train.nn: Train feed-forward ANN

In pierrecattin/thesis-resources: Thesis Resources

Description

Train feed-forward ANN

Usage

train.nn(x.train, x.dev, y.train, y.dev, epochs, batch.size, activations,
  structure, l2, learning.rate, tb.log, perf.metric)

Arguments

x.train	matrix/xts n.obs*n.features
x.dev	matrix/xts n.obs*n.features
y.train	factor with levels c(-1, 1)
y.dev	factor with levels c(-1, 1)
epochs	number of eposch
batch.size	min batch size
activations	activation functions used in hidden layer: c("relu", "tanh", ...). For the output layer, softmax is used
structure	numeric vector indicating number of nodes in each hidden layer
l2	l2 regularisation value
learning.rate	learning rate
tb.log	character inidicating run name for tensorflow. if missing, tensorboard is not ran
perf.metric	c("mean.sr", "mean.accuracy")

Value

fit

pierrecattin/thesis-resources documentation built on May 4, 2019, 8:42 a.m.