gaiahealpixcache 0.4.0

Download and cache a lightweight version of the Gaia catalog for offline work

gaiahealpixcache

Download and cache a lightweight version of the Gaia DR3 catalog for offline work.

This tool follows the official HEALPix level 8 partitioning of the Gaia archive, enabling on-demand partial download with a pure NumPy backend. Much faster than querying the online archive for large amounts of data.

Installation

pip install gaiahealpixcache

Or from source:

git clone https://github.com/betoule/gaiahealpixcache.git
cd gaiahealpixcache
uv venv
source .venv/bin/activate
uv pip install -e .

Quick Start

import gaiahealpixcache

# Query sources around a sky position
sources = gaiahealpixcache.query(ra_deg=76.377, dec_deg=52.831, radius_arcmin=30)
print(f"{len(sources)} sources found")
print(sources["source_id"][:5])
print(sources["phot_g_mean_mag"][:5])

Tiles are downloaded on first access and cached as compressed NumPy arrays for fast subsequent queries.

Rectangular Region Query

For rectangular sky regions, use query_rectangular with ra_min, ra_max, dec_min, dec_max:

sources = gaiahealpixcache.query_rectangular(
    ra_min=76.0, ra_max=78.0,
    dec_min=52.0, dec_max=54.0,
)

RA wrapping across the 0/360 boundary is handled automatically (e.g., ra_min=358, ra_max=2).

The spectroscopy equivalent is query_spectra_rectangular:

meta, flux = gaiahealpixcache.query_spectra_rectangular(
    ra_min=76.0, ra_max=78.0,
    dec_min=52.0, dec_max=54.0,
)

Products

A product defines which columns are loaded from the Gaia archive and which rows are kept. Multiple products can coexist, each with its own cache namespace.

Default Products

Product	Description	Filter
source	Full Gaia source catalog (selected columns)	none
bright_sources	Sources with G < 16	phot_g_mean_mag < 16
sampled_spectra	Sampled mean spectra (343 flux points)	none
continuous_spectra	Continuous mean spectra (calibrated & sampled via gaiaxpy)	none

# Use the bright_sources product to save disk space
sources = gaiahealpixcache.query(76.377, 52.831, product="bright_sources")

The default column set is: source_id, ra, ra_error, dec, dec_error, parallax, parallax_error, pmra, pmra_error, pmdec, pmdec_error, phot_g_mean_mag, phot_bp_mean_mag, phot_rp_mean_mag, radial_velocity, radial_velocity_error.

Custom Products

Define a product with a custom column selection and optional row filter:

from gaiahealpixcache import GaiaProduct, register_product, query

my_product = GaiaProduct(
    name="astrometry_lite",
    url="https://cdn.gea.esac.esa.int/Gaia/gdr3/gaia_source/",
    md5sum_file="_MD5SUM.txt",
    file_prefix="GaiaSource_",
    file_ext=".csv.gz",
    columns=["source_id", "ra", "dec", "parallax", "pmra", "pmdec", "phot_g_mean_mag"],
    where="(parallax > 0) & (phot_g_mean_mag < 18)",
)
register_product(my_product)

sources = query(76.377, 52.831, product="astrometry_lite")

Products are persisted to ~/.config/gaiahealpixcache/products/ and survive sessions.

# List all available products
print(gaiahealpixcache.list_products())

# Remove a custom product (also cleans its cached data)
gaiahealpixcache.unregister_product("astrometry_lite")

Filter Expressions (where)

The where field accepts a Python boolean expression evaluated over the loaded data. Column names refer to Gaia column names:

# Good proper-motion candidates
where="abs(pmra) > 10 and abs(pmdec) > 10"

# Bright sources with parallax
where="(phot_g_mean_mag < 14) & (parallax > 0)"

# Use numpy functions
where="np.isfinite(phot_g_mean_mag) & np.isfinite(parallax)"

Filter expressions are validated by AST analysis and evaluated in a restricted namespace (column arrays + np only). No filesystem access or arbitrary imports are possible.

Spectroscopy

The sampled_spectra product provides Gaia sampled mean spectra — 343 flux points per source spanning 336–1020 nm.

import gaiahealpixcache

# Query spectra around a sky position
meta, flux = gaiahealpixcache.query_spectra(
    ra_deg=76.377,
    dec_deg=52.831,
    radius_arcmin=30,
)
print(f"{len(meta)} sources with spectra")
print(meta["source_id"][:5])
print(flux.shape)  # (num_sources, 343)

query_spectra returns a tuple of (meta, flux):

• meta: structured array with source_id, ra, dec
• flux: 2D float array with shape (num_sources, 343), in W/m²/nm

Spectro tiles are cached separately as .npz files. The same where filter and product system applies:

# Custom spectro product with narrower flux range
from gaiahealpixcache import GaiaProduct, register_product, query_spectra

spec_product = GaiaProduct(
    name="spectra_uv_only",
    url="https://cdn.gea.esac.esa.int/Gaia/gdr3/Spectroscopy/xp_sampled_mean_spectrum/",
    md5sum_file="_MD5SUM.txt",
    file_prefix="XpSampledMeanSpectrum_",
    file_ext=".csv.gz",
    columns=["source_id"],
    spectro=True,
    spectro_meta_cols=["source_id", "ra", "dec"],
    spectro_flux_cols=(4, 100),  # first 96 flux points only
)
register_product(spec_product)

meta, flux = query_spectra(76.377, 52.831, product="spectra_uv_only")

Continuous Spectra (via gaiaxpy)

The continuous_spectra product downloads XP continuous mean spectra and processes them through gaiaxpy to produce calibrated, sampled flux arrays — identical in format to the sampled_spectra product. The continuous spectra archive is more complete (more sources available) than the pre-sampled archive.

