.

@inproceedings{cjh15akk,
  title ={{PASSCoDe}: {P}arallel {AS}ynchronous {S}tochastic dual {Co}-ordinate {De}scent},
  author={Cho-Jui Hsieh and Hsiang-Fu Yu and Inderjit S. Dhillon},
  booktitle = {International Conference of Machine Learning},
  year = {2015}
}