[Numpy-discussion] poly1d left versus right multiplication with np numbers

[josef.pktd at gmail.com]

I just had a hard to find bug in my program. poly1d  treats numpy
scalars differently than python numbers when left or right
multiplication is used.

Essentially, if the first term is the numpy scalar, multiplied by a
polynomial, then the result is an np.array.
If the order is reversed, then the result is an instance of np.poly1d.
The return types are also the same for numpy arrays, which is at least
understandable, although a warning would be good)

When using plain (python) numbers, then both left and right
multiplication of the number with the polynomial returns a polynomial.

Is this a bug or a feature? I didn't see it mentioned in the docs.

My problem for debugging was that in the examples I used python
numbers while the program used numpy scalars, and it took me a while
to figure out that this is the source of my bugs.

examples below

Josef



>>> polys
[poly1d([1]), poly1d([-1.,  0.]), poly1d([ 1.,  0., -1.]), poly1d([
1.,  0., -3.,  0.]), poly1d([ 1.,  0., -6.,  0.,  3.])]

np.array on left is fine
>>> (polys[2]*np.array(0.5/6.0) + polys[3]*np.array(0.5/24.0))
poly1d([ 0.02083333,  0.08333333, -0.0625    , -0.08333333])
>>> (polys[2]*0.5/6.0 + polys[3]*0.5/24.0)
poly1d([ 0.02083333,  0.08333333, -0.0625    , -0.08333333])
>>> (0.5/6.0*polys[2] + 0.5/24.0*polys[3])
poly1d([ 0.02083333,  0.08333333, -0.0625    , -0.08333333])

problems with np.array on left

>>> (np.array(0.5/6.0)*polys[2] + np.array(0.5/24.0)*polys[3])
Traceback (most recent call last):
  File "<pyshell#722>", line 1, in <module>
    (np.array(0.5/6.0)*polys[2] + np.array(0.5/24.0)*polys[3])
ValueError: shape mismatch: objects cannot be broadcast to a single shape
>>> np.array(0.5/6.0)*polys[2]
array([ 0.08333333,  0.        , -0.08333333])
>>> polys[2]*np.array(0.5/6.0)
poly1d([ 0.08333333,  0.        , -0.08333333])
>>> 0.5/6.0*polys[2]
poly1d([ 0.08333333,  0.        , -0.08333333])
>>> np.array(0.5/6.0)
array(0.083333333333333329)

same with numpy scalar

>>> np.array([0.5/6.0])[0]*polys[2]
array([ 0.08333333,  0.        , -0.08333333])
>>> np.array([0.5/6.0])[0]
0.083333333333333329