fix: officially remove top NormalizedIntensity

[spflueger]

Currently the expertsystem and tensorwaves are not compatible, as the highest level NormalizedIntensity is missing but is required when performing fits in tensorwaves.

To simplify things I propose the following:
We officially remove this highest level NormalizedIntensity as this is only required by fitting using a log likelihood estimator and belong there. This has several advantages:

1. It makes the amplitude model simpler as the normalization is not needed for the intensity definition
2. Data generation is fast as the normalization is skipped
3. The construction of the intensity in tensorwaves becomes more transparent

   builder = IntensityBuilder(part_list, kinematics) # note: the phase space sample here is optional
   intensity = builder.create_intensity(model)
   
   data_sample = generate_data(10000, kinematics, intensity)
   dataset = kinematics.convert(data_sample)
   
   phsp_sample = generate_phsp(100000, kinematics, random_generator, bunch_size=100)
   phspset = kinematics.convert(phsp_sample)
   
   estimator = UnbinnedNLL(intensity, dataset, phspset) # phspset is required here for the normalization

@redeboer @Leongrim @sebastianJaeger What do you think?

[redeboer]

Just a small note I can't put in the review: can __prepend_strength be removed?

expertsystem/src/expertsystem/amplitude/helicity_decay.py

Lines 477 to 489 in 79e59f3

	def __prepend_strength(
	self, intensity: IntensityNode
	) -> StrengthIntensity:
	strength = self.__register_parameter(
	name="strength_incoherent",
	value=1.0,
	fix=True,
	)
	return StrengthIntensity(
	component="incoherent_with_strength",
	strength=strength,
	intensity=intensity,
	)

[redeboer]

But generally I agree.

3. The construction of the intensity in tensorwaves becomes more transparent

Is that because a phase space sample is not required anymore?

[redeboer]

I'll fix the tests now

[spflueger]

Just a small note I can't put in the review: can __prepend_strength be removed?

expertsystem/src/expertsystem/amplitude/helicity_decay.py

Lines 477 to 489 in 79e59f3

	def __prepend_strength(
	self, intensity: IntensityNode
	) -> StrengthIntensity:
	strength = self.__register_parameter(
	name="strength_incoherent",
	value=1.0,
	fix=True,
	)
	return StrengthIntensity(
	component="incoherent_with_strength",
	strength=strength,
	intensity=intensity,
	)

It can be removed for now. It will be needed when using for example background parts in the amplitude model. So something like

Intensity = strength * SignalIntensity + (1-strength) * BackgroundIntensity

[redeboer]

#417 (comment)
Ok. Because next question I would have is whether the StrenghtIntensity and NormalizedIntensity classes are still needed at all.

[codecov]

Codecov Report

Merging #417 (d4aaaec) into master (0cb3d37) will decrease coverage by 0.70%.
The diff coverage is 60.00%.

@@            Coverage Diff             @@
##           master     #417      +/-   ##
==========================================
- Coverage   90.18%   89.48%   -0.71%     
==========================================
  Files          26       26              
  Lines        3750     3747       -3     
  Branches      925      926       +1     
==========================================
- Hits         3382     3353      -29     
- Misses        185      201      +16     
- Partials      183      193      +10     

Flag	Coverage Δ
unittests	89.48% <60.00%> (-0.71%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/expertsystem/amplitude/helicity_decay.py	88.84% <60.00%> (-0.89%)	⬇️
src/expertsystem/io/_xml/_dump.py	82.24% <0.00%> (-5.61%)	⬇️
src/expertsystem/io/_yaml/_dump.py	80.17% <0.00%> (-5.18%)	⬇️
src/expertsystem/io/_yaml/_build.py	84.87% <0.00%> (-5.05%)	⬇️
src/expertsystem/io/_xml/_build.py	81.52% <0.00%> (-3.83%)	⬇️

[spflueger]

But generally I agree.

1. The construction of the intensity in tensorwaves becomes more transparent

Is that because a phase space sample is not required anymore?

Yes, in cases where there is no internal normalizations going on, the phase space sample is not needed anymore. However it may still be needed in the the IntensityBuilder, if normalizations within the amplitude model appear. For instance when the model includes

• signal and a background term. Then it can be beneficial to normalize both parts to have meaningful strengths (value E [0,1])
• normalizing amplitudes. Same logic as above. Magnitudes then correspond to fit fractions.
• maybe more use cases

[redeboer]

LGTM!
Just one comment #417 (comment) about the if else statement

[spflueger]

#417 (comment)
Ok. Because next question I would have is whether the StrenghtIntensity and NormalizedIntensity classes are still needed at all.

See #417 (comment)