rbiswas4 / buildingsnmetric

Basic Idea:

We are trying to come up with metrics that are functions of Scheduler variables that will improve the science cases we are interested in. In supernovae, this has been done intuitively: We know better temporal coverage helps, better filter coverage helps and so we tend to have metrics that that reward these characteristics. What is not clear is how much we lose when we move off these ideal characteristics.

Steps

1. Take an opsim output. Pull out the output for a field using https://github.com/rbiswas4/OpSimSummary. Look at https://github.com/rbiswas4/OpSimSummary/blob/master/example/ExploringOpSimOutputs.ipynb
2. Use sncosmo realize_lc to realize a light curve
3. Use https://github.com/rhiannonlynne/OpsimObs to add observations
4. fit the light curves using snosmo.mcmc_lc , fixing z
5. Vary the number of observations by using opsimObs, randomly change the filters and time locations
6. Build a correlation from position to size of distance errors for which we have a formula

buildingsnmetric's Issues

Main steps to check

• Think of an observing sequence within a window of time as a model with the following parameters.
  {Number of visits, Delta Time_i : deviations in time, filter_i }
• Write something to get random observation sequences in a field defined by ra, dec in (mjd, filter choice) for a fixed number of visits. How do we make this sensible to avoid sampling during the day? Maybe we can pick the beginning and end times each day of observing from ENIGMA?
• use this as an input for opsimObs and get output in the form of an opsim output like ascii file
• opsimObs needs a couple of changes perhaps? send in as an array or write to ASCII and then reformat. But what we need in the ASCII finally is a change of some names fivesigmadepth - maglim, seeing - finseeing, and add fieldID and obsHistID with fake numbers. Reformat to take out too many comments. (can also be done on a new class method.
• use the method fromOpSimASCII(cls, opSimFlatFile, **kwargs) to go from opsimObs output to simlibs. The problem is that the columns may have aliases. Let us start by choosing field id 744 from opsim Enigma 1189 (ie. this gives you the ra, dec). Output a simlib using writesimlib method.
• Set a model for a SNIa using the SALT model and simulate the SN light curve using SNCosmo and the simlib. The relevant documentation page for SNCosmo is http://sncosmo.readthedocs.org/en/v1.1.x/simulation.html. This shows you how to go from an obstable to a list of SN.
• We would like to put in time translation invariance which we could do by brute force. Maybe smarter methods exist, by going to Fourier space and then padding? Don't know but if such ideas don't come up, we can continue with the brute force.
• fit light curves to get uncertainties on light curve parameters
• Asuuming constant alpha = 0.11, beta=3.1, get an error on mu for the SN.
• repeat all steps, recalculate sigma_mu. Try to run a sampler to explore the parameter space near the minimum in sigma_mu.

Double counting Poisson noise due to flux

The simlib values are such that they already consider the error due to both the point source flux and the sky background. But SNCosmo redoes this calculation in realize_lcs, by assuming that the skynoise is only due to the sky background. So, there is some double counting. Do we need a new calculation for the skynoise using the skyfilterbrightness from opsim converted to flux units per pixel?