A Python implementation of NNLS algorithm

lsqnonneg.py

"""
A Python implementation of NNLS algorithm

References:
[1]  Lawson, C.L. and R.J. Hanson, Solving Least-Squares Problems, Prentice-Hall, Chapter 23, p. 161, 1974.

Based on MATLAB's lsqnonneg function
Ported to Python by Klaus Schuch ([email protected]) and contributed by myself (especially unit tests).
"""

import numpy

def lsqnonneg(C, d, x0=None, tol=None, itmax_factor=3):
    '''Linear least squares with nonnegativity constraints.

    (x, resnorm, residual) = lsqnonneg(C,d) returns the vector x that minimizes norm(d-C*x)
    subject to x >= 0, C and d must be real
    '''

    eps = 2.22e-16    # from matlab
    def norm1(x):
        return abs(x).sum().max()

    def msize(x, dim):
        s = x.shape
        if dim >= len(s):
            return 1
        else:
            return s[dim]

    if tol is None: tol = 10*eps*norm1(C)*(max(C.shape)+1)

    C = numpy.asarray(C)

    (m,n) = C.shape
    P = numpy.zeros(n)
    Z = numpy.arange(1, n+1)

    if x0 is None: x=P
    else:
        if any(x0 < 0): x=P
        else: x=x0

    ZZ = Z

    resid = d - numpy.dot(C, x)
    w = numpy.dot(C.T, resid)

    outeriter=0; it=0
    itmax=itmax_factor*n
    exitflag=1

    # outer loop to put variables into set to hold positive coefficients
    while numpy.any(Z) and numpy.any(w[ZZ-1] > tol):
        outeriter += 1

        t = w[ZZ-1].argmax()
        t = ZZ[t]

        P[t-1]=t
        Z[t-1]=0

        PP = numpy.where(P != 0)[0]+1
        ZZ = numpy.where(Z != 0)[0]+1

        CP = numpy.zeros(C.shape)

        CP[:, PP-1] = C[:, PP-1]
        CP[:, ZZ-1] = numpy.zeros((m, msize(ZZ, 1)))

        z=numpy.dot(numpy.linalg.pinv(CP), d)

        z[ZZ-1] = numpy.zeros((msize(ZZ,1), msize(ZZ,0)))

        # inner loop to remove elements from the positve set which no longer belong
        while numpy.any(z[PP-1] <= tol):
            it += 1

            if it > itmax:
                max_error = z[PP-1].max()
                raise Exception('Exiting: Iteration count (=%d) exceeded\n Try raising the \
                                 tolerance tol. (max_error=%d)' % (it, max_error))

            QQ = numpy.where((z <= tol) & (P != 0))[0]
            alpha = min(x[QQ]/(x[QQ] - z[QQ]))
            x = x + alpha*(z-x)

            ij = numpy.where((abs(x) < tol) & (P != 0))[0]+1
            Z[ij-1] = ij
            P[ij-1] = numpy.zeros(max(ij.shape))
            PP = numpy.where(P != 0)[0]+1
            ZZ = numpy.where(Z != 0)[0]+1

            CP[:, PP-1] = C[:, PP-1]
            CP[:, ZZ-1] = numpy.zeros((m, msize(ZZ, 1)))

            z=numpy.dot(numpy.linalg.pinv(CP), d)
            z[ZZ-1] = numpy.zeros((msize(ZZ,1), msize(ZZ,0)))

        x = z
        resid = d - numpy.dot(C, x)
        w = numpy.dot(C.T, resid)

    return (x, sum(resid * resid), resid)


# Unittest
if __name__ == '__main__':
    C = numpy.array([[0.0372, 0.2869],
                     [0.6861, 0.7071],
                     [0.6233, 0.6245],
                     [0.6344, 0.6170]])

    C1 = numpy.array([[0.0372, 0.2869, 0.4],
                      [0.6861, 0.7071, 0.3],
                      [0.6233, 0.6245, 0.1],
                      [0.6344, 0.6170, 0.5]])

    C2 = numpy.array([[0.0372, 0.2869, 0.4],
                      [0.6861, 0.7071,-0.3],
                      [0.6233, 0.6245,-0.1],
                      [0.6344, 0.6170, 0.5]])

    d = numpy.array([0.8587, 0.1781, 0.0747, 0.8405])

    [x, resnorm, residual] = lsqnonneg(C, d)
    dres = abs(resnorm - 0.8315)          # compare with matlab result
    print('ok, diff:', dres)
    if dres > 0.001:
        raise Exception('Error')

    [x, resnorm, residual] = lsqnonneg(C1, d)
    dres = abs(resnorm - 0.1477)          # compare with matlab result
    print('ok, diff:', dres)
    if dres > 0.01:
        raise Exception('Error')

    [x, resnorm, residual] = lsqnonneg(C2, d)
    dres = abs(resnorm - 0.1027)          # compare with matlab result
    print('ok, diff:', dres)
    if dres > 0.01:
        raise Exception('Error')

    k = numpy.array([[0.1210, 0.2319, 0.4398, 0.9342, 0.1370],
                     [0.4508, 0.2393, 0.3400, 0.2644, 0.8188],
                     [0.7159, 0.0498, 0.3142, 0.1603, 0.4302],
                     [0.8928, 0.0784, 0.3651, 0.8729, 0.8903],
                     [0.2731, 0.6408, 0.3932, 0.2379, 0.7349],
                     [0.2548, 0.1909, 0.5915, 0.6458, 0.6873],
                     [0.8656, 0.8439, 0.1197, 0.9669, 0.3461],
                     [0.2324, 0.1739, 0.0381, 0.6649, 0.1660],
                     [0.8049, 0.1708, 0.4586, 0.8704, 0.1556],
                     [0.9084, 0.9943, 0.8699, 0.0099, 0.1911]])

    k1 = k-0.5

    l = numpy.array([0.4225, 0.8560, 0.4902, 0.8159, 0.4608, 0.4574, 0.4507, 0.4122, 0.9016, 0.0056])

    [x, resnorm, residual] = lsqnonneg(k, l)
    dres = abs(resnorm - 0.3695)          # compare with matlab result
    print('ok, diff:', dres)
    if dres > 0.01:
        raise Exception('Error')

    [x, resnorm, residual] = lsqnonneg(k1, l)
    dres = abs(resnorm - 2.8639)          # compare with matlab result
    print('ok, diff:', dres)
    if dres > 0.01:
        raise Exception('Error')

    C = numpy.array([[1.0, 1.0],
                     [2.0, 1.0],
                     [5.0, 1.0],
                     [6.0, 1.0],
                     [10.0, 1.0]])

    d = numpy.array([3, 5, 11, 13, 21])

    [x, resnorm, residual] = lsqnonneg(C, d)

    print([x, resnorm, residual])

Thank you for acknowledging the error. I wish one day robots will do something else than losing the developer's time with convincing but wrong results.