fixing LD bug for joint fits

[erinmmay]

addressing bug #743

I believe this is how we want to do it, limb darkening should have channel id appended regardless of if it's a free fit since we can fix them to different values.

[erinmmay]

also adding a minor fix for fleck. Currently if you set the star rotation (spotrot) in the s5 fit params epf file then fleck_fast is never set to anything and the code crashes, this adds the lines to set fleck_fast when spotrot is defined in the epf

[taylorbell57]

I haven't yet confirmed where this bug arises, but as it stands I don't think this patch is quite right. If there's a bug (which I'm certainly willing to believe, though it hasn't arisen for non-multwhite fits), it must be in the interactions between these snippets of code:

Eureka/src/eureka/S5_lightcurve_fitting/models/AstroModel.py

Lines 146 to 171 in a8b3a8a

	# Load in all the values for each astro parameter
	for item in self.__dict__.keys():
	item0 = item+self.pid_id
	try:
	if model.parameters.dict[item0][1] == 'free':
	item0 += self.channel_id
	value = getattr(parameterObject, item0)
	if eval:
	value = value.value
	setattr(self, item, value)
	except KeyError:
	if (item in [f'u{i}' for i in range(1, 5)] or
	'spot' == item[:4]):
	# Limb darkening and spots probably don't vary with planet
	try:
	item0 = item
	if model.parameters.dict[item0][1] == 'free':
	item0 += self.channel_id
	value = getattr(parameterObject, item0)
	if eval:
	value = value.value
	setattr(self, item, value)
	except KeyError:
	pass
	else:
	pass

Eureka/src/eureka/S5_lightcurve_fitting/models/AstroModel.py

Lines 306 to 309 in a8b3a8a

	ld_func = ld_profile(self.limb_dark)
	len_params = len(inspect.signature(ld_func).parameters)
	coeffs = ['u{}'.format(n) for n in range(1, len_params)]
	self.u = [getattr(self, coeff) for coeff in coeffs]

Eureka/src/eureka/S5_lightcurve_fitting/models/BatmanModels.py

Lines 38 to 73 in a8b3a8a

	# Store the ld_profile
	self.ld_from_S4 = kwargs.get('ld_from_S4')
	ld_func = ld_profile(self.parameters.limb_dark.value,
	use_gen_ld=self.ld_from_S4)
	len_params = len(inspect.signature(ld_func).parameters)
	self.coeffs = ['u{}'.format(n) for n in range(1, len_params)]
	self.ld_from_file = kwargs.get('ld_from_file')
	# Replace u parameters with generated limb-darkening values
	if self.ld_from_S4 or self.ld_from_file:
	log.writelog("Using the following limb-darkening values:")
	self.ld_array = kwargs.get('ld_coeffs')
	for c in range(self.nchannel_fitted):
	chan = self.fitted_channels[c]
	if self.ld_from_S4:
	ld_array = self.ld_array[len_params-2]
	else:
	ld_array = self.ld_array
	for u in self.coeffs:
	index = np.where(np.array(self.paramtitles) == u)[0]
	if len(index) != 0:
	item = self.longparamlist[c][index[0]]
	param = int(item.split('_')[0][-1])
	ld_val = ld_array[chan][param-1]
	log.writelog(f"{item}: {ld_val}")
	# Use the file value as the starting guess
	self.parameters.dict[item][0] = ld_val
	# In a normal prior, center at the file value
	if (self.parameters.dict[item][-1] == 'N' and
	self.recenter_ld_prior):
	self.parameters.dict[item][-3] = ld_val
	# Update the non-dictionary form as well
	setattr(self.parameters, item,
	self.parameters.dict[item])

If the bug is specifically with loading-in exotic-ld results (and/or custom ld files), then my pretty strong suspicion is that the bug is in the last of those three snippets (from BatmanModels.py or the equivalent chunks of code for fleck or starry). I won't have time to look into this more today, but I can try to investigate further tomorrow if you don't beat me to it.

[erinmmay]

I added tons of print statements to confirm that the code you reference in BatmanModels sets things up in properly when reading from the exotic values and that the bug arises when BatmanModels.py calls AstroModel/PlanetParams to initialize the planet. AstroModel line 308/309 was previously only looking for u1/u2, never u1_ch1/u2_ch1 and the result was that both visits used the same limb darkening (confirmed with the print statements, visually with the plots, and by switching the order of the visits to see which LD values were used in that case)

[taylorbell57]

I really would've thought that the ld parameters would be populated with the values for ch1 based on the snippet below which is used to set (in the pretend case of some parameter called par) the value of PlanetParams.par to the value of parameterObject.par_ch#:

Eureka/src/eureka/S5_lightcurve_fitting/models/AstroModel.py

Lines 149 to 155 in a8b3a8a

	try:
	if model.parameters.dict[item0][1] == 'free':
	item0 += self.channel_id
	value = getattr(parameterObject, item0)
	if eval:
	value = value.value
	setattr(self, item, value)

It is the same snippet that is used for every other variable, with the only exception being that lower down there's a snippet that allows ld parameters and spot parameters to be the same between different planets.

[taylorbell57]

The lines to which you are referring:

Eureka/src/eureka/S5_lightcurve_fitting/models/AstroModel.py

Lines 308 to 309 in a8b3a8a

	coeffs = ['u{}'.format(n) for n in range(1, len_params)]
	self.u = [getattr(self, coeff) for coeff in coeffs]

use self.u1 which should've been set to the value of parameterObject.u1_ch1 by those lines I just linked above.

[erinmmay]

yeah unfortunately they are not set to parameterObject.u1_ch1. self.u1 returns the actual u1 value. If there's a different way you want to fix this happy to do that instead, but as it stands it's not working as expected.

[erinmmay]

Ahh I see the issue. Lines 149-155 in AstroModel that you linked only set self.u1 to parameterObject.u1_ch1 if the parameter is free. But We often fix to the exotic-values, in which case it doesn't append the channel id.

[taylorbell57]

Ahh yeah, that'd do it! Honestly maybe we don't need that 'free' check at all since its already wrapped in a KeyError check, though maybe we still do (worth investigating).

[taylorbell57]

@erinmmay, can you click the "allow maintainers to push edits to this pull request" (or whatever the exact wording is) button so that I can make some small formatting tweaks at some point tomorrow?

[erinmmay]

On my end it says that's already enabled.

[taylorbell57]

Okay thanks, must just be an issue with the online GitHub Codespaces then - I'll try using my local git installation later

[taylorbell57]

@erinmmay, I just reworked this patch to (hopefully) more generally catch similar issues with fixed parameters during shared or multwhite fits. Can you please test this new code and make sure it still resolves your original problem?

[taylorbell57]

@erinmmay, have you had the chance to test my edits to this PR? I'd like to get this patch merged if it's good-to-go

[taylorbell57]

@kevin218, do you want to review this PR - it works for me, but would be good to get a second set of eyes on this