PyPy vs CPython, DerApproximator results

from DerApproximator import *
from numpy import sin, cos, arctan, arange
from time import time
N = 1000

#from _numpypy.pypy import set_invalidation
#set_invalidation(False)

vectorized_1 = lambda x: sin(x**2).sum() + cos(1-x**2).sum()

def nonvectorized_1(x):
    r = 0.0
    for i in range(N):
        r += 0.1 * r + 0.2 * sin(x[i]) * cos(0.2+x[i]) * arctan(15-x[i])
    return r

print('\n1st derivatives: ')
for name, N, f in [
                ('vectorized_1', 1000, vectorized_1),
                ('nonvectorized_1', 100, nonvectorized_1)
                ]:
    t = time()
    r = get_d1(f, arange(float(N)))
    print('%s: time elapsed for N = %d: % 0.1f' % (name,  N, time()-t))

print('2nd derivatives: ')
for name, N, f in [
                ('vectorized_1', 50, lambda x: (x**2).sum()),
                ('nonvectorized_1', 20, nonvectorized_1)
                ]:
    t = time()
    r = get_d2(f, arange(float(N)))
    print('%s: time elapsed for N = %d: % 0.1f' % (name,  N, time()-t))

'''
CPython results:
1st derivatives:
vectorized_1: time elapsed for N = 1000:  1.3
nonvectorized_1: time elapsed for N = 100:  3.1
2nd derivatives:
vectorized_1: time elapsed for N = 50:  4.7
nonvectorized_1: time elapsed for N = 20:  10.8

PyPy results
1st derivatives:
vectorized_1: time elapsed for N = 1000:  7.1 # slower
nonvectorized_1: time elapsed for N = 100:  1.0
2nd derivatives:
vectorized_1: time elapsed for N = 50:  5.3
nonvectorized_1: time elapsed for N = 20:  3.8

PyPy results with
from _numpypy.pypy import set_invalidation
set_invalidation(False)

1st derivatives:
vectorized_1: time elapsed for N = 1000:  3.8
nonvectorized_1: time elapsed for N = 100:  0.9
2nd derivatives:
vectorized_1: time elapsed for N = 50:  3.2
nonvectorized_1: time elapsed for N = 20:  2.7
'''