unumpy | A backend-dispatchable version of NumPy | Data Manipulation library

0 has physical significance for sure. But for computation, Computer Science can't handle NaN or Inf as they are indefinite. So usually how we handle this mathematical error by adding a small value. Which will not have much impact.

Usually in many mathematical libraries they handle like below;

lets say we have matrix/vector A which can contain 0. and I am going to divide by another matrix. So before that we can do below preprocessing.

Instead of using a regular expression you could use the rsplit() method to split a string into a list, starting from the right:

I can reproduce this, the print is what is taking forever. Or rather, it is the conversion to string implicitly called by print. I used line_profiler to measure the time of the __format__ function of AffineScalarFunc. (It is called by __str__, which is called by print) I decreased the array size from 8200 to 1000 to make it go a bit faster. This is the result (pruned for readability):

From the documentation:

floatfmt argument can be a list or a tuple of format strings, one per column, in which case every column may have different number formatting::