Uncertainties + pint

[SengerM]

I am trying to combine uncertainties and pint to avoid tedious calculations. When operations consist only in +,-,* and / everything works perfectly. The problem arises when I try to use other functions of the uncertainties package such as cosine(), sqrt(), etc. A MWE of what I am trying to do is as follows

import uncertainties as un
from uncertainties.umath import *
import pint
ureg = pint.UnitRegistry()
a = un.ufloat(10,1)*ureg.meter
b = un.ufloat(2,0.1)*ureg.meter
print(a/b) # This works nice
print(sqrt(un.ufloat(3,1))) # This works nice
print(sqrt(a/b)) # This fails

As far as I know the problem is that sqrt() is waiting for a ufloat object and I am passing a "pint type object". Is there any way to implement this?

[lebigot]

Indeed, Pint objects should be handled by Pint instead of the uncertainties module: units have to be handled somewhere.

Now, how would you do the same if your Pint quantities didn't have uncertainties?

[SengerM]

For Pint quantities without uncertainties it can be done using Numpy:

import pint
import numpy
ureg = pint.UnitRegistry()
a = 10*ureg.meter
b = 20*ureg.meter
print(numpy.sqrt(a/b))

[lebigot]

And so what happens if you do the same with uncertainties?

[SengerM]

This

import pint
import numpy
import uncertainties as un
import uncertainties.unumpy as unp
ureg = pint.UnitRegistry()
a = un.ufloat(10,1)*ureg.meter
b = un.ufloat(20,1)*ureg.meter
print(a)
print(numpy.sqrt(a/b))

gives

10.0+/-1.0 meter
Traceback (most recent call last):
  File "test.py", line 9, in <module>
	print(numpy.sqrt(a/b))
AttributeError: 'AffineScalarFunc' object has no attribute 'sqrt'

[lebigot]

Thanks. I'll investigate some more the interaction between NumPy and non-basic types, then. My guess is that uncertainties could interface more with NumPy (in addition to what is already available in uncertainties.unumpy).