CODEWITHSUNDEEP
pythonnumpyscipy
FooBar
FooBar

Reputation: 16518

Vectorize scipy.integrate.quad

I am using scipy.integrate.quad(f, a, b, args=(c,)) to integrate function f between a and b, adding another parameter c. You can think about f as

f = lambda x, y: x*y

As I understand it, the function requires a, b, c to be float values. However, I have many integrals to solve, all with the same function, just over vectors of a, b, and c. Iterating through all the vectors is quite inefficient. Is there any way to speed up / vectorize this operation?

Upvotes: 4

Views: 3858

Answers (3)

Nikolai Zaitsev
Nikolai Zaitsev

Reputation: 323

It looks that meanwhile numpy has responded to this by implementing numpy.vectorize with an example covered in integrate. Here is modified example with two variables:

fun2int = lambda x, a: np.sqrt(x+a)
intfun = lambda a: integrate.quad(fun2int, 0, 4, args=(a))[0]
vec_int = np.vectorize(intfun)
vec_int(np.linspace(0,2,5))
Out[1]: array([5.33333333, 6.12825877, 6.78689326, 7.37435069, 7.91234089])

There is also integrate.quad_vec (scipy.version of 1.4.1), but I did not test that.

Upvotes: 3

ev-br
ev-br

Reputation: 26090

Alas, this is not supported by integrate.quad.

You might find better luck with using gaussian quadratures yourself: scipy.integra.fixed_quad chokes on array_like input, but the lower-level routines which compute roots and weights should work. See "Roots and weights for orthogonal polynomials" section here. YMMV though.

Upvotes: 3

Fabio Lamanna
Fabio Lamanna

Reputation: 21574

If I understood well the problem, I would just run the function over an array of a,b,c parameters. For instance, set an array of two sets of a,b,c as:

a = np.array([[0.1,0.2,0.3], [0.9,0.8,0.3]])

then set an empty list to store results:

L = []

and loop through the array:

for a,b,c in a:

    b = integrate.quad(f, a, b, args=(c,))
    L.append(b)

The result is a list of tuples:

[(0.004500000000000001, 4.996003610813206e-17), (-0.025499999999999995, 2.8310687127941485e-16)]

Maybe there is a more elegant way, but hope that helps.

Upvotes: 0

Related Questions