Super, Mega, Ultra Pre-Alpha Users

[JSKenyon]

I am starting this issue to discuss basic installation and usage of QuartiCal.

Installation

pip install git+https://github.com/JSKenyon/QuartiCal.git

Hopefully it will remain this easy. Note that QuartiCal is strictly Python3.

Usage

QuartiCal can be invoked similarly to CubiCal:

goquartical [.yaml file] {args}

For a full list of currently (probably) working args, run:

goquartical -h

QuartiCal has a convenience function which can be invoked as follows:

goquartical-config [name.yaml]

which will generate an (optionally named) .yaml file containing the default configuration.

Caveats

• This is very much still under development and some very basic functionality is missing (e.g. maxiter is hard-coded as the interface which is required to pass it in is a WIP.)
• No PRs will be considered at present. This is just to protect the code base until such time as I have a more rigorous testing framework in place and the interfaces have solidified. The plan is to make QuartiCal as easy to maintain and update as possible.
• Not all CubiCal solvers have been reimplemented yet. These will come with time. For now diagonal and 2x2 chains of complex gains are supported. Parameterised phase-only terms are supported in diagonal chains. This is a proof of concept - parametric terms WILL be supported in both diagonal and non-diagonal chains, but this requires lots of complexity in the kernels which I have not yet tackled.
• There is a VERY EXPERIMENTAL Kalman filter-smoother implementation. This does not yet support chains, but does seem to work. I encourage users to experiment with it and report back on performance.
• CubiCal's recipe specification has been replicated in QuartiCal. However, it has only undergone limited testing. Please report failures.
• Issues are welcome and encouraged. Particularly issues pertaining to existing features which are currently broken. Issues can also be used to help me prioritise missing features.
• Almost no logging. The reason for this is because logging is combination of expensive and difficult when using dask. The ultimate solution to this will be to log summary statistics as chunks are completed. This will happen eventually.

[JSKenyon]

@o-smirnov @IanHeywood this issue is largely to help you, should you decide to get your feet wet. :-)

[o-smirnov]

Awesome! What do you mean by "CubiCal's recipe specification"?

[JSKenyon]

I mean that it supports the "COLUMN1:LSM + COLUMN2" style syntax for constructing model data. I think that the only major difference is that I changed negation/subtraction to be represented by "\~" rather than "+-", as I found the latter very confusing. Note that it important to always wrap the recipe in string quotes to make sure bash doesn't interpret "~" as a path.