Quickstart¶
A quick look¶
luv-find searches a measurement set or NPZ, corrects every pointing for its
primary beam, combines them, saves the best grid point’s S/N spectrum to plots/
and prints it. The smoke preset (ACA, 24 channels) runs in seconds:
luv-mock configs/mocks/smoke.yaml
luv-find --ms output/ms_files/smoke/smoke.aca.cycle13.noisy.ms --grid configs/grids/smoke.yaml --jackknife
--field N searches one pointing, --out response.npz keeps the responses, and
--continuum-order and --cores set the filter.
A catalogue¶
Catalogues come from a run: one YAML file in, every product in one directory.
# run.yaml in the repository root; relative paths are relative to this file
data: output/ms_files/line13_line9/line13_line9.alma.cycle13.3.noisy.ms
output: output/line13_line9
grid: configs/grids/line13_line9.yaml # or the grid itself, as a mapping
Optional keys are name, pb_limit, max_channels (the channels a unit of the
search holds at most), uv_min (kλ: only longer baselines), filter (continuum_order, channel_noise),
reliability (floor, k, fidelity_min, rule), near_miss
(fidelity, max) and cores; an unknown key raises an error.
luv-run run.yaml search # export, search and catalogue every unit; --workers N shares them
luv-run run.yaml fit # fit the detections and near misses in the visibilities
luv-run run.yaml summary
luv-run run.yaml plots
The output holds catalogue.ecsv (every group, with its likelihood ratio, fidelity
and detected), lines_fit.ecsv (the fitted lines), continuum.ecsv (the
fitted continuum sources), line_maps.npz (moment-0 maps), export/ and units/ (each unit’s catalogue, maps and weights).
What is on disk is not redone, and Run.open(output) reopens a run for its figures.
Python¶
from luv_finder.data import load
from luv_finder.matchedfilter import MatchedFilter, build_grid
data = load("line13_line9.npz", [0])
mf = MatchedFilter(data, build_grid(data, {"width": [200.0, 300.0]}))
mf.run(jackknife=True)
print(mf.best_params, mf.response.max())
mf.response is the statistic T(x): unit variance under H0, so the peak height of a
line is its SNR, and mf.response_jackknife shows the noise on the same axis.
Column i is a template centred on trial frequency i; rows follow
mf.grid_params, the product of the dra, ddec, bmin, bmaj, pa, width axes in
that order. A mock’s declared snr is the SNR the filter recovers.