Question

I am working on an algorithm that requires a Gaussian function and a lot of loops.
The Boost::math::pdf has the correct formula that I need to implement.

I am creating a normal distribution with boost::math::normal_distribution normal_distribution(0,sigma) where the mean will always be zero and the sigma can vary at different times when the function is called.

I then want to use the function boost::math::pdf( normal_distribution , dsitance_from_mean ) in my loop to avoid reinventing the wheel with my own look up table. I read through the boost documentation looking for this answer, but I am not quite clear on how the function works.

Does the pdf function use a look up table generated with the normal distribution, or does it calculate the exponent value every time it is called?

Answer 1

My expectation would be that the Boost authors would prioritize accuracy over performance.

Browsing through the source code for the version I have installed, I see plenty of
calls to trig/log/exp functions, and nothing resembling lookup tables or interpolation,
so I think it’s calling exp() every time.

I would recommend benchmarking it yourself, to see if the Boost code is really a
performance bottleneck, before implementing your own lookup table.