Closed tlienart closed 7 years ago
Best is probably to avoid making any strong claim of consistency and going more for the EPBP approach where you try to have stuff that make sense...
^ probably best since also normalised importance sampling is potentially biased meaning that the whole thing is biased anyway.
SAPB, Taghevi and Fearnhead's linear algorithms are not consistent. In order to be consistent, the algorithm must be quadratic to compute the corresponding weights. Is this discussed anywhere?
Try to see explicitly with Arnaud whether can be discussed.