update_weights: Traversing the network for fit weight sets.

In robertness/signalgraph: Signal Processing with Graphs

Description

To reduce dimensionality of the problem of fitting all the weights in the graph, for each vertex the weights on the incoming edges of that vertex are fit seperately. Priority goes to 'fitted' vertices (vertices that are used to calculate the mean squared error loss function), then vertices that are upstream by 1 degree, then 2 degrees, etc. 'update_weights' traverses the vertices and fits them individually. It uses 'vertex_updater" from the lucy package to apply the 'fit_weights_for_node' callback on a vertex. 'vertex_updater' will first check all the downstream vertices for those that are not yet updated (via the 'get_downstream_vertices' callback) and recursively fit those if they are not yet updated. However, 'order_vertices' is first applied to minimize the need for this recursion.

Usage

update_weights(g, verbose = FALSE)

order_vertices(g)

get_downstream_vertices(g, v)

Arguments

g	a signalgraph object.
verbose	if true prints messages about fitting performance.

Value

a signalgraph object with updated weights.

robertness/signalgraph documentation built on May 27, 2019, 10:33 a.m.