flash_conv_crit_max_chg_L

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

Methods for matrix factorization based on Wang and Stephens (2021)
<https://jmlr.org/papers/v22/20-589.html>.

Jason Willwerscheid

flashier

Empirical Bayes Matrix Factorization

Peter Carbonetto

Wei Wang

Matthew Stephens

Eric Weine

Gao Wang

flash_conv_crit_max_chg_L function

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

Arguments

Calculate the maximum absolute difference in scaled loadings — flash_conv_crit_max_chg_L

<dl>

<dt>curr</dt>
<dd>The <code>flash_fit</code> object from the current iteration.</dd>


<dt>prev</dt>
<dd>The <code>flash_fit</code> object from the previous iteration.</dd>


<dt>k</dt>
<dd>Only used during sequential backfits (that is, calls to
<code>flash_backfit</code> where <code>extrapolate = FALSE</code>). It then
takes the index of the factor/loadings pair currently being optimized.</dd>

</dl>

flash_conv_crit_max_chg_L: Calculate the maximum absolute difference in scaled loadings

Description

Usage

Value

Arguments

See Also