FTRLProximal (version 0.3.0)

ftrlprox.formula: FTRL Proximal formula

Description

Online elastic net regression using the FTRL Proximal algorithm for training.

Usage

"ftrlprox"(formula, data, lambda, alpha, a, b = 1, num_epochs = 1, save_loss = F, ...)

Arguments

formula

modeling formula

data

data.frame containing features and dependent variable

lambda

regularization term

alpha

mixing parameter, alpha=0 corresponds to L2 regularization and alpha=1 to L1.

a

learning rate parameter

b

learning rate parameter controlling decay, defaults to 1.

num_epochs

number of times we should traverse over the traiing set, defaults to 1.

save_loss

is to save the loss function during training.

...

additional args

Value

ftrlprox model object

Details

Test text

Examples

require(mlbench)

p <- mlbench.circle(100,2)
dat <- as.data.frame(p)

mdl <- ftrlprox(classes ~ x.1 + x.2 + I(x.1^2) + I(x.2^2), dat,
                a = 0.3, lambda = 5.0, alpha = 1.0)
print(mdl)