Compute gradient of Spreadgram — gradient • labyrinth

This function calculates the gradient of the computed activation rates for a given background graph. The gradient is computed based on the Sigmoid function and the log likelihood, which iteratively computes the error between each permutation.

This function requires the background graph and the computed activation rates as input.

Usage

gradient(graph, activation, threads = 0, verbose = TRUE)

Arguments

graph: A square matrix (or dgCMatrix representing the background graph. Inside this adjacency matrix, each row and column of the matrix represents a node in the graph. The values of the matrix should be either 0 or 1 (or either 0 or larger than 0), where a value of 0 indicates no relations between two nodes. The diagonal of the matrix should be 0, as there are no self-edges in the graph.
activation: A numeric vector representing the computed activation rates for each node in the graph. The length of the vector should be equal to the number of nodes in the graph. This vector should contain the activation rate for each node.
threads: A scalar numeric indicating the parallel threads. Default is 0 (auto-detected).
verbose: Show verbose message

Value

A scalar representing the gradient of the computed activation rates. The gradient represents the rate of change of the activation rate.

Examples

# The graph G
data("graph", package = "labyrinth")

# The initial info
last_activation = c(2, 4, 3, 2, 2, 1, 5)

gradient(graph, last_activation)
#> [1] 0.03732267