This snippet of code shows you how to use your own C++ functions on a numpy array instancied in Python You will find two simple examples on which you can expand :

• one with a simple c++ function to square the values of a numpy array
• one with an oop approach

The initial problem appeared when training RL models on fluid mechanics problems and more generally on simulations where returning the next step of the environment is very costly due to complex numerical schemes.

Solving these numerical schemes in Python makes it the bottleneck of the program

A good implementation of said numerical scheme on c++ can be a solution then - we still use Python to handle the environment and model with openai-gym for example

It can of course be used for other means

The library boost python, and you'll need to changeLIBRARIES and INCLUDES in the makefile to match your boost local path

If you have a clean installation of python3.6, the python3 example should work as it is

(On UiO computers, type module load python/anaconda in terminal and it will work as it is for python2)