algorithms.debiased_lasso

Module: algorithms.debiased_lasso

Functions

selectinf.algorithms.debiased_lasso.debiased_lasso_inference(lasso_obj, variables, delta)

Debiased estimate is .. math:

\hat{eta}^d = \hat{eta} - \hat{       heta} 

abla ell(hat{eta})

where \(\ell\) is the Gaussian loss and \(\hat{ heta}\) is an approximation of the inverse Hessian at \(\hat{eta}\). The term on the right is expressible in terms of the inactive gradient as well as the fixed active subgradient. The left hand term is expressible in terms of \(ar{eta}\) the “relaxed” solution and the fixed active subgradient. We need a covariance for \((ar{eta}_M, G_{-M})\). Parameters ———- lasso_obj : selection.algorithms.lasso.lasso

A lasso object after calling fit() method.

variablesseq

Which variables should we produce p-values / intervals for?

deltafloat

Feasibility parameter for estimating row of inverse of Sigma.

selectinf.algorithms.debiased_lasso.debiasing_matrix(X, rows, bound=None, linesearch=True, scaling_factor=1.5, max_active=None, max_try=10, warn_kkt=False, max_iter=50, kkt_stop=True, parameter_stop=True, objective_stop=True, kkt_tol=0.0001, parameter_tol=0.0001, objective_tol=0.0001)

Find a row of debiasing matrix using line search of Javanmard and Montanari.

selectinf.algorithms.debiased_lasso.pseudoinverse_debiasing_matrix(X, rows, tol=1e-09)

Find a row of debiasing matrix using algorithm of Boot and Niedderling from https://arxiv.org/pdf/1703.03282.pdf