Following #105, GenerateFromSeed is not deterministic.
This is legacy functionality from Planck primordial reconstruction days, for when sampling randomly from the hypercube is too inefficient due to e.g. theoretical constraints. Examples of this are when you have a sorted uniform prior with more than 12 components, sampling from 1/12! of the space is too inefficient. The inflation use case was that when reconstructing an inflationary potential using a polynomial spline, most potentials did not result in a universe being create.
The GenerateFromSeed functionality addressed this by requiring you to know a theoretically acceptable point, and then slice sampling from that. This is quite inefficient, and could in principle be more flexibly accommodated by having an external code which generates an initial set of live points, as the cube_samples code does
It is also inaccessible through the python UI, so I anticipate this affecting very few, if any, users. Happy to reinstate on request.
Following #105, GenerateFromSeed is not deterministic.
This is legacy functionality from Planck primordial reconstruction days, for when sampling randomly from the hypercube is too inefficient due to e.g. theoretical constraints. Examples of this are when you have a sorted uniform prior with more than 12 components, sampling from 1/12! of the space is too inefficient. The inflation use case was that when reconstructing an inflationary potential using a polynomial spline, most potentials did not result in a universe being create.
The GenerateFromSeed functionality addressed this by requiring you to know a theoretically acceptable point, and then slice sampling from that. This is quite inefficient, and could in principle be more flexibly accommodated by having an external code which generates an initial set of live points, as the cube_samples code does
It is also inaccessible through the python UI, so I anticipate this affecting very few, if any, users. Happy to reinstate on request.