sgd: (Batch) (Stochastic) Gradient Descent

In jonlachmann/irls.sgd: Stochastic IRLS with SGD

(Batch) (Stochastic) Gradient Descent

Description

(Batch) (Stochastic) Gradient Descent

Usage

sgd(
  gradient,
  data = NULL,
  ctrl = list(start, alpha, decay, subs, maxit, histfreq)
)

Arguments

gradient	The gradient of the objective function.
data	The dataset to use for the gradient, if applicable, if NULL, no data is used.
ctrl	A list of control parameters.

Details

The control parameters that can be set are

• "start" The starting point, NULL means random around 0.

• "alpha" The initial step size, defaults to 0.0005.

• "decay" The exponential decay of the step size per iteration, defaults to 0.999.

• "subs" The subsample percentage to use at each iteration, defaults to 1 (100%).

• "maxit" The maximum number of iterations to run for, defaults to 1000.

• "histfreq" The number of iterations between stored history frames, defaults to 50.

jonlachmann/irls.sgd documentation built on March 11, 2023, 7:42 a.m.