python并行执行中变量的范围

Question

My issue is summarized with a simple code as follows:

#!/usr/bin/env python
from joblib import Parallel, delayed
import numpy as np
h =  np.ones(3)
def func(i):
    global h
    h[i] = h[i]+i
    print(h[i])
print(h)
Parallel(n_jobs=3)(delayed(func)(i) for i in range(3))
print(h)

The output is:

[ 1.  1.  1.]
1.0
2.0
3.0
[ 1.  1.  1.]

However, I would like the h values to be modified as they are modified inside the loop. What am I doing wrong?

Edit: I tried the answer backend="threading" suggested by L_S . It served the purpose. However, the code behaves in a serial manner. The loop was started with 3 n_jobs and I could see only one python running on doing top , whereas there are 3 python executables running without specifying backend="threading" .

Answer 1

By default Parallel uses the Python multiprocessing module to fork separate Python worker processes to execute tasks concurrently on separate CPUs.

So, You need to pass backend="threading"

Parallel(n_jobs=3, backend="threading")(delayed(func)(i) for i in range(3))