MAINT: Fixup 9cbc0bc

- use `multiprocessing` in spawn mode
- finalise `Pool`
- remove `multiprocess` dependency

.travis.yml

@@ -40,7 +40,7 @@ install:
     - eval "$(conda shell.bash hook)"
     - conda activate test-env
     - travis_retry conda install -c conda-forge python=${TRAVIS_PYTHON_VERSION%-dev}
-    - travis_retry conda install -c conda-forge numpy scipy python-igraph h5netcdf multiprocess tqdm
+    - travis_retry conda install -c conda-forge numpy scipy python-igraph h5netcdf tqdm
     - travis_retry conda update  -c conda-forge --all
 
     # testing dependencies

setup.cfg

@@ -48,7 +48,6 @@ install_requires =
     scipy >= 1.10
     igraph >= 0.10
     h5netcdf >= 1.1
-    multiprocess >= 0.70
     tqdm >= 4.66
 python_requires = >=3.8
 packages = find:

src/pyunicorn/core/_ext/numerics.pyx

@@ -412,9 +412,9 @@ def _local_cliquishness_5thorder(
 def _nsi_betweenness(
     int N, ndarray[DWEIGHT_t, ndim=1] w,
     ndarray[DEGREE_t, ndim=1] k,
-    ndarray[NODE_t, ndim=1] targets,
     ndarray[NODE_t, ndim=1] flat_neighbors,
-    ndarray[MASK_t, ndim=1] is_source):
+    ndarray[MASK_t, ndim=1] is_source,
+    ndarray[NODE_t, ndim=1] targets):
     """
     Performs Newman's algorithm. [Newman2001]_
     """

src/pyunicorn/core/network.py

@@ -23,7 +23,8 @@
 
 import sys                          # performance testing
 import time
-from functools import wraps         # helper function for decorators
+from functools import partial, wraps
+from multiprocessing import get_context, cpu_count
 
 import numpy as np                  # array object and fast numerics
 from numpy import random
@@ -35,7 +36,6 @@
 
 import igraph                       # high performance graph theory tools
 
-from multiprocess import Pool, cpu_count
 from ..utils import mpi             # parallelized computations
 
 from ._ext.types import \
@@ -3323,15 +3323,15 @@ def nsi_betweenness(self, parallelize: bool = False, **kwargs):
         assert k.sum() == len(flat_neighbors) == 2 * self.n_links
         w, k = to_cy(w, DWEIGHT), to_cy(k, DEGREE)
 
-        def worker(batch):
-            return _nsi_betweenness(N, w, k, batch, flat_neighbors, is_source)
-
+        worker = partial(_nsi_betweenness, N, w, k, flat_neighbors, is_source)
         if parallelize:
             # (naively) parallelize loop over nodes
             n_workers = cpu_count()
             batches = np.array_split(to_cy(targets, NODE), n_workers)
-            pool = Pool()  # pylint: disable=not-callable
-            betw_w = np.sum(pool.map(worker, batches), axis=0)
+            with get_context("spawn").Pool() as pool:
+                betw_w = np.sum(pool.map(worker, batches), axis=0)
+                pool.close()
+                pool.join()
         else:
             betw_w = worker(to_cy(targets, NODE))
         return betw_w / w

tests/test_core/test_network.py

@@ -11,12 +11,14 @@
 # L. Tupikina, V. Stolbova, R.V. Donner, N. Marwan, H.A. Dijkstra,
 # and J. Kurths, "Unified functional network and nonlinear time series analysis
 # for complex systems science: The pyunicorn package"
+
 """
 Simple tests for the Network class.
 """
+
 from functools import partial
 from itertools import islice, product, repeat
-from multiprocess import Pool, cpu_count
+from multiprocessing import get_context, cpu_count
 
 import pytest
 import numpy as np
@@ -58,7 +60,7 @@ def compare_permutations(net, permutations, measures):
           map(np.random.permutation, repeat(net.N, permutations))))
     tasks = list(product(measures, range(permutations)))
     cores = cpu_count()
-    with Pool() as pool:  # pylint: disable=not-callable
+    with get_context("spawn").Pool() as pool:
         pool.map(partial(compare_measures, net, pnets, rev_perms),
                  (list(islice(tasks, c, None, cores)) for c in range(cores)))
         pool.close()