virgil.ensemble

Families drawn uniformly: "tv", "tsv", "maxent", "starlet" or "gp" (gp_members).

Per family, the (low, high) weight per data point; n_weights (at least 3) log-uniform weights are drawn, one per bin of log w.

Pixels per Nyquist pixel.

The field is field_mas times one of these, at most field_of_view(data) (and 1 on a uv lattice).

Base field (default image_field) and a field-of-view ceiling.

Starts drawn uniformly: "moments", "dirty" (needs phases), "flat".

Most pixels on a side; larger fields get coarser pixels, then a crop.

A group keeps weights from corner / 10**window_dex up to the corner.

Drop members whose raw χ² per point on any dataset exceeds chi2_ratio times the best's, then those more than mad_cut robust deviations above the median total. Both are relative to the members, so if all fit badly all pass: summary counts the drops per reason, and the pipeline's χ² checks judge the fit itself.

Deprecated absolute cut; use chi2_ratio.

Largest recentring shift without a star (default: beam major axis).

Tolerance on the mean's χ² excess over the best member's (default sqrt(2/N); 0 often keeps one member).

combine warns below this many members.

GP evidence grid (evidence_grid): starting amplitudes and lengths (beam minor axes), limits, growth steps.

GP members are fits within this log evidence of the best.

Add anisotropic fields if a Gaussian fits below this axis ratio.

Position of the group in the ensemble.

The regularizer family, or "gp".

Pixels on a side.

Pixel size in mas.

The starting image.

The regularizer weights, largest first (empty for "gp").

The group's settings.

Its fits, one per weight, largest weight first; None for a Gaussian-process group.

For a Gaussian-process group (gp_members), the fits it offers as members.

For a Gaussian-process group, the evidence grid they come from.

The settings of its group.

Its regularizer weight (NaN for a Gaussian-process member).

The fit.

Raw χ² per data point of each dataset (with the quoted errors).

Whether it is in the mean image.

For a Gaussian-process member, its hyperparameters and evidence.

Why it was left out: "window" (outside its group's L-curve window), "diverged" (non-finite χ²), "chi2" (a dataset fitted too badly), "outlier" (total χ² far above the others) or "mean" (adding it made the mean fit a dataset worse). None if kept.

The mean scene, exactly: the kept members' images on their own grids (member0, member1, ...), each carrying its share of the flux, and with a star a star component. chi2_red and trace are of this model.

The mean extended emission on a common grid (the finest pixels and the largest field of the kept members), summing to one; with a star its flux is the mean flux relative to the star. For display and scoring: resampling smooths it, so it fits the data less well than model.

Standard deviation across the kept members of mean.brightness, pixel by pixel.

Raw χ² per data point of the mean scene on each dataset.

chi2_red of the running mean after each accepted member; on every dataset it stays within spec.mean_rtol of the first entry.

Every reconstruction, kept or not.

The groups, with their L-curves.

"iso" or "aniso".

Correlation length in mas (for "aniso" the geometric mean of the two lengths).

Standard deviation of the log-brightness.

log_evidence of the fit.

The fit.

For each variant ("iso", and "aniso" if the data are elongated), the grid's axes: correlation lengths in mas and amplitudes, ascending.

Per variant, the log evidence, with rows over lengths_mas and columns over sigmas.

(variant, length_mas, sigma) to the GPFit.

The fit of largest evidence.

Whether the best point is inside its grid's edges: on an edge the optimum may lie beyond it, because the growth stopped at a limit.

The elliptical-Gaussian fit's minor-to-major ratio and position angle (None if it was not made).