R/simulations.R

In IshmaelBelghazi/bigoptim: Efficient Large Scale Convex Finite Sums Optimization

.simulate_binomial <- function(true_params, sample_size, intercept=FALSE) {

    n_params <- length(true_params)
    X <- matrix(rnorm(n_params * sample_size), ncol=n_params)
    if (intercept) {
        X[, 1] <- 1
    }
    z <- X %*% true_params
    pr <- 1/(1 + exp(-z))
    y <- 2 * rbinom(sample_size, 1, pr) - 1
    y <- matrix(y, ncol=1)
    return(list(X=X, y=y, true_params=true_params))
}

.simulate_gaussian <- function(true_params, sample_size, intercept=FALSE) {
    n_params <- length(true_params)
    X <- matrix(runif(n_params * sample_size), ncol=n_params)
    if (intercept) {
        X[, 1] <- 1
    }
    z <- X %*% true_params
    y <- z + rnorm(n=sample_size)
    y <- matrix(y, ncol=1)
    return(list(X=X, y=y, true_params=true_params))
}

.simulate_exponential <- function(true_params, sample_size, intercept=FALSE) {
    n_params <- length(true_params)
    X <- matrix(runif(n_params * sample_size), ncol=n_params)
    if (intercept) {
        X[, 1] <- 1
    }
    z <- X %*% true_params
    pr <- pmax(-1/z, 0.1)
    y <- rexp(sample_size, pr)
    y <- matrix(y, ncol=1)
    return(list(X=X, y=y, true_params=true_params))
}

.simulate_poisson <- function(true_params, sample_size, intercept=FALSE) {
    n_params <- length(true_params)
    X <- matrix(runif(n_params * sample_size), ncol=n_params)
    if (intercept) {
        X[, 1] <- 1
    }
    z <- X %*% true_params
    pr <- exp(z)
    y <- as.numeric(rpois(sample_size, pr))
    y <- matrix(y, ncol=1)
    return(list(X=X, y=y, true_params=true_params))
}