TML (version 2.3.0)

trop.logistic.regression: Tropical Logistic Regression

Description

Performs tropical logistic regression, by finding the optimal statistical parameters for the training dataset (D,Y), where D is the matrix of covariates and Y is the binary response vector

Usage

trop.logistic.regression(D, Y, penalty = 0, model_type = "two_species")

Value

vector; optimal model parameters (two normal vectors and two scaling factors)

Arguments

D

matrix of dimension N*e, where N is the number of observations which lie in R^e

Y

binary vector with dimension N, with each component corresponding to an observation

penalty

scalar; positive real number

model_type

string; options are "two-species" (default), "one-species", "general"

Author

Georgios Aliatimis

References

Aliatimis, Georgios, Ruriko Yoshida, Burak Boyaci and James A. Grant (2023). Tropical Logistic Regression on Space of Phylogenetic Trees.

Examples

# \donttest{
library(ROCR)
T0 = Sim_Trees15
T1 = Sim_Trees25
D  = rbind(T0,T1)
Y = c(rep(0,dim(T0)[1]), rep(1,dim(T1)[1]))
N = length(Y)
set.seed(1)
train_set = sample(N,floor(0.8 * N)) ## 80/20 train-test split
pars <- trop.logistic.regression(D[train_set,],Y[train_set], penalty=1e4)
test_set = (1:N)[-train_set]
Y.hat <- rep(0, length(test_set))
for(i in 1:length(test_set))   Y.hat[i] <- prob.class(pars, D[test_set[i],])
Logit.ROC <- performance(prediction(Y.hat, Y[test_set]), measure="tpr", x.measure="fpr")
plot(Logit.ROC, lwd = 2, main = "ROC Curve for Logistic Regression Model")
print(paste("Logit.AUC=", performance(prediction(Y.hat, Y[test_set]), measure="auc")@y.values))
# }