Requires gaiaxpy to be installed:

pip install gaiahealpixcache[spectro]

Usage is identical to sampled_spectra — just pass product="continuous_spectra":

import gaiahealpixcache

meta, flux = gaiahealpixcache.query_spectra(
    ra_deg=76.377,
    dec_deg=52.831,
    radius_arcmin=30,
    product="continuous_spectra",
)
print(flux.shape)  # (num_sources, 343)

If gaiaxpy is not installed, a clear ImportError is raised guiding you to install the [spectro] extra.

Wavelengths

Retrieve the wavelength array corresponding to the flux points:

import gaiahealpixcache

wavelengths = gaiahealpixcache.spectro_wavelengths()
print(wavelengths)  # [336., 338., 340., ..., 1018., 1020.]

Wavelengths are 343 values from 336 to 1020 nm with a step of 2 nm. For custom products with a narrower spectro_flux_cols range, only the corresponding wavelengths are returned. The same function works for the continuous_spectra product.

Catalog Matching

Match two catalogs by source_id using an efficient inner join:

import gaiahealpixcache

# Query photometry and spectra separately
sources = gaiahealpixcache.query(76.377, 52.831)
meta, flux = gaiahealpixcache.query_spectra(76.377, 52.831)

# Find common sources
idx_a, idx_b = gaiahealpixcache.match_catalogs(sources, meta)
print(f"{len(idx_a)} sources have both photometry and spectra")
print(sources["source_id"][idx_a][:5])
print(flux[idx_b].shape)

match_catalogs returns index arrays so that cat_a[idx_a][k] and cat_b[idx_b][k] correspond to the same source. Uses a hash-based algorithm with O(n+m) time complexity.

Coordinate Transforms

Topocentric Conversion

Convert ICRS catalog coordinates to apparent topocentric positions:

import gaiahealpixcache
from astropy.time import Time

now = Time.now()

sources = gaiahealpixcache.query(ra_deg=76.377, dec_deg=52.831)

topo = gaiahealpixcache.gaia_to_topocentric(
    sources,
    mjd=now.mjd,
    lon_deg=5.71,
    lat_deg=43.93,
    height_m=640.0,
)
print(topo["ra_apparent_deg"][:5])
print(topo["alt_deg"][:5])

Coordinate Normalization

Coordinates outside the standard convention (RA in [0, 360), Dec in [-90, 90]) are automatically normalized. You can also call the helper directly:

ra, dec = gaiahealpixcache.conform_coordinates(-10.0, 95.0)
# ra=170.0, dec=85.0

Cache Management

cache_dir = gaiahealpixcache.get_cache_dir()
print(f"Cache location: {cache_dir}")

gaiahealpixcache.clear_cache()

Cached tiles are stored as .npy files named with the product's configuration hash, so different products and filters maintain separate cache entries.

Custom Cache Directory

By default, the cache follows the XDG Base Directory specification (~/.cache/gaiahealpixcache on Linux). Override it with the GAIAXCACHE environment variable:

export GAIAXCACHE=/path/to/my/cache

This is useful when the default cache location has limited disk space or when you want to share the cache across projects.

Concurrency

Each cached tile is protected by a file-based lock (fcntl.flock on Unix). When two processes or threads request the same tile simultaneously, only one downloads it; the other waits and loads the result once it's ready. Downloads use atomic rename — if a download is interrupted, the partial file is discarded, preventing corrupted cache entries.

API Reference

Function	Description
query(ra_deg, dec_deg, radius_arcmin, product)	Query Gaia sources within a circular region
query_rectangular(ra_min, ra_max, dec_min, dec_max, product)	Query Gaia sources within a rectangular region
query_spectra(ra_deg, dec_deg, radius_arcmin, product)	Query Gaia spectra within a circular region (supports sampled_spectra and continuous_spectra)
query_spectra_rectangular(ra_min, ra_max, dec_min, dec_max, product)	Query Gaia spectra within a rectangular region (supports sampled_spectra and continuous_spectra)
gaia_to_topocentric(catalog, mjd, ...)	Convert ICRS catalog to topocentric coordinates
center_at_date(ra, dec, mjd)	Get apparent RA/Dec at a given date
conform_coordinates(ra, dec)	Normalize coordinates to standard convention
get_pixlist(ras, decs, level)	Get HEALPix pixels for coordinates
get_pix_range(ra, dec, product)	Get Gaia file pixel ranges for coordinates
retrieve_gaia_data(pixel_range, product)	Download/cache a single Gaia tile
haversine(ra1, dec1, ra2, dec2)	Great-circle distance in degrees
spectro_wavelengths(product)	Get wavelength array for spectro flux points
match_catalogs(cat_a, cat_b)	Inner join two catalogs by source_id
get_cache_dir()	Get cache directory path
clear_cache()	Remove all cached data
get_product(name)	Look up a product by name
list_products()	List all available product names
register_product(product)	Register and persist a custom product
unregister_product(name)	Remove a custom product and its cache

License

MIT