Using multiprocessing.Pool cannot share objects between parent and child process, which requires expensive serialize/deserialize operations. The downside of parallel_map is that the pool does not stay open, but that is OK since we were closing multiprocessing.Pool once per method invocation anyway.
Using
multiprocessing.Pool
cannot share objects between parent and child process, which requires expensive serialize/deserialize operations. The downside of parallel_map is that the pool does not stay open, but that is OK since we were closing multiprocessing.Pool once per method invocation anyway.