sparse_ho.algo.Implicit

class sparse_ho.algo.Implicit(max_iter=100, max_iter_lin_sys=100, tol_lin_sys=1e-06)

Algorithm to compute the hypergradient using implicit differentiation.

First the algorithm computes the regression coefficients beta, then the gradient is computed after resolution of a linear system on the generalized support of beta.

Parameters
max_iter: int (default=100)

Maximum number of iteration for the inner solver.

max_iter_lin_sys: int (default=100)

Maximum number of iteration for the resolution of the linear system.

tol_lin_sys: float (default=1e-6)

Tolerance for the resolution of the linear system.

__init__(max_iter=100, max_iter_lin_sys=100, tol_lin_sys=1e-06)

Initialize self. See help(type(self)) for accurate signature.

Methods

__init__([max_iter, max_iter_lin_sys, …])

Initialize self.

compute_beta_grad(X, y, log_alpha, model, …)

Compute beta and the hypergradient, with implicit differentiation.