gd: Gradient Descent.

In jeksterslabds/jeksterslabRds: Jekster's Lab Data Science and Statistics

Description

Implements the gradient descent algorithm. Weights are updated using the following equation \mathbf{w} \rightarrow \mathbf{w} η Δ where Δ = f(\mathbf{Xw}) - \mathbf{y} and f is the activation function.

Usage

gd(X, y, aFUN, eta, epochs, criteria = 1e-08, final = TRUE, ...)

Arguments

X	The data matrix, that is an n \times k matrix of n observations of k regressors, which includes a regressor whose value is 1 for each observation.
y	n \times 1 vector of observations on the dependent variable.
aFUN	Activation function.
eta	η learning rate.
epochs	Number of iterations.
criteria	Stopping criteria. The algorithm stops if the sum of the absolute values of delta is less than criteria.
final	Logical. If TRUE, returns the value of w for the final iteration. If FALSE, returns all values of w from the random start to the final iteration.
...	Arguments to pass to aFUN.

Value

If final is TRUE, returns a vector of k estimated weights w for the final iteration. If final is FALSE, returns all the values of w from the random start to the final iteration.

jeksterslabds/jeksterslabRds documentation built on July 16, 2020, 3:41 p.m.