pymc.pytensorf.make_shared_replacements

pymc.pytensorf.make_shared_replacements(point, vars, model)

Make shared replacements for all other variables than the ones passed.

This way functions can be called many times without setting unchanging variables. Allows us to use func.trust_input by removing the need for DictToArrayBijection and kwargs.

Parameters:

point: dictionary mapping variable names to sample values

vars: list of variables not to make shared

model: model

Returns:

Dict of variable -> new shared variable

Notes

A fresh shared variable per call means the compiled graph differs for every step method instance, so these functions can never be served from a frozen model’s compilation cache. Making the shared variables depend only on the model, and swapping in per-instance ones with Function.copy afterwards (as ValueGradFunction.copy() does), would let these steps reuse a cached graph as well.