Ok, I think I didn’t do a full justice.
So actually, numba is fairly fast, given everything is precompiled. Predicates
still make things slower, but not as much as I initially thought.
However, the way your code is structured, it has to re-compile both predicate
and target function (with new pr
Your comparisons do not paint correct picture.
a) Most of time here is spent for array allocation and the actual time of the
loop gets swallowed in the noise
a) You do not test early stop - your benchmarks simply test full loop as
condition is almost never hit - 5 standard deviations...
Here is
Thanks Juan, this is really great! I plan to make use of this right away.
On Wed, Nov 1, 2023 at 8:13 AM Juan Nunez-Iglesias wrote:
> Have you tried timing things? Thankfully this is easy to test because the
> Python source of numba-jitted functions is available at jitted_func.py_func.
>
> In [
Have you tried timing things? Thankfully this is easy to test because the
Python source of numba-jitted functions is available at jitted_func.py_func.
In [23]: @numba.njit
...: def _first(arr, pred):
...: for i, elem in enumerate(arr):
...: if pred(elem):
...:
