This function builds an extrapolated point from previous iterates during the
coordinate descent iterations. Using the K previous iterates, it builds an
extrapolated point by solving a quadratic program.
This is the backbone function to solve multi-task optimization problems. For a
detailed description, see [solvers::anderson_cd::coordinate_descent] function.
This function constructs the gradient of a datafit restricted to the features in
the working set. It is used in opt_cond_violation to rank features included in
the working set.
This function is used to construct a working set by sorting the indices
in descending order of the features having the largest violation to the optimality
conditions.
This function computes the distance of the negative gradient of the datafit to the
subdifferential of the penalty restricted to the working set. It returns
an array containing the distances for each feature in the working set as well
as the maximum distance.