backprop: Backpropagate data on a neural network

In wiper8/AI: Provide AI subject functions.

Description

Backpropagate data on a neural network

Usage

backprop(
  nn,
  newdata,
  n_epoch = 100,
  step_size = 0.1,
  lr = 1,
  batch_size = ceiling(nrow(newdata)/10),
  algo = "backprop",
  trace = FALSE
)

Arguments

nn	neural network
newdata	data.frame of the inputs and the outputs
n_epoch	integer : number of iterations of backpropagation going through every data once.
step_size	numeric : multiplier of the gradient.
lr	numeric : learning rate. new_param = old_param x (1-lr) + lrxbackproped_param
batch_size	integer : size of batch of backpropagation.
algo	character : convergence algorithm either "backprop" or "rprop+".
trace	logical : printing last ran epoch?

Value

list of neural network (class : nn) and the mean Loss

wiper8/AI documentation built on Dec. 23, 2021, 5:15 p.m.