parallel on Windows only uses parallel processing on iterating over first guess

[lenarddome]

In most of the optimisation functions, users can specify multiple starting guesses; the function iterates over those starting guesses for the optimisation.

It has been reported that on Windows, the number of cores specified only takes effect on the first guess but not on all subsequent iterations.

@frankhezemans reported this issue

[lenarddome]

One possible reason why it might happens is because we create a new multiprocess.Pool on every iteration within a function:

for i in range(len(self.initial_guess)):
            print(
                f"Starting optimization {i+1}/{len(self.initial_guess)} from {self.initial_guess[i]}"
            )
            self.__current_guess__ = self.initial_guess[i]
            if self.__parallel__:
                with mp.Pool(self.cl) as pool:
                    results = pool.map(__task, self.data)
            else:
                results = list(map(__task, self.data))

This might cause a lot of overhead and failure to re-initialise processes, especially if some processes are hanged on the cores that were already used.
Windows also has some issues with parallel processing so it might be better to set up parallel processing outside of the for loop.

[lenarddome]

Note that I cannot recreate the problem - so if anyone has a reproducible example, please post it here.