Datasets
Models
Downloads: 1
Diff of
/Functions-GenerateData.R
[000000]
..
[c4e594]
Switch to unified view
| a | b/Functions-GenerateData.R | ||
|---|---|---|---|
| 1 | Generate.Data <- function(sample_size,num_features,theta_pred,model){ |
||
| 2 | |||
| 3 | |||
| 4 | #### |
||
| 5 | #### To generate the treatment |
||
| 6 | #### |
||
| 7 | treatment <- rbinom(sample_size,1,0.5) |
||
| 8 | |||
| 9 | |||
| 10 | #### |
||
| 11 | #### To generate the labels I need to create for each model the logistic regression function |
||
| 12 | #### |
||
| 13 | switch(model, |
||
| 14 | { #### Model 1 |
||
| 15 | |||
| 16 | sigma <- diag(num_features) |
||
| 17 | correl <-0 |
||
| 18 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 19 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 20 | diag(sigma) <- 1 |
||
| 21 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 22 | |||
| 23 | |||
| 24 | prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part); |
||
| 25 | pred_part <- apply(covariates[,4:8],1,sum);# pred_part <- pred_part - mean(pred_part); |
||
| 26 | logit_pY1 <- prog_part + (treatment+0.1) * theta_pred * pred_part ; |
||
| 27 | prog_features <- 1:5; pred_features <- 4:8; |
||
| 28 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 29 | |||
| 30 | for (index_feature in 1:num_features){ |
||
| 31 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 32 | } |
||
| 33 | |||
| 34 | covariates <- as.data.frame.matrix(covariates) |
||
| 35 | }, |
||
| 36 | { #### Model 2 |
||
| 37 | |||
| 38 | sigma <- diag(num_features) |
||
| 39 | correl <-0 |
||
| 40 | sigma[seq(1,15,by=2),seq(1,15,by=2)] <-correl # correlation between odds features |
||
| 41 | sigma[seq(2,15,by=2),seq(2,15,by=2)] <-correl # correlation between even features |
||
| 42 | diag(sigma) <- 1 |
||
| 43 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 44 | |||
| 45 | prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part); |
||
| 46 | pred_part <- apply(covariates[,6:10],1,sum);# pred_part <- pred_part - mean(pred_part); |
||
| 47 | logit_pY1 <- prog_part + (treatment+0)*theta_pred * pred_part ; |
||
| 48 | prog_features <- 1:5; pred_features <- 6:10; |
||
| 49 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 50 | |||
| 51 | for (index_feature in 1:num_features){ |
||
| 52 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 53 | } |
||
| 54 | |||
| 55 | covariates <- as.data.frame.matrix(covariates) |
||
| 56 | }, |
||
| 57 | { #### Model 3 |
||
| 58 | |||
| 59 | sigma <- diag(num_features) |
||
| 60 | correl <-0.70 |
||
| 61 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 62 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 63 | diag(sigma) <- 1 |
||
| 64 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 65 | |||
| 66 | |||
| 67 | |||
| 68 | prog_part <- apply(covariates[,1:5],1,sum); # prog_part <- prog_part - mean(prog_part); |
||
| 69 | pred_part <- apply(covariates[,6:10],1,sum);# pred_part <- pred_part - mean(pred_part); |
||
| 70 | logit_pY1 <- prog_part + (treatment+0)*theta_pred * pred_part ; |
||
| 71 | prog_features <- 1:5; pred_features <- 6:10; |
||
| 72 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 73 | |||
| 74 | for (index_feature in 1:num_features){ |
||
| 75 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 76 | } |
||
| 77 | |||
| 78 | covariates <- as.data.frame.matrix(covariates) |
||
| 79 | }, |
||
| 80 | { #### Model 4 |
||
| 81 | |||
| 82 | sigma <- diag(num_features) |
||
| 83 | correl <-0.70 |
||
| 84 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 85 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 86 | diag(sigma) <- 1 |
||
| 87 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 88 | |||
| 89 | |||
| 90 | 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); |
||
| 91 | 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); |
||
| 92 | logit_pY1 <- prog_part + treatment * theta_pred * pred_part; |
||
| 93 | prog_features <- 1:5; pred_features <- 6:10; |
||
| 94 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 95 | |||
| 96 | for (index_feature in 1:num_features){ |
||
| 97 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 98 | } |
||
| 99 | |||
