obscode / snpy Goto Github PK

I'm a final year physics undergrad modelling photometric data on sn2022vqz. When attempting to fit or call the template class with method="gp", I get error messages saying that Mean(const) is not a valid attribute of gp object. Browsing the pymc codes on GitHub, there seems to be Mean and Covariance objects, although the keyword arguments the Covariance constructor may take do not match those input in the fit1dcurve.py of SnooPy. Not finding what the first gp.matern.euclidean in PyMC past documentations on their website, I couldn't replace it on my local machine.

Hi, I’m a grad student attempting to use snpy to fit some SNe Ia data. Specifically I’m attempting to use the fitMCMC method to enforce priors. I have two corrections which I believe should be made:

1. In the documentation for fitMCMC (https://users.obs.carnegiescience.edu/cburns/SNooPyDocs/html/fitting_MCMC.html), the first fit is run using

s.fit(Rv=2.0)

This correctly fixes the Rv parameter. However, the MCMC fit is run using

s.fitMCMC(bands=['u','B','V','g','r','i','Y','J','H'], R_V="N,2.3,0.9")

R_V is not a parameter, Rv is. I believe passing this extra argument does nothing (e.g. if I pass foobar=’N,2.3,0.9’ then the fit will run without errors, but obviously that’s not a parameter).

I believe that R_V should be changed to Rv in the MCMCfit documentation.

2. However, making sure parameter names match introduces an error. Here’s the partial traceback when I try to enforce a prior on EBVhost:

emcee: Exception while calling your likelihood function:
  params: [3.25815455e+01 1.55605255e+00 5.81841998e+04]
  args: ({'varlist': ['DM', 'dm15', 'EBVhost', 'Tmax'], 'fitflux': True, 'free': ['DM', 'dm15', 'Tmax'], 'DM': {'fixed': False, 'index': 0, 'prior_type': 'model'}, 'dm15': {'fixed': False, 'index': 1, 'prior_type': 'model'}, 'EBVhost': {'value': 'U,0,1', 'fixed': True}, . . .
. . .
"/Users/jamisonfrost/miniconda3/envs/earlysne/lib/python3.7/site-packages/snpy/model.py", line 473, in __call__
    temp = temp + self.Robs[band]*self.EBVhost + R*self.parent.EBVgal
TypeError: can't multiply sequence by non-int of type 'numpy.float64'

The parameter is getting fixed as a string, instead of translated into the desired prior. In order for the built-in priors (uniform, exponential, and normal) to be instantiated, I believe this line should be changed:

Currently it’s checking for an arg with type bytes, and I think this should be changed to str. Passing the argument as a byte string doesn’t work, I get ValueError: I don't understand the prior code b'U,0,1' (since it checks if st[0] in [‘U’, ‘G’, ‘E’], and b’U,0,1’[0] is apparently 85). I’ve tested it and making the change to str allows priors to be enforced as expected.

For completeness, I’m running this on python 3.7.9, snpy version 2.5.3.

I have been using SNooPy to plot light curves in 3 bands and have been plotting them on a single plot with each band offset to allow for better viewing. A feature I think would be really nice would be to be able to plot residuals under this main plot to show how well the model curves fit to the data. Perhaps there could also be the option to specify which bands you'd want to plot the residuals for. This would be a really useful feature.