Tasks

Top Level

• Complete Gaussian line fitter

Prior Framework

• Implement dependent prior for covariant Tex/N

Parameter products

• Parameter maps of marginals (maximum likelihood, median, etc)
• Property cubes by LSR velocity

HDF5 post-processing

• Re-organize datasets into hierarchy of groups (metrics, quantiles, pdfs, etc.)

Visualization

• Parameter map plots, number of components
• Corner plots using GetDist. See Krumholz & McKee (2019) for example of good corner plots.

Performance

• Test impact of chunking on HDF5 write speeds.

Testing

• Check astropy package template repository for example of pytest
• Refactor tests for (1) consistency with pyspeckit and (2) benchmarks.

Applications

• Other molecular lines with hyperfine structure: N2H+, HCN
• Faraday rotation measure
• Zeeman effect profiles
• RADEX grid interpolation for model fitting

Completed

• Refactored prior transformers to accept arrays from Python and not have it hard-coded for the IRDCs within the Cython code.
• Refactored prior transformation function into Prior objects for each variable. Now it is more flexible to swap in different kinds of priors (free independent, ordered, dependent) for each parameter.
• Change to folder-based HDF5 store file to hide process-based files and virtual store. Create a wrapper class StoreFile that takes a number of threads and handles the linking.
• Refactor CubeStack to remove dependency on pyspeckit, only spectral-cube. The implementation now uses the pyspeckit code to generate the spectral axes, but uses spectral-cube for the cubes and not the pyspeckit CubeStack class.
• Add "integrated intensity" to the HDF5 as a post-processing step. Compute the spectra from the maximum a posteriori value and then discrete sum along the axis to get summed brightness temperature (multiplied by channel width to get K km/s). Add method to spectrum class to sum spectrum and return value. An alternative method is to add an extra (unused) parameter in the MultiNest run, and store the value in utheta. This way the integrated intensity of all posterior samples will be stored.
• Weighted model prior for updated number of components. The convolved products are pre-pended with the "conv_" label.
• PDF cubes for each parameter. These are stored under "post_pdfs" and "conv_post_pdfs".
• Implemented dependent velocity prior with forced separation offset based on the widths of both components, the "resolved width prior". This should work for other parameters that are similarly functionally dependent, but is meant specifically for centroid and dispersion.
• Added flag for using the kinetic temperature and converting to a rotation temperature using the Swift et al. approximation ("cold ammonia").
• Abstracted the model framework for other models. New extensions may be added in the nestfit/models directory or built locally by including the nestfit/core/core.pxd declarations.