optimizep: Optimizing algorithm parameters

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

Description

Function to learn the parameters "alpha", determining neurons activation values, and the neuron threshold "c".

Usage

optimizep(pos_vect, neg_vect, training_labels)

Arguments

pos_vect	vector of abscissae of the points to be separated
neg_vect	vector of ordinates of the points to be separated
training_labels	1/-1 vector of point labels

Details

This function computes the optimal angle "alpha" and the optimal threshold "c". For each labeled neuron k, a point (pos_vect[k], neg_vect[k]) is considered. Points are labeled according to the labels contained in the vector "training_labels". Then the straight line, among those with positive slope, which separates these points by maximizing the F-score is learned. The line is represented by the angle alpha formed with the "x" axis, and its intercept "q" with "y" axis. When separating points by a straight line, there are two possibility: 1) Considering as positive the half-plane above the line; 2) Considering as positive the half-plane below the line. The procedure investigates both these possibilities, and also returns which choice between 1) and 2) corresponds to the best F-score.

Value

list "res" with 4 components:

res$alpha	value of the optimum angle alpha
res$c	value of the optimum threshold c
res$Fscore	value of the optimum F-score
res$pos_half	the position of the positive half-plane : > 0 in case 1), < 0 in case 2)

See Also

optimize_pos_above

Examples

library(bionetdata);
data(Yeast.STRING.data);
data(Yeast.STRING.FunCat);
n <- nrow(Yeast.STRING.data);
## removing dummy node 00
Yeast.STRING.FunCat <- Yeast.STRING.FunCat[, 
                        -which(colnames(Yeast.STRING.FunCat)=="00")];
## selecting the class with index 1
class <- 1;
labels <- as.vector(Yeast.STRING.data[, class]);
names(labels) <- rownames(Yeast.STRING.FunCat);
labels <- as.vector(Yeast.STRING.FunCat[, class]);
names(labels) <- rownames(Yeast.STRING.FunCat);
## partitioning the data
folds <- find.division.strat(labels, 1:n, 3);
labels[labels <= 0] <- -1;
test.set <- folds[[1]];
training.set <- setdiff(1:n, test.set);
labels[test.set] <- 0;
## generating the points to be separated
points <- generate_points(Yeast.STRING.data, test.set, labels);
opt_parameters <- optimizep(points$pos_vect[training.set],
                    points$neg_vect[training.set], labels[training.set]);
str(opt_parameters);

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