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
because MAP is just an alias to AP. Can we (and do we want) alias names to flow through? I'm not sure we can set name on a per-instance basis, but could probably move NAME to be an instance-level field.
This potentially brings up a bunch of questions, though. E.g., do alias names affect equality (MAP == AP)? What happens if metrics are requested for both MAP and AP (for some reason)?
If we want this behavior, I think alias names should affect equality; essentially, they'll be treated as different measures. Otherwise, there could be weird behavior when using them as dictionary keys and such. To get the (potentially aliased) name, use NAME. To get the canonical name, use __name__ (e.g., providers will want to start using __name__ instead of NAME, and support matching should also be done with __name__).
What about aliases that include parameters, like NumRelRet -> NumRet(rel=1)? With the above proposal, it'd end up being NumRelRet -> NumRelRet(rel=1), without specific handling for this case. Maybe it's worth just making it its own measure?
The text was updated successfully, but these errors were encountered:
Right now:
because MAP is just an alias to AP. Can we (and do we want) alias names to flow through? I'm not sure we can set name on a per-instance basis, but could probably move NAME to be an instance-level field.
This potentially brings up a bunch of questions, though. E.g., do alias names affect equality (
MAP == AP
)? What happens if metrics are requested for both MAP and AP (for some reason)?If we want this behavior, I think alias names should affect equality; essentially, they'll be treated as different measures. Otherwise, there could be weird behavior when using them as dictionary keys and such. To get the (potentially aliased) name, use
NAME
. To get the canonical name, use__name__
(e.g., providers will want to start using__name__
instead ofNAME
, and support matching should also be done with__name__
).What about aliases that include parameters, like
NumRelRet -> NumRet(rel=1)
? With the above proposal, it'd end up beingNumRelRet -> NumRelRet(rel=1)
, without specific handling for this case. Maybe it's worth just making it its own measure?The text was updated successfully, but these errors were encountered: