Source code for nipy.algorithms.utils.matrices

# emacs: -*- mode: python; py-indent-offset: 4; indent-tabs-mode: nil -*-
# vi: set ft=python sts=4 ts=4 sw=4 et:
""" Utilities for working with matrices """
from __future__ import absolute_import

import numpy as np
import scipy.linalg as spl

[docs]def matrix_rank(M, tol=None): ''' Return rank of matrix using SVD method Rank of the array is the number of SVD singular values of the array that are greater than `tol`. This version of matrix rank is very similar to the numpy.linalg version except for the use of: * scipy.linalg.svd istead of numpy.linalg.svd. * the MATLAB algorithm for default tolerance calculation ``matrix_rank`` appeared in numpy.linalg in December 2009, first available in numpy 1.5.0. Parameters ---------- M : array-like array of <=2 dimensions tol : {None, float} threshold below which SVD values are considered zero. If `tol` is None, and `S` is an array with singular values for `M`, and `eps` is the epsilon value for datatype of `S`, then `tol` set to ``S.max() * eps * max(M.shape)``. Examples -------- >>> matrix_rank(np.eye(4)) # Full rank matrix 4 >>> I=np.eye(4); I[-1,-1] = 0. # rank deficient matrix >>> matrix_rank(I) 3 >>> matrix_rank(np.zeros((4,4))) # All zeros - zero rank 0 >>> matrix_rank(np.ones((4,))) # 1 dimension - rank 1 unless all 0 1 >>> matrix_rank(np.zeros((4,))) 0 >>> matrix_rank([1]) # accepts array-like 1 Notes ----- We check for numerical rank deficiency by using ``tol=max(M.shape) * eps * S[0]`` (where ``S[0]`` is the maximum singular value and thus the 2-norm of the matrix). This is one tolerance threshold for rank deficiency, and the default algorithm used by MATLAB [#2]_. When floating point roundoff is the main concern, then "numerical rank deficiency" is a reasonable choice. In some cases you may prefer other definitions. The most useful measure of the tolerance depends on the operations you intend to use on your matrix. For example, if your data come from uncertain measurements with uncertainties greater than floating point epsilon, choosing a tolerance near that uncertainty may be preferable. The tolerance may be absolute if the uncertainties are absolute rather than relative. References ---------- .. [#1] G. H. Golub and C. F. Van Loan, _Matrix Computations_. Baltimore: Johns Hopkins University Press, 1996. .. [#2] http://www.mathworks.com/help/techdoc/ref/rank.html ''' M = np.asarray(M) if M.ndim > 2: raise TypeError('array should have 2 or fewer dimensions') if M.ndim < 2: return int(not np.all(M==0)) S = spl.svd(M, compute_uv=False) if tol is None: tol = S.max() * np.finfo(S.dtype).eps * max(M.shape) return np.sum(S > tol)
[docs]def full_rank(X, r=None): """ Return full-rank matrix whose column span is the same as X Uses an SVD decomposition. If the rank of `X` is known it can be specified by `r` -- no check is made to ensure that this really is the rank of X. Parameters ---------- X : array-like 2D array which may not be of full rank. r : None or int Known rank of `X`. r=None results in standard matrix rank calculation. We do not check `r` is really the rank of X; it is to speed up calculations when the rank is already known. Returns ------- fX : array Full-rank matrix with column span matching that of `X` """ if r is None: r = matrix_rank(X) V, D, U = spl.svd(X, full_matrices=0) order = np.argsort(D) order = order[::-1] value = [] for i in range(r): value.append(V[:,order[i]]) return np.asarray(np.transpose(value)).astype(np.float64)
[docs]def pos_recipr(X): """ Return element-wise reciprocal of array, setting `X`>=0 to 0 Return the reciprocal of an array, setting all entries less than or equal to 0 to 0. Therefore, it presumes that X should be positive in general. Parameters ---------- X : array-like Returns ------- rX : array array of same shape as `X`, dtype np.float, with values set to 1/X where X > 0, 0 otherwise """ X = np.asarray(X) return np.where(X<=0, 0, 1. / X)
[docs]def recipr0(X): """ Return element-wise reciprocal of array, `X`==0 -> 0 Return the reciprocal of an array, setting all entries equal to 0 as 0. It does not assume that X should be positive in general. Parameters ---------- X : array-like Returns ------- rX : array """ X = np.asarray(X) return np.where(X==0, 0, 1. / X)