Reputation: 4510
As I understand it, vectorized numpy functions are faster than Python loops because loops are done in C or Fortran. I would like to know where in the source code this happens.
For example, the scipy.special.bdtr
binomial CDF function accepts array-like arguments k,n,p
and will return an ndarray provided the arguments are broadcastable. The documentation says that scipy.special.bdtr
is a wrapper for a routine in the Cephes Mathematical Functions Library. Digging through the source code on Github, I found a scipy/special/cephes/bdtr.c
file containing the C code for the routine; here are what I believe to be the first three lines of the relevant C function:
double bdtr(k, n, p)
int k, n;
double p;
It appears that the underlying C function does not operate on arrays, and I can't find the source code where this function is converted to a Python function that operates on arrays.
Upvotes: 1
Views: 339
Reputation: 59701
In the case of scipy.special
functions, the C code only contains the "kernels" of the functions, that is, how to apply the function to scalars. Each of these is then wrapped into a ufunc
with automatically generated Cython code. To do this, it uses C header files, like scipy/special/cephes.h
, Cython declaration files, like scipy/special/_cephes.pxd
, the file scipy/special/functions.json
, where all the functions to be generated for scipy.special
are listed, and finally scipy/special/_generate_pyx.py
, which is where the Cython code is actually produced.
Upvotes: 2