compute_gauss_lasso: Do a lm on top of a lasso regression.
Description
compute_gauss_lasso takes the variables selected by a lasso procedure, and
uses them to do a simple linear least square regression. Function used is
lm for non-transformed data (root = NULL), and lm.fit for
transformed data (root = an integer).
Usage
compute_gauss_lasso(Ypt, Xp, delta, root, projection = which(rowSums(delta) != 0))
Arguments
Xp
(transformed) matrix of regression
delta
regression coefficients obtained with a lasso regression
root
the position of the root (intercept) in delta
Value
Named list, with "E0.gauss" the intercept (value at the root);
"shifts.gauss" the list of shifts found on the branches; and "residuals" the
residuals of the regression
Details
Depending on the value of root, the behaviour is different. If root is null, then
we fit a linear regression with an intercept. If root is equal to an integer,
then the "intercept" column of the matrix Xp (that has possibly been trough a
multiplication with a cholesky decomposition of the variance) is included, rather
than the intercept.