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
Good catch snl-bconder and thanks for the nice write-up. np.isclose has a weird history and we got bit by its odd definition here. We're working through a fix that makes the printout consistent with the checked arithmetic and will link the PR to this issue once it is posted.
Per the numpy documentation for np.isclose, the total tolerance of that function is equal to
atol + rtol * abs(b)
SarPy calls the np.isclose method when doing approximations.
When the CPHD validator provides verbose output, the printed tolerances are incorrectly shown per the calculation in consistency.py:438.
This syndrome I observe is that tolerances are under-represented, and the representation shown for negative values will not account for rtol at all.
The text was updated successfully, but these errors were encountered: