missing multiplication operator between float and unumpy.matrix in one direction about uncertainties

[lebigot]

Thank you for the heads up. I'll look into this missing feature.

EOL

On Thu, 16 Jun 2011, seibert wrote:

One direction of the multiplication operator between a float and a unumpy matrix seems to be missing:

>>> from uncertainties import unumpy, ufloat
>>> m = unumpy.matrix([[ufloat((1.0, 0.2))]])
>>> m * 0.5
matrix([[0.5+/-0.1]], dtype=object)
>>> 0.5 * m
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for *: 'float' and 'matrix'

However, this works just fine for a normal numpy matrix:

>>> import numpy
>>> m = numpy.matrix([[1.0]])
>>> m * 0.5
matrix([[ 0.5]])
>>> 0.5 * m
matrix([[ 0.5]])

[lebigot]

0.5*m uses NumPy's dot function. The core of the problem seems to be the following non-symmetrical behavior of NumPy's dot functions (either from numpy.core.multiarray or from the optimized numpy.core._dotblas, which are the two possible sources for numpy.core.numeric.dot()):

>>> m2 = numpy.matrix([uncertainties.ufloat((1, 0.1))])
>>> numeric.dot(3, m2)
NotImplemented
>>> numeric.dot(m2, 3)
matrix([[3.0+/-0.3]], dtype=object)

Both work, for a matrix m2 of floats, but not for a matrix of objects. I am not sure that this behavior of NumPy is intended. I'll investigate this.

[lebigot]

3*unumpy.matrix(…) works, in the new uncertainties version 1.7.3. I went around the apparent limitation in NumPy's dot() function. I'll contact NumPy's maintainers about the issue. Thank you for the heads up!

[seibert]

Great, thanks!