You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I see punpy uses builtin multiprocessing pools to parallelise its computations. Would it be possible to support other parallelism backends? Assuming a pure functional measurement function, it would be theoretically possible to launch it anywhere?
I am thinking about Dask RN. Supporting such a framework would allow to easily launch computations on many machines, which could be helpful when the measurement function is expensive to compute.
I believe it would be feasible to abstract the computation backend, so that users may use whatever they need/want, instead of importing a non-builtin parallelism backend as a new dependency for Punpy. multiprocessing.Pool would then be the default backend.
The text was updated successfully, but these errors were encountered:
It would be interesting to provide an access to the parallelism backend from the measurement function, so that this later can launch parallel subproblems on the same backend, and resume when these terminate. This would require an asynchronous implementation however, which may be more complicated to do for the time being.
I see punpy uses builtin multiprocessing pools to parallelise its computations. Would it be possible to support other parallelism backends? Assuming a pure functional measurement function, it would be theoretically possible to launch it anywhere?
I am thinking about Dask RN. Supporting such a framework would allow to easily launch computations on many machines, which could be helpful when the measurement function is expensive to compute.
I believe it would be feasible to abstract the computation backend, so that users may use whatever they need/want, instead of importing a non-builtin parallelism backend as a new dependency for Punpy. multiprocessing.Pool would then be the default backend.
The text was updated successfully, but these errors were encountered: