cumf / cumf_ccd Goto Github PK

## How to build  

make clean && make

## How to run  

$./ccdp_gpu toy_example 
(more examples below)

##Input format

The input should be in binary format (See toy_example directory
for details). An example to create binaries can be found in https://github.com/cuMF/cumf_als/blob/master/data/ml10M/prepare_ml10M_data.py

"meta_modified_all" has the name of the input files.

Details:

line 1 has #rows #cols
line 2 has nnz in training dataset
line 3 has nnz in test dataset
line 4 to 12 has file names


## Arguments

    $ ./ccdp_gpu [options] [input file directoty containing meta_modified_all] 

    options:

        -k rank/feature : set the rank (default 10)
        -l lambda : set the regularization parameter lambda (default 0.05)
        -a tile size: set tile size for input matrix R (default 499999999)
        -b tile size: set tile size for input matrix R Transpose (default 499999999)
        -t max_iter: number of iterations (default 5)
        -T max_iter: number of inner iterations (default 1)


## Examples:
To run Netflix:
./ccdp_gpu -T 1 -a 100000 -b 100000 -l .058 -k 40 -t 10 ../Netflix 

To run Yahoo Music:
$./ccdp_gpu -T 1 -a 100000 -b 100000 -l 1.2 -k 40 -t 10 ../Yahoo_music