The current version of sampled basin entropy (controlkernel.modularity.sampled_basin_entropy) implicitly assumes that attractors found starting from the sampled initial conditions are the only possible attractors. In many cases, we will be able to find all attractors exactly using the modularity code, but won't be able to find the exact basin entropy (because we won't be able to compute the exact transition graph).
If we instead give sampled_basin_entropy the known complete set of attractors, we should be able to use that to find the expected basin sizes of all attractors given our finite sampling (e.g. in a Bayesian analysis), even for attractors that we never find using sampling. This will likely give a much better estimate of basin entropy for cases with attractors with very small basins.
The current version of sampled basin entropy (
controlkernel.modularity.sampled_basin_entropy
) implicitly assumes that attractors found starting from the sampled initial conditions are the only possible attractors. In many cases, we will be able to find all attractors exactly using the modularity code, but won't be able to find the exact basin entropy (because we won't be able to compute the exact transition graph).If we instead give
sampled_basin_entropy
the known complete set of attractors, we should be able to use that to find the expected basin sizes of all attractors given our finite sampling (e.g. in a Bayesian analysis), even for attractors that we never find using sampling. This will likely give a much better estimate of basin entropy for cases with attractors with very small basins.