Parallel and Efficient Testing

Numba Community Support
1

I am currently working in jupyter lab and want to do a quick assertion in order to validate that a function is working as expected. Here is the recreation of what I am doing:

np.random.seed(0)

@njit(parallel=True)
def sum_shape(repeats,samples):
    
    sum = 0
    
    for i in prange(0,repeats):
        x = np.random.uniform(0,100,int(samples))
        sum += np.sum(x)
        
    return sum

print(sum_shape(100,1000))
    
time1 = time.time() - start1
    
print(time1)

%%ipytest

np.random.seed(0)

def test():
    assert sum_shape(100,1000) == 5015855.398294773 # expected output given seed

The problem I currently face is with how parallelization works, pulling numbers from different seeds per thread, so I am wondering how to combat this. I want to have the test still compare to an expected output while being efficient.

Thanks.