Features

• Data containers — SingleSidedOpdData, DoubleSidedOpdData, OneSidedOpdData, SpecData, and the retro-compatible FTSData. OpdData is the geometry-agnostic entry point (from_array / read).

• Interferogram → spectrum — Forman phase correction, apodisation, zero-padding and ITG binning correction.

• Spectrum → interferogram — to_itg for forward modelling.

• I/O — FITS products registered with Astropy’s unified I/O, plus export to the IDIA HDF5 schema used by CARTA.

• Large cubes — thread-based or dask-backed blockwise processing.

• Co-addition — ftsdata_average combines cubes with per-pixel inverse-variance weighting.

• Optional kidsdata integration — opacity correction can use kidsdata.forward_model.ATM when kidsdata is installed.

Support policy

ftsdata follows SPEC 0 — Minimum Supported Dependencies for Python and core dependency support, with one explicit exception: Python 3.8 remains supported for the CONCERTO environment used by kidsdata.

On Python 3.8 the package uses the last compatible NumPy, SciPy and Astropy releases; newer Python versions use the current SPEC 0 floors. Minimum versions are raised on minor or major releases, never on patch releases. The current floors and supported versions are listed in docs/min_versions.md.

Python 3.8 and the associated Astropy 5.2 stack are targeted for removal in ftsdata 0.2.0, or once the CONCERTO environment used by kidsdata moves to Python 3.11+ and Astropy 7+.

Installation

The package is not on PyPI yet; install it from this repository (Python 3.8+):

git clone git@gitlab.lam.fr:KISS/ftsdata.git
cd ftsdata
python -m pip install -e .

Optional features are grouped in extras:

python -m pip install -e ".[all]"     # h5py, dask, matplotlib, psutil
python -m pip install -e ".[hdf5]"    # HDF5 / IDIA export
python -m pip install -e ".[dask]"    # dask-backed blockwise transforms
python -m pip install -e ".[plot]"    # cube previews
python -m pip install -e ".[docs]"    # Sphinx documentation build
python -m pip install -e ".[test]"    # pytest and pytest-cov

Only NumPy, SciPy and Astropy are required for the core containers and transforms. h5py enables HDF5 I/O, dask enables dask-backed transforms, matplotlib enables plot_cube_preview, and psutil improves memory detection.

Quick start

import astropy.units as u
import numpy as np
from ftsdata import SingleSidedOpdData

# 101 OPD samples with a 100 µm step, symmetric around the ZPD
opd = np.arange(-50, 51) * 0.1 * u.mm
data = np.random.default_rng(42).normal(size=(opd.size, 16, 16))

# from_array inspects the OPD geometry and returns the matching subclass
itg = SingleSidedOpdData.from_array(opd, data)
print(type(itg).__name__)          # DoubleSidedOpdData

# Apodise, zero-pad and FFT to a spectral cube
spectrum = itg.to_spectra(apodization=np.hanning, n_pad=3)
print(spectrum.data.shape)         # (nfreq, 16, 16)
print(spectrum.spectral_axis.to("GHz")[:3])

spectrum.write("spectrum.fits", overwrite=True)

Read a product back with a compatible class:

from ftsdata import FTSData

cube = FTSData.read("spectrum.fits")
print(cube.data.shape)

Data model

All containers derive from astropy.nddata.NDDataArray and carry:

I/O and co-addition

FITS products use the Astropy unified I/O interface:

from ftsdata import FTSData

spectrum.write("cube.fits", overwrite=True)
spectrum.write("cube-compressed.fits", compress=True, overwrite=True)
cube = FTSData.read("cube.fits")

For CARTA, write the IDIA HDF5 layout directly — the .hdf5 extension selects the registered writer and reader:

from ftsdata import OpdData

cube.write("cube.hdf5", mip_levels=[2, 4, 8], dtype="float32")
cube = OpdData.read("cube.hdf5")   # returns the matching geometry subclass

The writer creates the /0 group and the VERSION attribute required by the IDIA schema. Pass format="hdf5" explicitly when the filename lacks an .hdf5 or .h5 extension.

Co-add cubes that share the same WCS grid:

from pathlib import Path
from ftsdata import FTSData, ftsdata_average

cubes = [FTSData.read(path) for path in sorted(Path("products").glob("*_T.fits"))]
combined = ftsdata_average(cubes)
combined.write("combined_T.fits", overwrite=True)

Relationship with kidsdata

ftsdata does not import kidsdata at module import time, so it can be used on its own. When the kidsdata distribution is installed, two integrations become available lazily:

• available_memory() uses kidsdata.utils.available_memory when available, which knows about SLURM allocations.

• SpecData.correct_opacity() uses kidsdata.forward_model.ATM by default. Without kidsdata, pass a compatible model explicitly:

  cube.correct_opacity(elevation=60, pwv=1.0, atm=my_atm_model)

Existing code can keep using the old import path:

from kidsdata.ftsdata import FTSData

New code should import from ftsdata directly.

Documentation

The Sphinx sources live in docs/:

python -m pip install -e ".[docs]"
python -m sphinx -b html docs docs/_build/html

The built pages are published in three places from the main branch:

• GitLab Pages: https://ftsdata-7333b4.docs.lam.fr

• GitHub Pages: https://abeelen.github.io/ftsdata/ (GitHub Actions)

• Read the Docs: https://ftsdata.readthedocs.io/ (configured by .readthedocs.yaml)

Tests

python -m pip install -e ".[all,test]"
python -m pytest --pyargs ftsdata -q

The GitLab and GitHub CI pipelines run the test-suite on several Python versions and build the documentation with warnings treated as errors.

Repository layout

ftsdata/            package modules and pytest suite (ftsdata/tests)
docs/               Sphinx sources
licenses/           license files
pyproject.toml      packaging and tool configuration
tox.ini             test and documentation environments
.gitlab-ci.yml      tests and GitLab Pages
.github/workflows/  GitHub Actions tests and Pages

License

BSD 3-clause. See licenses/LICENSE.rst.