Long time, no post, but now I have news to share!
First, I just merged a PR to make SymPy's
ufuncify create actual instances of
numpy.ufunc. This function takes in a tuple of arguments, and an expression,
and returns a binary function that will broadcast (apply the function to each
argument in turn) the arguments through the function, returning an array. An
example would be:
from sympy import * from sympy.utilities.autowrap import ufuncify import numpy as np # Create an example expression a, b, c = symbols('a, b, c') expr = sin(a) + cos(b**2)*c # Create a binary (compiled) function that broadcasts it's arguments func = ufuncify((a, b, c), expr) func(np.arange(5), 2.0, 3.0)
array([-1.96093086, -1.11945988, -1.05163344, -1.81981085, -2.71773336])
Previously all broadcasting was done using hardcoded loops. These limited arguments to:
Now, through the magic of
This makes this functionality incredibly more useful. My next step is to add the ability for multiple outputs, and a custom wrapper so that matrix calculations can be broadcast as well. This should help with repeated computation of matrices, which is something Jason needs for his work. This should be done (hopefully) by the end of next week.
I gave a talk yesterday on code generation in SymPy for our local Python User Group. Discusses the why and how of the code generation tool chains we developed, and gives a little demo. Slides are here, and the corresponding demo here.
School recently started back up. In an effort to keep myself on task, I've started doing AIOs:
I've been hosting them on github in an effort to be open about my work, and update them every Friday. So far it seems to be a good idea - at the end of each week I get a chance to reflect on what I did that week, and what I plan to do over the next week. As a plus, my non-existant internet readers get to hold me accountable to my to-do list :).