R/transfer.R

In qrnn: Quantile Regression Neural Network

Defines functions softmax lrelu lrelu.prime logistic logistic.prime relu relu.prime softplus softplus.prime sigmoid sigmoid.prime elu elu.prime linear linear.prime

Documented in elu elu.prime linear linear.prime logistic logistic.prime lrelu lrelu.prime relu relu.prime sigmoid sigmoid.prime softmax softplus softplus.prime

linear.prime <-
function(x)
{
    x*0 + 1
}

linear <-
function(x)
{
    x
}

elu.prime <-
function(x, alpha=1)
{
    ifelse(x >= 0, 1, elu(x, alpha) + alpha)
}

elu <-
function(x, alpha=1)
{
    ifelse(x >= 0, x, alpha*(exp(x)-1))
}

sigmoid.prime <-
function(x)
{
    (0.5)*(1-tanh(0.5*x)^2)
}

sigmoid <-
function(x)
{
    tanh(0.5*x)
}

softplus.prime <-
function(x, alpha=2)
{
    y <- exp(alpha*x)/(1+exp(alpha*x))
    ifelse(is.nan(y), 1, y)
}

softplus <-
function(x, alpha=2)
{
    y <- log(1+exp(alpha*x))/alpha
    ifelse(is.infinite(y), x, y)
}

relu.prime <- function(x){
    ifelse(x >= 0, 1, 0)
}

relu <- function(x){ 
    ifelse(x >= 0, x, 0)
}

logistic.prime <- function(x){
    0.25/(cosh(x/2)^2)
}

logistic <- function(x){
    0.5 + 0.5*tanh(x/2)
}

lrelu.prime <- function(x){
    ifelse(x >= 0, 1, 0.01)
}

lrelu <- function(x){
    ifelse(x >= 0, x, 0.01*x)
}

softmax <- function(x){
    exp(x)/sum(exp(x))
}

Try the qrnn package in your browser