python numba fingerprint error

Question

I'm attempting numba to optimise some code. I've worked through the initial examples in section 1.3.1 in the 0.26.0 user guide (http://numba.pydata.org/numba-doc/0.26.0/user/jit.html) and get the expected results, so I don't think the problem is installation.

Here's my code:

import numba
import numpy
import random

a = 8
b = 4

def my_function(a, b):
    all_values = numpy.fromiter(range(a), dtype = int)
    my_array = []
    for n in (range(a)):
        some_values = (all_values[all_values != n]).tolist()
        c = random.sample(some_values, b)
        my_array.append(sorted([n] + c))
    return my_array

print(my_function(a, b))

my_function_numba = numba.jit()(my_function)

print(my_function_numba(a, b))

Which after printing out the expected results from the my_function call returns the following error message:

ValueError                                Traceback (most recent call last)
<ipython-input-8-b5d8983a58f6> in <module>()
     19 my_function_numba = numba.jit()(my_function)
     20 
---> 21 print(my_function_numba(a, b))

ValueError: cannot compute fingerprint of empty list

Fingerprint of empty list?

Answer 1

I'm not sure about that error in particular, but in general, to be fast numba requires a particular subset of numpy/python (see here and here for more). So I might rewrite it like this.

@numba.jit(nopython=True)
def fast_my_function(a, b):
    all_values = np.arange(a)
    my_array = np.empty((a, b + 1), dtype=np.int32)
    for n in range(a):
        some = all_values[all_values != n]
        c = np.empty(b + 1, dtype=np.int32)
        c[1:] = np.random.choice(some, b)
        c[0] = n
        c.sort()
        my_array[n, :] = c
    return my_array

Main things to note:

1. no lists, I'm pre-allocating everything.
2. no use of generators (in both python 2 & 3 for n in range(a) will get converted to a fast native loop)
3. adding nopython=True to the decorator makes it so numba will complain if I use something that can't be efficiently JITed.