Tuning Parameter Selection: Network Sparsity

Our XMRF package implements two data-driven methods to determine the sparsity of a fitted network. The first method is the stability selection over many bootstrap samples for a single regularization value (Meinshausen and Bühlmann, 2010). The second method is the StARS selection, which is computed over a range of regularization values to select a network with the smallest regularization value that is simultaneously sparse and reproducible in random samples (Liu et al., 2010). In this section, both of these methods are demonstrated for an example using the local Poisson graphical model (LPGM).



Subsections

2015-05-29