simonsobs / pspipe Goto Github PK

For testing parameter recovery and the analytical covariance matrix computation, we need some semi-realistic lensed CMB + foreground maps.
We would need both galactic and extra galactic foreground in healpix and CAR pixellisation (we can project from one to another using the so map class).
We can probably start with simple gaussian extra galactic foregrounds and then iterate.

Seems like there is no need to multiply a binary mask twice?

Write a bunch of function that create the spectra covariance matrix. Both analytically and using monte carlo.
A place holder can be found in pspy/so_cov.py

The code in https://github.com/simonsobs/PSpipe/blob/master/project/maps2params/python/maps_to_params_prepare_sim_data.py does not seem consistent with current mflike master. Obsolete? If not, may need updating for simonsobs/LAT_MFLike#86 (along with simonsobs/pspipe_utils#53).

In order to install pspy and make it work with python3 I had to:

1. change ConfigParser to configparser in so_config.py
2. put parentheses around a bunch of print statements in so_dict.py
3. change except ImportError, exc: into except ImportError, exc: in so_mpi.py

I can open a PR with these if useful

Write a bunch of function that do spherical transform, compute power spectra, bin them, and debiased the spectra using the mode coupling matrices.
A place holder can be found in pspy/so_spectra.py

This is a general issue for discussing the logic of the power spectrum pipeline.
We will have tons of map to cross correlate, from different detector arrays, different frequency channel, possibly add the Planck data.
The code structure should be adaptable enough to deal with this.

The LAT scanning strategy will probably lead to an inhomogeneous scanning pattern leading to inhomogeneous noise properties. Simulations of this effect are needed to test the optimality of the estimator.

I'd like to add some testing at some point, and it seems like the first step is to replace the travis config with some github configuration.

FYI, pip installing mpi4py in the docker environment doesn't seem to be possible.

As suggested by @adrien-laposta the signal simulations should have a seed, determined by the sim task number like the mnms noise sims.

I'm not sure where the signal sim gets the seed from -- numpy?

Ground pick up hurts us badly in AdvACT and might be there in the LAT data.
We want to construct a bunch of filters that can be applied to the data as the preprocessing stage.
A place holder can be found in pspy/so_map_preprocessing.py

Write a bunch of python functions allowing to generate mode coupling matrices from window function.
The mode coupling matrices should be able to handle Cls and Dls and pure/standard B mode estimation. They should be general enough to also produce the input for the covariance matrix calculation (see appendix of https://arxiv.org/pdf/1609.09730.pdf).
A placeholder can be found in pspy/so_mcm.py

Write a bunch of python functions allowing to create window functions for power spectrum estimation.
These windows should ideally be formed from:
-Masks: point source, galactic, extended sources, cluster, missing pixels
-Hit count/Noise properties
-Amount of cross linking in the data
-Apodisation routine
Ideally you will use the so map class provided in pspy/so_map.py allowing to handle both CAR and Healpix pixellisation. A placeholder can be found in pspy/so_window.py.
One important thing to be done is to code up a general distance function, that is, a function that return the distance to the closest masked pixel for HEALPIX and CAR pixellisation.