rbiswas4 / analyzesn Goto Github PK

It has been unclear how to setup the environment variable THROUGHPUTS_DIR in Windows. Also, in fliters.py using os.getenv twice needs to be avoided.

Add the lsst throughputs directory to install and setup.

In order for fits to work with LSST, we need the LSST throughputs. This is available from the lsst sims stack, but also through using the throughputs repository

• Investigate if it is better to conda install the throughputs directory and set it up through eups or it is better to git clone/curl in the throughtputs from the git repository and add a environment variable
• Either way, use a setup.sh script to allow the user to set the throughputs directory. This way, the user having the throughputs directory already is not forced to download it again.

• Move lightcurve.py and aliases.py module from SNSims repository to this repository.
• Make SNANASims.get_SNANA_photometry return an instance of LightCurve. This involves making sure the correct minimal columns are provided.
• make sure current tests pass

infer_params is a method we are using in pipeline scripts. It would be good to transfer this into the analyzeSN package.

• fix import in snanaio.py
• fix the new names of registered LSST bands lsstu rather than lsst_u
• fix the way remap_filters is applied so that it creates a dictionary with keys in lower case, for comparison where ignore_case==True

The magnitude histogram and the redshift distribution plot shown in JLA_example.ipynb in line 18 and 19 are different from the ones shown in example notebook when I run it on my computer (Mac OS 10.9.5) with sncosmo version 1.4.0

The histogram I get is here, and the histogram shown in the example notebook is here.

note: This includes prior on redshift.

1. We need to a. register the megacam filters from the throughputs directory in the same way as LSST filters. This is important partly because the examples stored are SNLS supernovae.

2. We need a way to prevent SNANASims from defaulting to use LSST throughputs.

There should be a script to combine the installs from conda and pip to produce a combined requirements document.

Running reschar.mu_variance_linear() in the example notebook gives an AttributeError "AttributeError: 'Series' object has no attribute 'ix'"

There are two issues to sort through:

• The official megacam filters have an i and a i2 band. For SN from SNLS 3 year data, what we care about is the i band rather than the newer i2 band.
• radial dependence of the bandpasses: This will be implemented in SNCosmo. Till then I was thinking about matching up SN for their positions as was done in SNANA. However, I am unsure of how different epochs find the same set of radial positions, unless the dithers in SNLS were small compared to the gradient of the transmission values.

Allow input of SALTFormat datasets using a class method for LightCurve Class.

• Add class method for using a SALT Format
• Change the use of megacam /lsst filters so that it follows the same pattern as sdss ie. register the filters by names like megacamg and lsstg
• Allow methods
• Do examples

simdata = ans.SNANASims.fromSNANAfileroot('LSST_Ia',
                                          location='MINION_1016_10YR_DDF_v2/',
                                           coerce_inds2int=True)
simdata.get_SNANA_photometry(snid=13)

raises

172             assert np.shape(ptrs) == (2,)
    173         elif snid is not None:
--> 174             ptrs = self.get_photrows(snid=snid.strip().lower())
    175         else:
    176             raise ValueError('Both {0} and {1} cannot be None'

AttributeError: 'int' object has no attribute 'strip'

where the method assumes that snid is a string.

The LightCurve class reads in the bandpass as strings which could be 'ugrizy'. However, it might be that we need the strings to be aliased for certain programs to function. For example filters module imports lsst filters using the registered throughputs lsst_u, lsst_g etc. We should have a method of taking care of this.

Allow for having a double gaussian which can be imported by SNsims.

Add the megacam filters to throughputs in the same way as LSST filters are added

In analyzeSN/aliases.py iteritems is used which is not compatible with python 3 use items instead

@djreiss has a parser for old SNANA ASCII data (gzipped) and it would be good to split the parsing methods for SNANA to use this.

Currently, LightCurve cannot accept column aliases, and works with its set of aliases. Add the ability to add a column alias in the form of a dictionary.

While using the lsst bandpasses from THROUGHPUTS, we have added an approximate_bandpass directory that removes zeros making it possible to use finite range spectra like the SALT model at different redshifts more easily. The filters.py code importing the locations and registering bandpasses should be changed to take advantage of these.

Demonstrate the usage of LightCurve class using SNANASims as an example.

Figure out how to output verbatim strings. Or change this to a python script.

In partticular the current script tries to interpret bash variables like $HOME which I want to write out verbatim. Another major problem is that the markdown code block syntax is not being echoed correctly .

In order to write the results in ResChar to disk, we should have methods for serializing these results. Obviously, all of the results are not calculated in every method. But there are a minimal number which will work on every successful run. We need to have methods for converting these into arrays suitable for storage as a flat array of floats. The list of such quantities includes:

• parameter values for at least x0, x1, c, t0
• covariances : we need to store the upper triangular parts only. And we need to order the variables correctly.
• An (under) estimate of the uncertainty on the distance modulus where we choose standard values of alpha and beta. In principle, this cannot be obtained from a single supernova, but we can assume values of alpha and beta.

The two problems are:
_lightcurve attribute is changed by the remap_filters function. This means that the second time we call lightCurve it tries to use remap_filters on _lightcurve and fails.

in #8 , we created a copy of _lightcurve in lightCurve. This means that everytime lightCurve is called a new copy of _lightcurve will be created. This should not result in larger memory as by this time the previous copy can be deleted. However, this may turn out to be inefficient particularly in MCMC sampling.

• Investigate if this is a problem by profiling
• Use a version of lazy property to fix this.