`scipy.interpolate._ndbspline:_make_lsq_ndbspl`

source: build-install/usr/lib/python3.14/site-packages/scipy/interpolate/_ndbspline.py:449

Signature

def _make_lsq_ndbspl(x, y, t, k = 3, *, w = None, solver = <function lsqr at 0x0000>, **solver_args)

Summary

Construct a least-squares NdBSpline from scattered data.

Parameters

x : array_like, shape (npts, ndim): Data point coordinates. Each row gives one point in ndim dimensions.
y : array_like, shape (npts, ...): Data values at x. Any trailing dimensions are treated as batch dimensions and fitted with the same design matrix.
t : tuple of array_like, shape (ndim,): Full knot vectors for each dimension. Boundary knots must already be included.
k : int or array_like, shape (ndim,), optional: Spline degree for each dimension. A scalar value is applied to all dimensions. Default is cubic, k=3.
w : array_like, shape (npts,), optional: Non-negative weights for weighted least squares.
solver : callable, optional: Sparse least-squares solver. Default is scipy.sparse.linalg.lsqr. The solver is called as solver(A, b, **solver_args), where A is the sparse design matrix with shape (npts, ncoeffs) and b is a 1D right-hand side with shape (npts,). It must return the same result format as scipy.sparse.linalg.lsqr and scipy.sparse.linalg.lsmr: the fitted coefficients are read from result[0] and the convergence status from result[1]. Status values 0, 1, and 2 are treated as successful termination, following the conventions of lsqr and lsmr.
**solver_args: Additional keyword arguments passed to solver.

Returns

spl : NdBSpline: Tensor-product spline with coefficients fitted by least squares.

Notes

This solves the sparse least-squares problem

min_c || W @ (A @ c - y) ||_2,

where A is the sparse tensor-product B-spline design matrix built by NdBSpline.design_matrix and W is a diagonal matrix of weights.

The default solver is scipy.sparse.linalg.lsqr. The related scipy.sparse.linalg.lsmr solver has the same basic interface and can be useful as a comparison when convergence depends on the conditioning of the least-squares problem or on the selected tolerances. Solver tolerances are not modified by this helper; pass them explicitly through solver_args if the solver defaults are not appropriate.

Other least-squares estimators can be used through a small wrapper, as long as the wrapper accepts a sparse matrix A and a one-dimensional right-hand side b and returns coefficients in result[0] and a status in result[1]. Status values other than 0, 1, or 2 are treated as solver failures.

Aliases

scipy.interpolate._ndbspline._make_lsq_ndbspl