When copied across multiple chains via ParallelChain, IndependenceSampler will sample the same samples in every chain, aka it starts on the same random state. While this is not too much of an issue in itself, since IndependenceSampler isn't that useful, I wonder whether a similar issue could affect other proposals or other parts of tinyDA. I suspect it has to do with scipy's distributions, but I have yet to see alarming behavior elsewhere. DREAMZ's initial archive for example seems to not have the same issue.
When copied across multiple chains via
ParallelChain,IndependenceSamplerwill sample the same samples in every chain, aka it starts on the same random state. While this is not too much of an issue in itself, sinceIndependenceSamplerisn't that useful, I wonder whether a similar issue could affect other proposals or other parts of tinyDA. I suspect it has to do withscipy's distributions, but I have yet to see alarming behavior elsewhere.DREAMZ's initial archive for example seems to not have the same issue.