Feature extractors¶

light_curve provides 40+ hand-crafted feature extractors for astrophysical light curves. All share a common interface: callable objects with .names and .descriptions attributes.

import light_curve as licu
import numpy as np

t = np.array([0.0, 1.2, 3.5, 7.1, 9.0, 11.0, 13.0, 15.0, 17.0, 19.0])
m = np.array([15.1, 15.3, 14.9, 15.0, 15.2, 15.1, 14.8, 15.3, 15.0, 15.1])
err = np.array([0.05] * 10)
band = np.tile(["g", "r"], 5)

ext = licu.Extractor(licu.Chi2Pvar(), licu.BeyondNStd(nstd=1), licu.LinearFit(bands=["g", "r"]))
result = ext(t, m, err, band)
print(dict(zip(ext.names, result)))

Use Extractor to combine multiple features into a single callable, or call each class directly for a single feature. The .many() method processes a list of light curves in batch with reduced Python–Rust overhead.

See the API reference for full signatures, parameters, and equations.

Feature table¶

Variability¶

Fast: all 26 features combined ~100 µs on 1,000 observations, single band.

Feature	Description	Outputs
`Amplitude`	Half peak-to-peak amplitude	1
`AndersonDarlingNormal`	Unbiased Anderson–Darling normality test statistic	1
`BeyondNStd`	Fraction of observations beyond \(n\,\sigma_m\) from \(\langle m \rangle\)	1
`Chi2Pvar`	Probability of variability from the \(\chi^2\) test	1
`Cusum`	Range of cumulative sums	1
`Eta`	Von Neumann \(\eta\)	1
`EtaE`	\(\eta^e\), adapted for unevenly sampled time series	1
`ExcessVariance`	Measure of intrinsic variability amplitude	1
`InterPercentileRange`	\(Q(1-p) - Q(p)\) inter-percentile range	1
`Kurtosis`	Excess kurtosis of magnitude	1
`LaflerKinmanStringLength`	Lafler–Kinman string-length smoothness statistic	1
`MaximumSlope`	Maximum slope between consecutive observations	1
`Mean`	Mean magnitude	1
`MeanVariance`	Standard deviation to mean ratio	1
`Median`	Median magnitude	1
`MedianAbsoluteDeviation`	Median of \(\lvert m_i - \mathrm{median}(m) \rvert\)	1
`MedianBufferRangePercentage`	Fraction of points within \(q \times \mathrm{amplitude}\) of median	1
`OtsuSplit`	Otsu thresholding: bimodality measure (subset means, std devs, fraction)	4
`PercentAmplitude`	Maximum deviation of magnitude from its median	1
`PercentDifferenceMagnitudePercentile`	Ratio of inter-percentile range to median	1
`ReducedChi2`	Reduced \(\chi^2\) of magnitude measurements	1
`Roms`	Robust median statistic (experimental)	1
`Skew`	Skewness \(G_1\) of magnitude	1
`StandardDeviation`	Standard deviation \(\sigma_m\) of magnitude	1
`StetsonK`	Stetson \(K\) light-curve shape coefficient	1
`WeightedMean`	Inverse-variance weighted mean magnitude	1

Linear trend¶

Fast: both features combined ~10 µs on 1,000 observations, single band.

Feature	Description	Outputs
`LinearFit`	Slope, its error, and reduced \(\chi^2\) of the weighted linear fit	3
`LinearTrend`	Slope, its error, and noise level of the unweighted linear fit	3

Time sampling¶

Fast: all 6 features combined ~5 µs on 1,000 observations, single band.

Feature	Description	Outputs
`Duration`	Total time span of the light curve	1
`MaximumTimeInterval`	Maximum gap between consecutive observations	1
`MinimumTimeInterval`	Minimum gap between consecutive observations	1
`ObservationCount`	Number of observations	1
`TimeMean`	Mean observation time	1
`TimeStandardDeviation`	Standard deviation of observation times	1

Periodogram¶

Feature	Description	Outputs
`Periodogram`	Lomb–Scargle periodogram: period and power of the strongest peaks; supports `phase_features` to extract features from the phase-folded light curve at the best period	≥2

Non-linear parametric fits (transients, flux only)¶

Feature	Description	Outputs
`BazinFit`	Bazin function — 5-parameter rising/falling exponential fit for core-collapse SNe	6
`LinexpFit`	Linexp function — linear-times-exponential fit for core-collapse SNe	5
`VillarFit`	Villar function — 7-parameter Gaussian+plateau fit for SN classification	8

Multiband parametric fit¶

Feature	Description	Outputs
`RainbowFit`	Multiband blackbody fit (Russeil+23) — bolometric flux + temperature evolution + SED spectral model	4 to 10+

Multiband¶

Fast: all 4 features combined ~20 µs on 1,000 observations.

Feature	Description	Outputs
`ColorOfMaximum`	Difference between maximum magnitudes of two bands	1
`ColorOfMedian`	Difference between median magnitudes of two bands	1
`ColorOfMinimum`	Difference between minimum magnitudes of two bands	1
`ColorSpread`	Population std dev of per-band weighted mean magnitudes	1

Detection-based (experimental)¶

Feature	Description	Outputs
`FluxNNotDetBeforeFd`	Number of non-detections (flux) before the first detection	1
`MagnitudeNNotDetBeforeFd`	Number of non-detections (magnitude) before the first detection	1

Meta¶

Feature	Description
`Extractor`	Combine multiple feature extractors into a single callable
`Bins`	Bin a time series, then apply any set of features to each bin