Solves \(\min_\beta \sum_i w_i (y_i - X_i \beta)^2\) on the unweighted design, where the row weight \(w_i = w^{cellrow}_i \cdot w^{\psi}_i \cdot w^{resp}_i\) and \(w^{cellrow}_i\) is the geometric mean of the predictor cell weights of row \(i\). Cell weights thus downweight the influence of rows with contaminated cells without distorting the design values, so the returned \(\beta\) is a valid coefficient for \(X \beta\). Uses QR decomposition for numerical stability.
.weighted_qr_solve(X_int, y, w_cell_int, w_response, w_psi)numeric (p+1)-vector of regression coefficients
\(n \times (p+1)\) design matrix with intercept
numeric n-vector
\(n \times (p+1)\) cell weight matrix
numeric n-vector of response weights
numeric n-vector of psi-weights from residuals
Matthias Templ