The Pathfinder algorithm is an approximate inference algorithm supported by Stan.
Pathfinder can be used as an inference algorithm in its own right, but the primary usage is to support faster initialization of Stan programs. As such, really "supporting" it would require we have a nice way of turning some draws from pathfinder into the init JSON objects sampling expects. This is not trivial, unfortunately - in Python I ended up writing an entire package for essentially that use case, and it heavily uses numpy to do it
The Pathfinder algorithm is an approximate inference algorithm supported by Stan.
Pathfinder can be used as an inference algorithm in its own right, but the primary usage is to support faster initialization of Stan programs. As such, really "supporting" it would require we have a nice way of turning some draws from pathfinder into the init JSON objects sampling expects. This is not trivial, unfortunately - in Python I ended up writing an entire package for essentially that use case, and it heavily uses numpy to do it