I suspect it might be related to fact that rank of this matrix is 2 (np.linalg.matrix_rank(your_matrix)), which means that there are some redundant features.