Getting started

import sweepystats as sw
import numpy as np

The SweepMatrix class

SweepMatrix is a thin wrapper over numpy darrays. The input must be a symmetric matrix. We convert all matrices into double-precision numpy arrays with column-major order, if they aren’t already.

# intantiate a SweepMatrix from symmetric input
A_numpy = np.array([[1., 2, 3],
                    [2, 5, 6],
                    [3, 6, 9]],
                   order='F') # column-major format
A = sw.SweepMatrix(A_numpy)
A

SweepMatrix(array([[1., 2., 3.],
       [2., 5., 6.],
       [3., 6., 9.]]))

NOTE: No data is copied if the input is already double-precision in column-major format. The latter requirement is because the sweeping operation is blessed by level-3 BLAS, which we call internally.

# modifying entries of A also changes the original
A[0, 0] = 10
A_numpy

array([[10.,  2.,  3.],
       [ 2.,  5.,  6.],
       [ 3.,  6.,  9.]])

A SweepMatrix can be swept forward and backwards:

A.sweep(verbose=False) # forward sweep through all diagonal entries of A
A

SweepMatrix(array([[-1.11111111e-01, -4.85722573e-17,  3.70370370e-02],
       [-4.85722573e-17, -1.00000000e+00,  6.66666667e-01],
       [ 3.70370370e-02,  6.66666667e-01, -5.67901235e-01]]))

A.sweep(inv=True, verbose=False) # inverse sweep recovers the original data
A

SweepMatrix(array([[10.,  2.,  3.],
       [ 2.,  5.,  6.],
       [ 3.,  6.,  9.]]))

We can also sweep a on the kth diagonal element

A.sweep_k(1) # sweep the kth diagonal
A

SweepMatrix(array([[ 9.2,  0.4,  0.6],
       [ 0.4, -0.2,  1.2],
       [ 0.6,  1.2,  1.8]]))

A.sweep_k(1, inv=True) # unweep the kth diagonal
A

SweepMatrix(array([[10.,  2.,  3.],
       [ 2.,  5.,  6.],
       [ 3.,  6.,  9.]]))