Configuration ============= Search grids ------------ A grid YAML (``luv-find --grid``, a run's ``grid``, or the ``cfg`` of :func:`~luv_finder.matchedfilter.build_grid`) sets the search axes. A value is a scalar (fixed), a list, or ``{start, stop, step}``; the grid is the product of all axes, so the cost is too. Omitted keys follow from the data: ``dra``, ``ddec`` arcsec east and north of the reference; default a lattice in steps of half the resolution out to where the primary beam falls to ``pb_limit``. ``pb_limit`` primary-beam level that bounds a pointing's search (default 0.2). ``bmaj``, ``bmin``, ``pa`` the trial source's axes (FWHM, arcsec) and position angle (degrees east of north); default 0.2355 of the resolution (a sigma of a tenth), effectively a point source. ``width`` line FWHM in km/s; default 100, 200, 300 and 400. The line centre is not an axis: every template is centred on every channel. Any other key raises an error; ``total_flux`` in particular cancels in the normalisation. :class:`~luv_finder.matchedfilter.MatchedFilter` takes the grid as a dict of the same keys, a value or a list each; a key left out takes its value from ``GRID_DEFAULTS``. .. code-block:: yaml # configs/grids/default.yaml (comments trimmed): a blind search pb_limit: 0.2 width: [50, 100, 300, 500] # km/s Filter options -------------- ``continuum_order`` (``luv-find --continuum-order``) degree of the polynomial continuum fitted with each line template per window and position (default 1; ``none`` turns it off). A line loses the part of its S/N the polynomial can mimic, a few per cent. Lines at S/N >= 5 are left out of the continuum of the others at their position. ``channel_noise`` ``"measure"`` (default) measures the channel correlation on the jackknife per spectral window; ``None`` assumes independent channels; or pass ``{spw: ChannelNoise}``. ``cores`` (``luv-find --cores``) the search pins itself to a quarter of the cores, at most all but two, because the machines are shared (Linux only). Mock presets ------------ ``configs/mocks/`` holds :class:`~luv_finder.mock.MockObservation` inputs (``luv-mock ``) and ``configs/grids/`` the matching search grids, paired by file name. A mock's declared ``snr`` is the optimal S/N, calibrated against the simulated data. Antenna configurations are not vendored: ``alma_config`` names a file CASA ships, such as ``alma.cycle13.3.cfg``. The presets and their pairing are listed in `configs/README.md `_. Runs ---- A run YAML (:doc:`../quickstart`) holds ``data`` and ``output`` and optionally ``grid``, ``pb_limit``, ``max_channels``, ``uv_min``, ``filter``, ``reliability``, ``near_miss`` and ``cores``. ``uv_min`` (kλ at the data's central frequency) keeps only the longer baselines, for example to filter a cluster's SZ decrement out of the search; the exports hold the cut, so another cut needs a new output directory. The search is planned in units: windows that overlap or whose gap is narrower than the widest template are one unit, so a line across a window edge is searched and fitted whole. ``max_channels`` cuts a long unit into chunks of whole blocks (windows overlapping by more than a channel) that load their neighbours and catalogue only the groups peaking in their own part, so the chunks give the unchopped catalogue.