Add a figure for 1d visibility distributions

MPoL-dev / MPoL

A flexible Python platform for Regularized Maximum Likelihood imaging

https://mpol-dev.github.io/MPoL/

MIT License

33 stars 11 forks source link

Add a figure for 1d visibility distributions #213

Closed jeffjennings closed 9 months ago

jeffjennings commented 9 months ago

~~NOTE: This should only be reviewed after #212 is merged into main and main into here.~~ [done]

Adds a plotting function to generate a figure with 1d visibilities: observed, mpol model, residual. Plots both real and imaginary, can plot either projected or deprojected (if a geometry is supplied). Will be used to generate quick looks at model performance at end of cross-val loop.

jeffjennings commented 9 months ago

Example figure with projected visibilities: $IMLup_dartboard_{'entropy' {'lambda' 0 0001, 'guess' False, 'prior_intensity' 5e-07}, 'sparsity' {'lambda' 0 001, 'guess' False}}_projected_vis$

With deprojected vis: $IMLup_dartboard_{'entropy' {'lambda' 0 0001, 'guess' False, 'prior_intensity' 5e-07}, 'sparsity' {'lambda' 0 001, 'guess' False}}_deprojected__rescaled_vis$

jeffjennings commented 9 months ago

Yes that's right, if they'res using the same binning there would be perfect correspondence in principle. E.g. since the points at short baseline aren't matching, it quickly indicates the model is underestimating the total flux. Or if there were a source that were highly asymmetric, seeing the structure in Im(V) and how well the model is tracing that could be interesting. I also looked at plotting the 2D residual visibilities, but haven't thought of a way to make it clear what's happening by eye (unless the fit is very bad), short of the imaged 2D residuals in the image_comparison_fig output.