Build on LUMI

1. Clone this repository to a directory: git clone https://github.com/cscjlan/qcqubo.git
2. cd qcqubo
3. Create a directory for binaries: mkdir bin
4. Run scripts/build.sh on e.g a login node. This builds an executable qubo in the bin/ directory.

Run on LUMI

First, use the data/inputs.json to specify the inputs to the program. Specifically:

• seed: what random seed to use
• num_to_search: how many random binary vectors to generate and test
• matrix_filename: the .csv file containing the matrix, either in csr format or in "dense" format, i.e. containing all the elements, including the zero elements
• output_filename: where to output the best binary vector

After you're satisfied with the inputs, use sbatch -A project_YOUR_PROJECT_NUMBER scripts/lumi_run.sh to queue the program, where YOUR_PROJECT_NUMBER is a number of the project you want the billing of the used resources to go to.

Input matrix

The matrix can be given as a dense .csv file, in which case the zeroes should be included.

If it's given as a csr matrix, the .csv file should contain three rows (for a N x N matrix):

1. indptr
2. indices
3. data.

See the SciPy csr matrix for the meaning of the names.

Given a SciPy csr matrix csr_mat, the following Python snippet writes the matrix in the correct format for this program:

    with open(f"data/csr_matrix.csv", "w") as f:
        csr_mat.indptr.tofile(f, sep=",")
        f.write("\n")
        csr_mat.indices.tofile(f, sep=",")
        f.write("\n")
        csr_mat.data.tofile(f, sep=",")