You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Nov 17, 2025. It is now read-only.
Our current approach of tracking Variables that have been converted to MeasurableVariables and reverting them when they're not eventually associated with a value variable (i.e. "bound" in some sense), is rather inefficient. We could—at the very least—make sure that these reversions are always performed in a reverse-topological order, so that fewer replacements are necessary (e.g. assuming that reversion is only ever necessary closest to the "base" valued/bound variable).
The reason we currently convert all terms to measurable terms—and not just the nodes that are directly bound/valued—is that some node conversions can only occur once the node's inputs are determined to be measurable. We can take a number of entirely different approaches, but the current is thorough and simple (e.g. MeasurableVariable-specific guided and short-circuited depth-first conversions, "inverse" functions associated with each MeasurableVariable type, consistent use of OpFromGraph, etc.) A fix to this issue could easily involve one of those approaches.
Our current approach of tracking
Variables that have been converted toMeasurableVariables and reverting them when they're not eventually associated with a value variable (i.e. "bound" in some sense), is rather inefficient. We could—at the very least—make sure that these reversions are always performed in a reverse-topological order, so that fewer replacements are necessary (e.g. assuming that reversion is only ever necessary closest to the "base" valued/bound variable).The reason we currently convert all terms to measurable terms—and not just the nodes that are directly bound/valued—is that some node conversions can only occur once the node's inputs are determined to be measurable. We can take a number of entirely different approaches, but the current is thorough and simple (e.g.
MeasurableVariable-specific guided and short-circuited depth-first conversions, "inverse" functions associated with eachMeasurableVariabletype, consistent use ofOpFromGraph, etc.) A fix to this issue could easily involve one of those approaches.