The Python interpreter is not completely isolated: importing numpy in 2 consecutive runs will lead to a Segmentation fault (compare https://github.com/numpy/numpy/issues/3837). Thus, all simulations have to be executed in a single interpreter. This class makes sure that the python interpreter is started and is kept alive, as long as on instance of the object exists. When the last instance is destroyed, the interpreter will be safely shutdown before reaching the end of main, preventing a segfault at the end of the application.
Note: Sub-interpreters and also this approach in general are not thread-safe. Instead of parallelizing simulations, you should use several threads in e.g. NEST (call pynn.nest='{"threads": 8}'), or combine several networks into one.
1.8.11
