HD 143006 imaging tutorial overview

[iancze]

This issue serves as a summary or table of contents for all other issues part of this tutorial.

The overarching goal is to create a comprehensive, multi-chapter set of tutorials covering RML imaging with real data. A noteworthy, accessible dataset is the DSHARP HD143006 continuum measurement set.

The scope of the tutorials should be broad. Basically, we want a user, completely unfamiliar with RML, to start reading the tutorial series and end up with a solid understanding of what they need to do in order to try imaging their own datasets. In general each tutorial should try to be self-contained without duplicating information already covered in the shorter tutorials for the ALMA logo mock dataset.

Some of the beginning data preparation tasks are best covered in the visread and they are noted as such. A npz file containing the processed visibilities is available here.

Part I (#61 )

• Download continuum data from DSHARP site (visread)
• Plotting and examine DSHARP CLEAN FITS (visread)
• Extract visibilities using CASA table tools (visread)
• Examine whether weights are scaled correctly (visread)
• Make dirty image using gridder

Part II (#62)

• Set up MPoL optimization loop, including residual imager
• Initialize model to dirty image
• Explore unregularized fit + training loop w/ Tensorboard
• Explore basic cross-validation and hyperparameter testing w/ Tensorboard

Part III (#63)

• Explore "production-ready" scripting layouts (no Jupyter notebooks)
• Explore thorough hyperparameter testing with Ray Tune

Describe the solution you'd like
We have some preliminary images of HD 143006, which shows that we can get interesting results (same arcsinh stretch as DSHARP)

[RCF42]

For this tutorial should the final product stay broken up like how it is under examples/HD143006?

[iancze]

I was thinking that the scripts in examples/HD143006 would form the basis of the "production-ready" tutorial (part III). I think Parts I and II would each be their own individual jupyter notebook rendered as a tutorial.

[RCF42]

I am slightly confused on the scope of the tutorial because it says it should be self contained and provide a solid understanding of everything but also it says to not duplicate any info from the ALMA logo tutorials so should things like cross validation and initializing to the dirty image be described again or no? I guess one thing would just be, is there the assumption that all the previous tutorials have already been gone over before someone would look at this one?

[iancze]

It will be a judgement call as to how much detail to provide in each tutorial, but I think we should aim to be as efficient as possible so as not to bog down readers. The assumption will be that the reader has at least glanced at the previous tutorials.

To give an example for the initializing from the dirty image. This tutorial should do this in the code (to speed optimization), but it doesn't need to go into depth explaining how or why this is being done. A brief 2-sentence summary of the underlying idea and then a link to the optimize from dirty image tutorial is sufficient.