-
-
Notifications
You must be signed in to change notification settings - Fork 74
Closed
Labels
code cleanuphelp wantedExtra attention is neededExtra attention is neededmethod inputsPython objects to CmdStan inputsPython objects to CmdStan inputsmethod outputsCmdStan outputs to Python objectsCmdStan outputs to Python objects
Description
EDIT: This is worse than I thought.
If I take a fitted model and do this:
fit = model.sample(...)
metric = fit.metric
then the right way to restart is as follows.
metric={"inv_metric": metric}`
This is confusing because of the mismatch. Something must be misnamed. If I use 1 / metric
instead of metric
to get the inverse of what fit.metric
returns, then the sampler grinds to a halt because of the anti-conditioning.
What I would really like is to be able to write is the following.
metric=metric
It would be helpful to call out the various cases of specifying an (inverse) metric with bullets the same way inits
is documented---as is, it's hard to visually identify the cases.
Metadata
Metadata
Assignees
Labels
code cleanuphelp wantedExtra attention is neededExtra attention is neededmethod inputsPython objects to CmdStan inputsPython objects to CmdStan inputsmethod outputsCmdStan outputs to Python objectsCmdStan outputs to Python objects