why has the function for pool.map to be defined before multiprocessing.pool?

Question

When using multiprocessing.pool and pool.map, it seems that the map function has to be defined before creating the pool. Why is this necessary?

Here is the working version:

import multiprocessing

def my_power(x):
  return x * x

_pool = multiprocessing.Pool()
my_list = _pool.map(my_power, list(range(10)))
print(my_list)

The version that gets the error, AttributeError: Can't get attribute 'my_power'...

_pool = multiprocessing.Pool()

def my_power(x):
  return x * x

my_list = _pool.map(my_power, list(range(10)))
print(my_list)

Answer 1

For multiprocessing, the code needs to be serialised. I guess it is easier for the runtime to serialise everything before creating the pool. Otherwise it has to analyse and decide what to serialise.