discover_network(..., n_jobs=-1) is in the signature and documented in the API reference, but there is no joblib/multiprocessing/Parallel anywhere in the package. Timing confirms it:
n_jobs=1 1.05 s
n_jobs=-1 0.99 s
We would like to support a parallelized workflow for larger networks. oCSE should work better for larger graphs than PCMCI. Further parallelizations can improve on this much more.
The reason it wasn't initially implemented is that the additional code bloat made the math hard to check.
discover_network(..., n_jobs=-1)is in the signature and documented in the API reference, but there is nojoblib/multiprocessing/Parallelanywhere in the package. Timing confirms it:We would like to support a parallelized workflow for larger networks. oCSE should work better for larger graphs than PCMCI. Further parallelizations can improve on this much more.
The reason it wasn't initially implemented is that the additional code bloat made the math hard to check.