Simultaneously handling uncertainties and units in python

Question

I building a python library to analyze data (e.g. spectral data) with units and uncertainties.

For example, we have data on wavelength (nm), counts, and power (W). I want to be able to easily convert between units e.g. from wavelength in nm to um. I found that functionality in pint. I also want to handle uncertainties, and be able to, for example perform an sqrt(power) where the power uncertainty will be carried over. I found that functionality in uncertainties. Ideally, all of this information is stored and handled via a pandas.DataFrame, though that is not necessary.

My problem is that I can not figure out how to properly integrate pint quantities and quantity arrays with uncertainties.unumpy module. I would be happy to receive any suggestions on how to do this with the aforementioned packages, or any other packages.

There are two main issues I have noticed:

1. unumpy operations are not supported with pint.Quantity:

   This code that tries to take the square root of a Quantity defined with uncertainty

   from uncertainties import unumpy as unp
   from pint import UnitRegistry
   
   ureg = UnitRegistry()
   Qty = ureg.Quantity
   
   a = Qty(3, 'um').plus_minus(0.1)
   b =  unp.sqrt(a)
   

   returns an error message that requires the Quantity to be dimensionless:

   File "C:\Users\USERNAME\AppData\Local\Programs\Python\Python310\lib\site-packages\pint\facets\plain\quantity.py", line 707, in __float__
    raise DimensionalityError(self._units, "dimensionless")
   pint.errors.DimensionalityError: Cannot convert from 'micrometer' to 'dimensionless'
   

   As a note, trying the same with numpy just returns the same error an uncertainties.ufloat would:

   TypeError: loop of ufunc does not support argument 0 of type Measurement which has no callable sqrt method
   

2. Neither unumpy.uarray(), or pint.Measurement.from_list() can initialize an array of pint.Measurement objects. The first approach yields this error:

   File "C:\Users\USERNAME\AppData\Local\Programs\Python\Python310\lib\site-packages\pint\facets\plain\quantity.py", line 707, in __float__
    raise DimensionalityError(self._units, "dimensionless")
   pint.errors.DimensionalityError: Cannot convert from 'micrometer' to 'dimensionless'
   

   while the second approach yields this error, that is also related to the uncertainties package:

     File "C:\Users\USERNAME\AppData\Local\Programs\Python\Python310\lib\site-packages\uncertainties\core.py", line 2714, in raise_error
    raise TypeError("can't convert an affine function (%s)"
   TypeError: can't convert an affine function (<class 'uncertainties.core.Variable'>) to float; use x.nominal_value
   

Answer 1

The trick is to make the magnitude of the Quantity an uncertainty. Once you do this a lot of things just work. The biggest thing that doesn't work is that uncertainties don't work well with Pint's REPR. There's a pending pull request to fix that: https://github.com/hgrecco/pint/pull/1615

And there's also a pending pull request to make this all work well with Pandas: https://github.com/hgrecco/pint-pandas/pull/140