sfh=1, updating both sf_start and tage via Prospector

[k-a-webb]

Hello,

I am working with Prospector to build a parametric SFH model. I've encountered an issue when trying to modify both 'tage' and 'sf_start' with the sfh=1 FSPS model.

The error traces back to the assertion statements (here, lines 1314-1318).
It looks like because Prospector loops through the parameter list, and in each loop there is this check, it can fail if "tage" updates before "sf_start", and the new "tage" is younger than the old "sf_start". I imagine this could be avoided if the check was done after updating the full parameter list.

I've skirted this issue by commenting out the assertion statement, and it appears to be running fine. I have built my actual model such that sf_start < tage < sf_trunc.

Here is a basic example of how I am encountering this issue:

from prospect.models.templates import TemplateLibrary
model_params = TemplateLibrary["parametric_sfh"]

model_params["tage"] = {'N': 1, 'init': 0, 'units': 'Gyr'}

model_params["sf_start"] = {'N': 1, 'init': 1, 'units': 'Gyr', }

model_params["sf_trunc"] = {'N': 1, 'init': 3, "units": "Gyr", }


model_params["tage"]["init"] = 0.5*( model_params["sf_start"]['init'] + model_params["sf_trunc"]['init'] )
from prospect.models import sedmodel
model = sedmodel.PolySpecModel(model_params)


from prospect.sources import CSPSpecBasis
sps = CSPSpecBasis(zcontinuous=1)

# sf_start < tage < sf_trunc
# this will work
_ = model.predict( model.theta, sps=sps, obs={"wavelength":None, "filters":None})

# Now make the new sf_start larger than the old tage
model.params["sf_start"] = model.params["tage"] + 0.1
model.params["tage"] = 0.5*( model.params["sf_start"] + model.params["sf_trunc"] )
# this will work
_ = model.predict( model.theta, sps=sps, obs={"wavelength":None, "filters":None})


# Now make the new tage younger than the old sf_start
model.params["sf_start"] = 0.1
model.params["tage"] = 0.5*( model.params["sf_start"] + model.params["sf_trunc"] )
# this will fail
_ = model.predict( model.theta, sps=sps, obs={"wavelength":None, "filters":None})

Here is the full error code:

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-12-77dec1eb0718> in <module>
----> 1 _ = model.predict( model.theta, sps=sps, obs={"wavelength":None, "filters":None})

~/venv3/lib/python3.9/site-packages/prospect/models/sedmodel.py in predict(self, theta, obs, sps, sigma_spec, **extras)
     72         # generate and cache model spectrum and info
     73         self.set_parameters(theta)
---> 74         self._wave, self._spec, self._mfrac = sps.get_galaxy_spectrum(**self.params)
     75         self._zred = self.params.get('zred', 0)
     76         self._eline_wave, self._eline_lum = sps.get_galaxy_elines()

~/venv3/lib/python3.9/site-packages/prospect/sources/galaxy_basis.py in get_galaxy_spectrum(self, **params)
    106         # Loop over mass components
    107         for i, m in enumerate(mass):
--> 108             self.update_component(i)
    109             wave, spec = self.ssp.get_spectrum(tage=self.ssp.params['tage'],
    110                                                peraa=False)

~/venv3/lib/python3.9/site-packages/prospect/sources/galaxy_basis.py in update_component(self, component_index)
     80                     this_v = v
     81 
---> 82                 self.ssp.params[k] = deepcopy(this_v)
     83 
     84     def get_galaxy_spectrum(self, **params):

~/venv3/lib/python3.9/site-packages/fsps/fsps.py in __setitem__(self, k, v)
   1304 
   1305             self._params[k] = v
-> 1306             self.check_params()

~/venv3/lib/python3.9/site-packages/fsps/fsps.py in check_params(self)
   1281             6
   1282         ), "imf_type={0} out of range [0, 5]".format(self._params["imf_type"])
-> 1283         assert (self._params["tage"] <= 0) | (
   1284             self._params["tage"] > self._params["sf_start"]
   1285         ), "sf_start={0} is greater than tage={1}".format(

AssertionError: sf_start=2.1 is greater than tage=1.55

[dfm]

Thanks! It looks to me like this is probably more a prospector issue than an fsps one, since the looping is happening there, and the implementation here looks sensible. @bd-j will have a better sense, but it might be worth moving this issue over there!

[bd-j]

Yes this is probably better as a prospector issue though python-fsps could be set up to only run check_params() before or within the call to _compute_csp(). In practice I've never run into this because I usually don't use sf_start; I think sf_start>0 is perfectly degenerate with tage=tage-sf_start; sf_trunc=sf_trunc-sf_start; sf_start=0

[k-a-webb]

Thank you both for your help. Perhaps it is a Prospector issue, however I couldn't figure out how to update the fsps parameters and call check_params() only after-the-fact.
In any case, I appreciate the advice for how to avoid the issue!

[bd-j]

ah, right I guess there's no easy way to ensure the order that the parameters are updated won't trigger the error, without putting another identical check in prospector and reordering the update there. I think having StellarPopulation. _update_params() call self.params.check_params() might be the cleanest way to solve this issue.

[dfm]

Thanks both! This makes sense to me. Are there any other places where we should add a call to check_params or is that sufficient?

[bd-j]

I think that’s the only place driver.update_* get called, so should be sufficient

[dfm]

perfect! Let's do it!

[bd-j]

@k-a-webb this issue should be resolved by #187