Reputation: 275
I found the following example on this website somewhere:
import multiprocessing
import ctypes
import numpy as np
shared_array_base = multiprocessing.Array(ctypes.c_double, 10*10)
shared_array = np.ctypeslib.as_array(shared_array_base.get_obj())
shared_array = shared_array.reshape(10, 10)
# No copy was made
assert shared_array.base.base is shared_array_base.get_obj()
# Parallel processing
def my_func(i, def_param=shared_array):
shared_array[i,:] = i
if __name__ == '__main__':
pool = multiprocessing.Pool(processes=4)
pool.map(my_func, range(10))
print shared_array
The above code works fine, but if I want to add an array to the shared array, something like shared_array += some_other_array (instead of the above shared_array[i,;] = i) I get
local variable 'shared_array' referenced before assignment
Any ideas why I cannot do that?
Upvotes: 4
Views: 3923
Reputation: 251568
If a variable is assigned to anywhere in a function, it is treated as a local variable. shared_array += some_other_array
is equivalent to shared_array = shared_array + some_other_array
. Thus shared_array
is treated as a local variable, which does not exist at the time you try to use it on the right-hand side of the assignment.
If you want to use the global shared_array
variable, you need to explicitly mark it as global by putting a global shared_array
in your function.
The reason you don't see the error with shared_array[i,:] = i
is that this does not assign to the variable shared_array
. Rather, it mutates that object, assigning to a slice of it. In Python, assigning to a bare name (e.g., shared_array = ...
) is very different from any other kind of assignment (e.g., shared_array[...] = ...
), even though they look similar.
Note, incidentally, that the error has nothing to do with multiprocessing.
Upvotes: 6