More than one module for lambdify in sympy

Question

I am trying to make lambdify understand to expect more than one type of input using the modules keyword argument. According to the source code of lambdify (http://docs.sympy.org/dev/_modules/sympy/utilities/lambdify.html), this can be done by using lists of the arguments, but i am not able to do so.

import sympy
from sympy import lambdify
x,y=sympy.symbols('x y')
from sympy.parsing.sympy_parser import parse_expr
func=lambdify(x,parse_expr(exp(x)),modules=["numpy","sympy"])

func(array([3,4]))

gives

array([ 20.08553692,  54.59815003])

but when i try

func(y)

i get an

Attribute error:exp

What am i doing wrong here? Shouldn't func accept both numpy and sympy types? Any help appreciated!!

Answer 1

The modules don't dispatch or anything like that. The way that lambdify works is that it creates

lambda x: exp(x)

where exp comes from the namespace of the module(s) you chose. lambdify(x, exp(x), ['numpy', 'sympy']) is roughly equivalent to

from sympy import *
from numpy import *
# Various name replacements for differences in numpy naming conventions, like
# asin = arcsin
return lambda x: exp(x)

If you want to provide a custom function that dispatches, you can use something like Saullo Castro's example. You can also use this with lambdify by providing a dict, like

import numpy as np
import sympy

def myexp(x):
    if isinstance(x, np.ndarray):
        return np.exp(x)
    else:
        return sympy.exp(x)

func = lambdify(x, exp(x), [{'exp': myexp}, 'numpy'])

This gives

>>> func(np.array([1, 2]))
array([ 2.71828183,  7.3890561 ])
>>> func(sympy.Symbol('y'))
exp(y)