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
Currently, qp treats all parametrizations the same in calculating metrics (and stacking) by forcing evaluations on a grid, which doesn't always make sense, particularly for samples. At the very least, there should be more options for user control of that process, and ideally, there should be a way around it by calculating metrics from the native format without an extra approximation. (This would also speed up calculating metrics for the histogram representation, for example.)
The text was updated successfully, but these errors were encountered:
Currently,
qp
treats all parametrizations the same in calculating metrics (and stacking) by forcing evaluations on a grid, which doesn't always make sense, particularly for samples. At the very least, there should be more options for user control of that process, and ideally, there should be a way around it by calculating metrics from the native format without an extra approximation. (This would also speed up calculating metrics for the histogram representation, for example.)The text was updated successfully, but these errors were encountered: