jpivarski
commented
4 years ago It's because Uproot 3 makes Python classes that have the same inheritance as the underlying C++ classes, TProfile2D inherits from TH2D, and the values method has not been overridden with profile-specific logic. So the problem is that the method you want hasn't been implemented, but due to historical inheritance issues, it manifests itself as the wrong answer, rather than a "not implemented" error or missing method, as it should.
Let's try to fix this in Uproot 4, because Uproot 3 is quickly approaching deprecation-time. In Uproot 4, the Python classes have a flat hierarchy and so we more fully control which class gets which method. The Uproot 4 TProfile2D has an implementation for edges but not for values:
But TProfile (1-dimensional) has a carefully worked out implementation for values and errors:
with extensive tests that were carefully checked against ROOT's output:
Probably all that needs to be done is to replace expressions like len(array) with array.shape because handling 2-dimensional profile values/errors should be just like handling 1-dimensional profile values/errors with a 2-dimensional shape. It might be even simpler to turn the 1-dimensional implementation into a helper function, flatten a 2-dimensional array on input and restore its 2-dimensional shape on output, and therefore not even have to touch the inner implementation.
Would you like to try that?
wiso
I am reading a tprofile2d with uproot with the numpy method
The edges are read correctly, but what I got as values seems to be the number of entries for each bin, while for a TProfile I am interested to the average of the quantity I am profiling