reg_data: Function to compute the regularized version of COSNet (Frasca...

In COSNet: Cost Sensitive Network for node label prediction on graphs with highly unbalanced labelings

Description

This function modifies the weights and the thresholds of the network to realized the COSNet regularization.

Usage

reg_data(W, theta, eta, M, m, pos_num)

Arguments

W	square symmetric named matrix of the network weights. The components of W are in the [0,1] interval. The i,j-th component is the weight between neuron i and neuron j. The components of the diagonal of W are 0
theta	vector of the neuron activation thresholds
eta	real value corresponding to the eta regularization coefficient in the energy function (Frasca et al. 2013). If eta = 0 no regularization is applied. The higher the value of eta, the more the influence of the regularization term
M	positive neuron activation value
m	negative neuron activation value
pos_num	number of expected positive neurons in the equilibrium state of the network

Value

list of two element:

W	the regularized connection matrix
theta	regularized threshold vector

References

Frasca M., Bertoni A., Re M., Valentini G.: A neural network algorithm for semi-supervised node label learning from unbalanced data. Neural Networks, Volume 43, July, 2013 Pages 84-98.

Examples

library(bionetdata);
data(Yeast.STRING.data);
n <- nrow(Yeast.STRING.data);
dim(Yeast.STRING.data);
range(Yeast.STRING.data);
## setting values for parameter alpha, for the rate of positive examples,
## for neuron thresholds and for eta parameter
alpha <- 1;
pos.rate <- 0.01;
thresholds <- runif(n);
range(thresholds);
eta <- 0.001;
a <- reg_data(Yeast.STRING.data, thresholds, eta, sin(alpha),
    -cos(alpha), ceiling(pos.rate*n));
## new connection matrix
dim(a$W);
range(a$W);
## new thresholds
range(a$theta);

COSNet documentation built on Nov. 8, 2020, 8:12 p.m.