flash_conv_crit_max_chg: Calculate the maximum absolute difference in scaled loadings...

In flashier: Empirical Bayes Matrix Factorization

Calculate the maximum absolute difference in scaled loadings and factors

Description

An alternative objective function that can be used to determine convergence when fitting a flash object. Calculates the maximum (absolute) change over all (posterior expected values for) loadings \ell_{ik} and factors f_{jk}. At each iteration, the loadings vectors \ell_{\cdot 1}, \ldots, \ell_{\cdot K} and factors f_{\cdot 1}, \ldots, f_{\cdot K} are L^2-normalized.

Usage

flash_conv_crit_max_chg(curr, prev, k)

Arguments

curr	The flash_fit object from the current iteration.
prev	The flash_fit object from the previous iteration.
k	Only used during sequential backfits (that is, calls to flash_backfit where extrapolate = FALSE). It then takes the index of the factor/loadings pair currently being optimized.

Value

A scalar, which is compared against the tolerance parameter tol to determine whether a fit has converged.

See Also

flash_conv_crit_elbo_diff, flash_conv_crit_max_chg_L flash_conv_crit_max_chg_F

flashier documentation built on Oct. 17, 2023, 5:07 p.m.