| 100 | covariates <- as.data.frame.matrix(covariates) |
||
| 101 | }, |
||
| 102 | |||
| 103 | |||
| 104 | { #### Model 5 |
||
| 105 | sigma <- diag(num_features) |
||
| 106 | correl <-0.70 |
||
| 107 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 108 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 109 | diag(sigma) <- 1 |
||
| 110 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 111 | prog_part <- covariates[,1]*(covariates[,2]*covariates[,3]-covariates[,4]^2) ;# prog_part <- prog_part - mean(prog_part); |
||
| 112 | pred_part <- covariates[,5]*exp(covariates[,6]*covariates[,7]-covariates[,8]^2) ;# pred_part <- pred_part - mean(pred_part); |
||
| 113 | logit_pY1 <- prog_part + treatment*theta_pred * pred_part; |
||
| 114 | prog_features <- 1:4; pred_features <- 5:8; |
||
| 115 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 116 | |||
| 117 | for (index_feature in 1:num_features){ |
||
| 118 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 119 | } |
||
| 120 | |||
| 121 | covariates <- as.data.frame.matrix(covariates) |
||
| 122 | }, |
||
| 123 | |||
| 124 | { #### Model 6 |
||
| 125 | sigma <- diag(num_features) |
||
| 126 | correl <-0.70 |
||
| 127 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 128 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 129 | diag(sigma) <- 1 |
||
| 130 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 131 | |||
| 132 | prog_part <- covariates[,1] + (covariates[,2]>-0.545) * (covariates[,3]<0.545) ;# prog_part <- prog_part - mean(prog_part); |
||
| 133 | pred_part <- covariates[,4] + (covariates[,5]>-0.545) * (covariates[,6]<0.545) ;# pred_part <- pred_part - mean(pred_part); |
||
| 134 | logit_pY1 <- prog_part + treatment*theta_pred * pred_part; |
||
| 135 | prog_features <- 1:3; pred_features <- 4:6; |
||
| 136 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 137 | |||
| 138 | for (index_feature in 1:num_features){ |
||
| 139 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 140 | } |
||
| 141 | |||
| 142 | covariates <- as.data.frame.matrix(covariates) |
||
| 143 | |||
| 144 | }, |
||
| 145 | { #### Model 7 |
||
| 146 | sigma <- diag(num_features) |
||
| 147 | correl <-0.70 |
||
| 148 | sigma[seq(1,num_features,by=2),seq(1,num_features,by=2)] <-correl # correlation between odds features |
||
| 149 | sigma[seq(2,num_features,by=2),seq(2,num_features,by=2)] <-correl # correlation between even features |
||
| 150 | diag(sigma) <- 1 |
||
| 151 | covariates <- mvrnorm(sample_size, rep(0, num_features), sigma) # I need install.packages("MASS") and library(MASS) |
||
| 152 | prog_part <- covariates[,1] + (covariates[,2]>-0.545) * (covariates[,3]<0.545)*(covariates[,4]>0);# prog_part <- prog_part - mean(prog_part); |
||
| 153 | pred_part <- covariates[,5] + (covariates[,6]>-0.545) * (covariates[,7]<0.545)*(covariates[,8]>0);# pred_part <- pred_part - mean(pred_part); |
||
| 154 | logit_pY1 <- prog_part + treatment*theta_pred * pred_part; |
||
| 155 | prog_features <- 1:4; pred_features <- 5:8; |
||
| 156 | irr_features <-setdiff(seq(1,num_features,by=1),c(prog_features,pred_features)) |
||
| 157 | |||
| 158 | for (index_feature in 1:num_features){ |
||
| 159 | covariates[,index_feature] = t(discretize( covariates[,index_feature], disc="equalwidth", nbins= sample(c(2,3,4,5),1))) |
||
| 160 | } |
||
| 161 | |||
| 162 | covariates <- as.data.frame.matrix(covariates) |
||
| 163 | }, |
||
| 164 | |||
| 165 | |||
| 166 | |||
| 167 | stop("Wrong model") |
||
| 168 | ) |
||
| 169 | |||
| 170 | |||
| 171 | |||
| 172 | pY1 <- 1/(1+exp(-logit_pY1)) |
||
| 173 | labels <- rbinom(sample_size,1,pY1); |
||
| 174 | |||
| 175 | |||
| 176 | |||
| 177 | synthetic_datasets <- numeric(0) |
||
| 178 | synthetic_datasets$data <- covariates |
||
| 179 | synthetic_datasets$treatment <- treatment |
||
| 180 | synthetic_datasets$labels <- labels |
||
| 181 | synthetic_datasets$prog_features <- prog_features |
||
| 182 | synthetic_datasets$pred_features <- pred_features |
||
| 183 | synthetic_datasets$irr_features <- irr_features |
||
| 184 | |||
| 185 | return(synthetic_datasets) |
||
| 186 | |||
| 187 | } |
||
| 188 | |||
| 189 | |||
| 190 | |||
| 191 |