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
We have debated for length about how to handle returned values from a DynamicPPL model, and use that as a way to store extra non-parameter model variables (e.g. transformed variables and generated quantities; see here). One straightforward approach is to introduce an extra field in SimpleVarInfo, e.g.
struct SimpleVarInfo
logp
values
ret_values
transformation
end
We can safely compress the ret_values and values into one field and skip ret_values during inference.
More generally, the returned value can be a closure, allowing us to easily perform post-inference tasks, e.g., prediction.
The text was updated successfully, but these errors were encountered:
We have debated for length about how to handle returned values from a DynamicPPL model, and use that as a way to store extra non-parameter model variables (e.g. transformed variables and generated quantities; see here). One straightforward approach is to introduce an extra field in
SimpleVarInfo
, e.g.We can safely compress the
ret_values
andvalues
into one field and skipret_values
during inference.More generally, the returned value can be a closure, allowing us to easily perform post-inference tasks, e.g., prediction.
The text was updated successfully, but these errors were encountered: