calc_lambda_max_cml: Calculate Lambda max for a sequence

In submission-user/cmlDIFlasso: DIF-detection with Lasso-methods in a cML-Framework

Description

Calculates the maximum Lambda-value for the sequence (the minimum Lambda-value, for which all parameters are equal to zero). For L1, this is the maximum-norm of the score function for the full modell evaluated at the cml-estimates of the restricted model (hence all delta-parameters equal to zero). For Group-lasso this is the maximum-norm of a vector containing the Euclidian norms per parameter group of the score vector evaluated at the cml-estimates of the restricted model

Usage

calc_lambda_max_cml(beta_intercepts, covariates, Y)

Arguments

beta_intercepts	beta_intercepts from restricted model, eg Rasch model without DIF
covariates	Matrix of covariate values of all persons (n x num. of covs)
Y	matrix of responses

Value

A list with the maximal Lambda-value for the sequence of Lambdas to be used for optimization with three elements: selection for items, covariates and all interactions. Note that for grouplasso this value is actually lambda_max * sqrt(groupsize), the groupsize being either number of items or covariates, depending on the selection of penalization model. The product is returned for these because group sizes are equal for all groups, so this simplification is justified.

submission-user/cmlDIFlasso documentation built on Dec. 23, 2021, 6:40 a.m.