No documentation yet, see the tutorial

Models

Base class Model, subclasses DustModel etc. Customization by:
- tweaking parameters of methods
- create subclass of DustModel
- new subclass of Model
Arguments specified, not from dictionary
Add new classes for multi component models, not list of dictionaries as inputs.

Model Setup:

Model functions will still exist to construct e.g. 'd4' from the component classes. Will return Sky objects.
Within init of each component do the reading of maps etc.

Parallelism:

Combination of MPI and Dask? How to interface with toast at same time.
Assuming no Dask, use MPI4Py.
Inputs read per process or broadcast?
- ordering issues: don't convert ring and nest in one process.
- conflicts between processes reading same file.
Distributing maps in libsharp
- scatterv and gatherv
keep get_emission mostly local, also has access to communicator

Object oriented structure

Reduce levels of abstraction. Initialize components from string. Keep pysm.Sky as convenience initialized from string or from component objects.

Interface with `Instrument`

Most important factor is bandpass integration.

Try as Component.get<sub>emission</sub>(bandpass) and Sky.get<sub>emission</sub>(bandpass).
Smoothing via libsharp, and trivial healpy routine separate to Instrument object, and called from Sky.

Extra Galactic

Current maps 2048, each 192 MB. Code exists for spline interpolation. Available for validation, determine the type of interpolation as a study. In the first instance:

CIB ready to go at 2048 for validation.
Incorporate into PySM kSZ and Compton-y.
Need to code from scratch kSZ and Compton-y.

In second stage:

Lenspix / taylens used on kappa map to lensed primary

Maintain Example usage:

sky = pysm.get_model("s1", "d3")
cmb = pysm.cmb("c1")
sky.get_emission(nu)
sky.apply_bandpass(instrument_settings)
sky.apply_smoothing(instrument_settings)

hoangducthuong/pysm