weights.py

def weight_proc(matrix, name, k, diag=0, checks=False):

    if checks:
        if not all(matrix[i][j] == matrix[j][i] for i, neighbors in matrix.neighbors.items() for j in neighbors):
            print("Matrix is not symmetric!")

    full_matrix, ids = matrix.full()
    np.fill_diagonal(full_matrix, diag)

    # Convert to a sparse matrix
    sparse_matrix = csr_matrix(full_matrix)

    # Convert to WSP (sparse weights) and then to W
    wsp = WSP(sparse_matrix)
    w = wsp.to_W()

    if checks:
        # Sanity checks
        # Negative weights?
        if any(w_ij < 0 for neighbors in w.weights.values() for w_ij in neighbors):
            print("There are negative weights!")

        # Isolated rows?
        isolated_rows = [
            i for i, neighbors in w.neighbors.items() if not neighbors]
        if isolated_rows:
            print(f"There are {len(isolated_rows)} isolated rows.")

        # Diagonal elements one?
        full_w_matrix, _ = w.full()
        if not all(np.diagonal(full_w_matrix) == 1):
            print("Not all diagonal elements are ones!")

    # Save as NEW kernel matrix
    with open(name + "_" + k + "_Proc_Diag_"+diag+".pkl", 'wb') as f:
        pickle.dump(w, f)

    return w.full()


with open('gaussian_2.pkl', 'rb') as f:
    kernel = pickle.load(f)

    weight_proc(kernel, "gaussian", "2", diag=0, checks=False)
    weight_proc(kernel, "gaussian", "2", diag=1, checks=False)