I've noticed that results change rapidly between running timeit with the regular arugments and timeit with the argument number=1. Here is the example that I found.
import sympy as sp
import numpy as np
var = (sp.Symbol("x"), sp.Symbol("y"), sp.Symbol("z"))
x,y,z = var[0], var[1], var[2]
monos = np.array([1, z, y, x, z**2, y*z, y**2, x*z, x*y, x**2, z**3, y*z**2, y**2*z, y**3, x*z**2, x*y*z, x*y**2, x**2*z, x**2*y, x**3, z**4, y*z**3, y**2*z**2, y**3*z, y**4, x*z**3, x*y*z**2, x*y**2*z, x*y**3, x**2*z**2, x**2*y*z, x**2*y**2, x**3*z, x**3*y, x**4])
f = sp.lambdify(var, monos)
import testit
timeit.timeit("(f(2,3,4))", setup="from __main__ import f")
#returns 2.0760600566864014
timeit.timeit("(f(2,3,4))", setup="from __main__ import f", number=1)
#returns 1.5974044799804688e-05
I'm using python2 version 2.7.12 and sympy version 1.2. What is going on here?