What you get back

GSPICE returns read-only container objects whose array fields have explicit shapes and physical units. A value without a unit is a count, a mask, or a dimensionless ratio. In the shapes below, nspec is the number of spectra and npix is the number of pixels.

Model

gspice.fit() returns a gspice.Model and a mask:

Python

model, clipped = gspice.fit(flux, ivar)

The mask clipped has the shape (nspec, npix). It is True at each pixel that fit left out of its final estimate. It is also True at every pixel of a spectrum that fit did not use.

A model has one mean and one covariance for each S/N bin. There is one bin unless you call fit(..., snr_bins=n). The mean and the covariance are in scaled units, not flux units. Continuum normalization explains the scaling.

Field	Type and shape	Holds
mean	array (nbin, npix)	The mean of each bin, in scaled units
cov	tuple of nbin arrays (npix, npix), or of LowRank	The covariance of each bin, in scaled units squared
n_train	tuple of nbin integers	The number of training spectra of each bin
edges	tuple of nbin − 1 floats	The signal-to-noise ratios (S/N) between the bins
center	float or None	The constant center in flux units, or None for a mean-centered model
noise	bool	True for a noise-aware model
weighted	bool	True if fit had unequal weights
blocks	Blocks or None	The block thresholds of fit_blocks, or None without them

A noise-aware model, from fit(..., noise=True), has one bin. Its mean is in flux units, and its cov is the covariance of the flux without noise, in flux².

Blocks

gspice.fit_blocks() returns a copy of the model whose blocks is a gspice.Blocks. gspice.repair() reads it.

Field	Type and shape	Holds
starts	array of int64 (nblock,)	The first pixel of each block
width	int	The number of pixels of each block
nguard	int	The half-width of the guard window around each block, in pixels
thresholds	array (nbin, nblock)	The threshold of each bin and block

Prediction

gspice.pixelwise() predicts each pixel from the pixels outside a guard window around it. It returns a gspice.Prediction.

Field	Shape	Holds
flux	(nspec, npix)	The prediction of each pixel, in flux units
var	(nspec, npix)	Its variance, which includes the noise, in flux²
z	(nspec, npix)	The input flux minus the prediction, divided by √var; NaN at bad pixels

Repaired

gspice.infill() and gspice.repair() return a gspice.Repaired. The replaced pixels hold their prediction. The ignored pixels are neither replaced nor used to predict the others.

Field	Shape	Holds
flux	(nspec, npix)	The prediction at the replaced pixels, and the input flux elsewhere, in flux units
var	(nspec, npix)	In flux²: the variance of the prediction at the replaced pixels, which includes the noise; 1/ivar at the other good pixels; inf at the other bad pixels
replaced	(nspec, npix), bool	The replaced pixels
ignored	(nspec, npix), bool	The ignored pixels
chi2	(nspec,)	The χ² of the input values at the replaced pixels that are good in the input; 0 if there are none
dof	(nspec,), int64	The number of those pixels
fs	(nspec,)	The finite-sample factor of chi2
draws	(nspec, npix), or None	One random draw at the replaced pixels, and the input flux elsewhere, in flux units
blocks	(nspec, nblock), bool, or None	The blocks that repair flagged in each spectrum

Three fields depend on the call:

• draws is None unless you pass rng, a numpy.random.Generator.
• blocks is None for a result of infill, and of repair(..., blockwise=False).
• Repaired.covariance(s) returns the replaced pixels of spectrum s and their covariance, in flux². It needs a call with covariance=True.