Benchmark of building an array with numba

Here I try out different ways of making arrays of tuples with Numba where the number of tuples that need to be produced is not known in advance. The fastest way is to make a list of tuples, and then convert this into a Numpy array before returning. The baseline shown in the last plot is the fastest possible way of pre-allocating a Numpy array with the right size upfront, but this cannot be used if the number of tuples is not known in advance.

nb.__version__='0.53.1'
ak.__version__='1.1.2'

array([[0.25473852, 0.47593451, 0.04606011, 0.55805331],
       [0.77287874, 0.52879803, 0.91944361, 0.98775869],
       [0.01221509, 0.30608935, 0.90908168, 0.07629862]])

26.1 µs ± 1.51 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

array([[0.40883519, 0.14577046, 0.20275019, 0.24327066],
       [0.7325605 , 0.76214408, 0.47819782, 0.93920595],
       [0.24872141, 0.92187207, 0.6691103 , 0.5643187 ]])

36.9 µs ± 1.55 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

array([[0.13088925, 0.25692562, 0.48062596, 0.39696723],
       [0.3552985 , 0.99054454, 0.38778594, 0.94289909],
       [0.76625383, 0.01588522, 0.43885749, 0.73488083]])

419 µs ± 18.6 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

array([[0.60766162, 0.66840636, 0.2019663 , 0.92669818],
       [0.69739471, 0.33933288, 0.9356699 , 0.61572307],
       [0.43173953, 0.40320305, 0.80278753, 0.46843452]])

49.8 µs ± 2.36 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

<Array [(0.376, 0.269, 0.149, ... 0.409, 0.38)] type='3 * (float64, float64, flo...'>

356 µs ± 11.1 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

<Array [[0.216, 0.481, ... 0.247, 0.0629]] type='3 * var * float64'>

263 µs ± 10.9 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

Other Formats