Skip to content

light-curve

High-performance time-series feature extraction for astrophysics.

light-curve is a Python package for analyzing photometric light curves at the scale of millions of objects. It provides multiple tools for ML pre-processing pipelines as well as 40+ statistical and variability features for filtering, classification, and catalog analysis.

Install
pip install 'light-curve[full]'
A P

Feature extractors

40+ features: magnitude and flux statistics, time-series shape descriptors, period extraction, and parametric fits for transients. Supports multi-band light curves and optimized to process 10⁶–10⁹ objects.

 0.73,  0.31, –1.22,  0.84, –0.28,  ...,  2.03

ML embeddings

Map raw light curves to dense vectors using pretrained transformer models. Suitable for classification, anomaly detection, and similarity search at different scales

Δm lg Δt

dm-dt maps

2D histograms of Δmag vs log-Δt for all observation pairs, providing a fixed-size image representation for CNN-based variability classifiers.

Quick start

import light_curve as lc
from light_curve.embed import Astromer2
import numpy as np

rng = np.random.default_rng(0)
t = np.sort(rng.uniform(0, 100, 100))
m = 15.0 + 0.01 * t + rng.normal(0, 0.1, 100)
err = np.full(100, 0.1)

# Feature extraction
extractor = lc.Extractor(lc.Amplitude(), lc.BeyondNStd(nstd=1), lc.LinearFit())
result = extractor(t, m, err)

# ML embedding with pretrained Astromer2 (downloads on first use)
model = Astromer2.from_hf(output="mean")
embedding = model(t, m).squeeze()   # shape (256,)

# dm-dt map for CNN-based variability classifiers
dmdt = lc.DmDt.from_borders(min_lgdt=0, max_lgdt=2, max_abs_dm=1.0, lgdt_size=16, dm_size=16, norm=[])
matrix = dmdt.points(t, m)   # shape (16, 16)

Used by

Alert brokers

AMPEL, ANTARES, and Fink use light-curve for real-time feature extraction when classifying on the order of a million alerts per night from the Zwicky Transient Facility and the Rubin Observatory.

The SNAD team

The SNAD anomaly-detection group uses light-curve to analyze hundreds of millions of Zwicky Transient Facility light curves from data releases, powering large-scale analyses and the public SNAD ZTF Viewer.

Software

Feature extraction library for time series with Dask-parallel processing and a scikit-learn-style API, using light-curve as its computational backend
Photometric supernova classification pipeline for LSST, with active learning, developed by LSST DESC and COIN
Real-time supernova light curve fitting and classification for ZTF/Rubin

Publications

Used in 30+ research publications.