Generate.Data <- function(sample_size,num_features,theta_pred,model){
####
#### To generate the treatment
####
treatment <- rbinom(sample_size,1,0.5)
####
#### To generate the labels I need to create for each model the logistic regression function
####
switch(model,
{ #### Model 1
sigma <- diag(num_features)
correl <-0
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part);
pred_part <- apply(covariates[,4:8],1,sum);# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + (treatment+0.1) * theta_pred * pred_part ;
prog_features <- 1:5; pred_features <- 4:8;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 2
sigma <- diag(num_features)
correl <-0
sigma[seq(1,15,by=2),seq(1,15,by=2)] <-correl # correlation between odds features
sigma[seq(2,15,by=2),seq(2,15,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part);
pred_part <- apply(covariates[,6:10],1,sum);# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + (treatment+0)*theta_pred * pred_part ;
prog_features <- 1:5; pred_features <- 6:10;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 3
sigma <- diag(num_features)
correl <-0.70
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part);
pred_part <- apply(covariates[,6:10],1,sum);# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + (treatment+0)*theta_pred * pred_part ;
prog_features <- 1:5; pred_features <- 6:10;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 4
sigma <- diag(num_features)
correl <-0.70
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- covariates[,1] + apply(covariates[,c(2,3)],1,prod) + apply(covariates[,c(4,5)],1,prod) ;# prog_part <- prog_part - mean(prog_part);
pred_part <- ( covariates[,6] + apply(covariates[,c(7,8)],1,prod) + apply(covariates[,c(9,10)],1,prod) );# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + treatment * theta_pred * pred_part;
prog_features <- 1:5; pred_features <- 6:10;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 5
sigma <- diag(num_features)
correl <-0.70
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- covariates[,1]*(covariates[,2]*covariates[,3]-covariates[,4]^2) ;# prog_part <- prog_part - mean(prog_part);
pred_part <- covariates[,5]*exp(covariates[,6]*covariates[,7]-covariates[,8]^2) ;# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + treatment*theta_pred * pred_part;
prog_features <- 1:4; pred_features <- 5:8;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 6
sigma <- diag(num_features)
correl <-0.70
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- covariates[,1] + (covariates[,2]>-0.545) * (covariates[,3]<0.545) ;# prog_part <- prog_part - mean(prog_part);
pred_part <- covariates[,4] + (covariates[,5]>-0.545) * (covariates[,6]<0.545) ;# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + treatment*theta_pred * pred_part;
prog_features <- 1:3; pred_features <- 4:6;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
{ #### Model 7
sigma <- diag(num_features)
correl <-0.70
sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features
sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features
diag(sigma) <- 1
covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS)
prog_part <- covariates[,1] + (covariates[,2]>-0.545) * (covariates[,3]<0.545)*(covariates[,4]>0);# prog_part <- prog_part - mean(prog_part);
pred_part <- covariates[,5] + (covariates[,6]>-0.545) * (covariates[,7]<0.545)*(covariates[,8]>0);# pred_part <- pred_part - mean(pred_part);
logit_pY1 <- prog_part + treatment*theta_pred * pred_part;
prog_features <- 1:4; pred_features <- 5:8;
irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features))
for (index_feature in 1:num_features){
covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1)))
}
covariates <- as.data.frame.matrix(covariates)
},
stop("Wrong model")
)
pY1 <- 1/(1+exp(-logit_pY1))
labels <- rbinom(sample_size,1,pY1);
synthetic_datasets <- numeric(0)
synthetic_datasets$data <- covariates
synthetic_datasets$treatment <- treatment
synthetic_datasets$labels <- labels
synthetic_datasets$prog_features <- prog_features
synthetic_datasets$pred_features <- pred_features
synthetic_datasets$irr_features <- irr_features
return(synthetic_datasets)
}
Datasets
Models
