Assembly getParamOfZFunction Interpolation Warning

[wcscherer]

The assembly method getParamOfZFunction interpolates a param axially to find it at any value of elevation z by using the scipy.interpolate.interp1d method. There are two issues with the implementation of this assembly method:

1. It was found in the unit tests of the armi/physics/neutronics/globalFlux/globalFluxInterface.DoseResultsMapper after it was moved out of ARMI in Removing DoseResultsMapper #1952, that if an assembly only returns one z location to interpolate over, the scipy.interpolate.interp1d method will return a Nan value and a Runtime warning stating an invalid value was encountered. This behavior is undesirable as the associated unit tests still passed but with the warnings. The error was resolved by using a test reactor with more than one z location for interpolation. It would be beneficial to include logic in the getParamOfZFunction to check for more than one z location in the assembly to interpolate over.
   

2. The scipy.interpolate.interp1d method is now an unmaintained legacy method: https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp1d.html . The docs recommend updating any use of interp1d with numpy.interp, so getParamOfZFunction should be updated to use the current recommended interpolation method. It's important to note that numpy.interp doesn't even provide a warning if a single value is passed for interpolation, meaning the check proposed in the first point would still apply to using this interpolation method.

[john-science]

For (1), we already test getParamOfZFunction with an Assembly that is 3 blocks tall:

armi/armi/reactor/tests/test_assemblies.py

Lines 921 to 938 in e777af7

	def test_getParamValuesAtZ(self):
	# single value param
	for b, temp in zip(self.assembly, [80, 85, 90]):
	b.p.percentBu = temp
	percentBuDef = b.p.paramDefs["percentBu"]
	originalLoc = percentBuDef.location
	try:
	self.assertAlmostEqual(
	87.5, self.assembly.getParamValuesAtZ("percentBu", 20.0)
	)
	percentBuDef.location = parameters.ParamLocation.BOTTOM
	self.assertAlmostEqual(
	82.5,
	self.assembly.getParamValuesAtZ("percentBu", 5.0, fillValue="extend"),
	)
	percentBuDef.location = parameters.ParamLocation.TOP
	self.assertAlmostEqual(
	82.5, self.assembly.getParamValuesAtZ("percentBu", 15.0)

That test says it tests getParamValuesAtZ(), but that method is just a one-line pass-through to getParamOfZFunction().

[john-science]

I can look at updating our usage of the scipy API to remove the warnings. Sure.

If it's just an API change, it's easy to handle. If it comes with numerical differences, it will take much longer to make the change.

#fingers-crossed

[wcscherer]

Thanks John. The issue was the offending test in question used the smallest test reactor of one assembly made of one block. When passed into the interpolator, it "failed" to interpolate over one z location and returned Nan with a warning.

Apparently, this method used to fail when passed an array of one value to interpolate over but that functionality was changed a decade ago to only warn the user. The warnings were noticed in the unit tests, even though the tests passed.

The original writers of the test should have known to use a larger test reactor, but neither the assembly method nor the interpolator gave them the information to know that unless they dug into the warnings - which they didn't. It would be helpful to have some guardrails around these kind of methods to avoid similar errors in the future.

This is a nice to have, not urgent.

[john-science]

Ouch.

Yeah, switching from scipy.interp1d() to numpy.interp() is not a drop-in replacement. The NumPy version is much simpler, and more limited, and doesn't have features from the SciPy version that we actually use:

armi/armi/reactor/assemblies.py

Lines 1137 to 1143 in cd616ed

	interpolater = interpolate.interp1d(
	z,
	values,
	kind=interpType,
	fill_value=fillValue,
	assume_sorted=True,
	bounds_error=boundsError,

The NumPy version is much simpler: