reg_data {COSNet} | R Documentation |
This function modifies the weights and the thresholds of the network to realized the COSNet regularization.
reg_data(W, theta, eta, M, m, pos_num)
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 |
list of two element:
W |
the regularized connection matrix |
theta |
regularized threshold vector |
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.
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);