g-factor is a matrix factorization implementation with parallel stochastic gradient descent on AMD or NVIDIA GPUs. It uses HIP and CUDA languages.

These are the base requirements to build and use g-factor:

• POSIX-standard shell
• GNU-compatible Make
• G++ compiler with C++11 support
• HIP or CUDA toolkit

make
cd bin
./mf -t train-file [-e test-file] [-c config-file] [-d gpu-id]

-t train-file: Train data file in the COO format. The first line is a (number-of-users, number-of-items, number-of-ratings) tuple, where the elements are divided by spaces/tabs. Each following line is a (user-id, item-id, rating) tuple, where the elements are divided by spaces/tabs. Comment lines start with '%'s.

[-e test-file]: Test data file in the COO format. The first line is a (number-of-users, number-of-items, number-of-ratings) tuple, where the elements are divided by spaces/tabs. Each following line is a (user-id, item-id, rating) tuple, where the elements are divided by spaces/tabs. Comment lines start with '%'s.

[-c config-file]: Configuration file. Comment lines start with '#'s.

[-d gpu-id]: GPU used to accelerate computation. If skipped, CPU is used.