Function concurrent_regression 

pub fn concurrent_regression(
    response: &FdMatrix,
    predictors: &[FdMatrix],
    argvals: Option<&[f64]>,
    bandwidth: f64,
    kernel: &str,
) -> Result<ConcurrentRegrResult, FdarError>

Concurrent (varying-coefficient) functional regression.

Fits the model y_i(t) = β₀(t) + Σₖ βₖ(t)·xₖᵢ(t) + εᵢ(t) for a dense shared grid of m evaluation points and p ≥ 1 functional predictors. Estimation follows the fdaconcur two-step convention: pointwise OLS at each grid column, then local-linear kernel smoothing of the resulting discrete coefficient sequences.

Arguments

• response — Functional response matrix (n × m).
• predictors — Slice of functional predictor matrices, each (n × m), all evaluated on the same shared grid.
• argvals — Shared evaluation grid (length m); if None, a uniform 0..1 grid is used (project-wide convention).
• bandwidth — Kernel bandwidth for β(t) smoothing (must be positive). Larger values yield smoother coefficient curves.
• kernel — Kernel type: "gaussian" (default), "epanechnikov", "tricube". Passed directly to crate::smoothing::local_linear.

Errors

Returns FdarError::InvalidDimension if:

• predictors is empty,
• response has fewer than 2 rows or zero columns,
• response has at most as many rows as there are predictors (n <= p), which would make each per-column design matrix underdetermined,
• any predictor’s shape does not match (n, m), or
• argvals is Some with a length different from m.

Returns FdarError::InvalidParameter if bandwidth is not strictly positive (i.e., zero, negative, f64::NAN, or f64::INFINITY).

Notes

A small ridge regulariser (eps = 1e-10 * (n + 1)) is applied to the diagonal of each column’s normal equations to prevent numerical blow-up when predictors are collinear at a grid point.