Function fplsr 

pub fn fplsr(
    data: &FdMatrix,
    ncomp: usize,
    argvals: &[f64],
) -> Result<FplsrResult, FdarError>

Functional PLS forecasting variant (a PLS-score alternative to FPC-score AR).

Fits a lag-1 PLS design — predictor = current curve (rows 0..n-2), response = next curve (rows 1..n-1) — and forecasts the next curve one step ahead. Because the shipped PLS machinery (crate::scalar_on_function::fregre_pls) takes a scalar response, the functional (curve) response is handled by fitting one scalar PLS regression per evaluation point (Option A of the research note): for each grid point j, the next-curve column j is regressed on the current curve via PLS, the in-sample fits populate column j of fitted, and the last observed curve is projected + predicted to give the forecast at j.

Divergence from ftsa

ftsa::fplsr uses a unified NIPALS/SIMPLS functional operator; this implementation uses per-evaluation-point scalar PLS regression (functionally equivalent for point prediction, less elegant). Deterministic — no RNG.

Errors

Returns FdarError::InvalidDimension for empty data / mismatched argvals, and FdarError::InvalidParameter for ncomp == 0 or n < 3 (need at least two lag-1 rows plus a forecast origin). PLS/rank failures propagate as FdarError::ComputationFailed.