Iterates \(g_j \leftarrow S_j(Y - \mu - \sum_{k \ne j} g_k)\).
Additivity buys back the one-dimensional \(n^{-2/5}\) rate in any
dimension. Components are centred every sweep because only their
SUM is identified. Mirrors morie.fn.hrzbkft.
Usage
morie_backfitting(X, y, h = NULL, max_iter = 50L, tol = 1e-06)
Arguments
- X
numeric covariate matrix.
- y
numeric response.
- h
bandwidth.
- max_iter
sweeps.
- tol
convergence tolerance.
Value
list: mu, components, fitted, n_iter, converged.
References
Horowitz, Ch. 2.
Examples
set.seed(1)
X <- matrix(runif(200), ncol = 2)
morie_backfitting(X, X[, 1]^2 + sin(3 * X[, 2]))$converged
#> [1] TRUE
