OptimizerCosineAnnealing: Coordinate descent with cosine annealing

In schalkdaniel/compboost: Efficient Component-Wise Boosting Implementation

Coordinate descent with cosine annealing

Description

Same as OptimizerCoordinateDescent but with a cosine annealing scheduler to adjust the learning rate during the fitting process.

Arguments

ncores	(integer(1)) Number of cores over which the base learner fitting is distributed.
nu_min	(numeric(1)) Minimal learning rate.
nu_max	(numeric(1)) Maximal learning rate.
cycles	(integer(1)) Number of annealing cycles form nu_max to nu_min between 1 and anneal_anneal_iter_max. ⁠anneal_iter_max (⁠integer(1)⁠)\cr Maximal number of iterations for which the annealing is conducted. ⁠nu_min⁠is used as fixed learning rate after⁠anneal_iter_max'.

Format

S4 object.

Usage

OptimizerCosineAnnealing$new()
OptimizerCosineAnnealing$new(ncores)
OptimizerCosineAnnealing$new(nu_min, nu_max, cycles, anneal_iter_max, cycles)
OptimizerCosineAnnealing$new(nu_min, nu_max, cycles, anneal_iter_max, cycles, ncores)

Fields

This class doesn't contain public fields.

Methods

• ⁠$getOptimizerType()⁠: ⁠() -> character(1)⁠

• ⁠$getStepSize()⁠: ⁠() -> numeric()⁠

Examples

# Define optimizer:
optimizer = OptimizerCosineAnnealing$new()

schalkdaniel/compboost documentation built on April 15, 2023, 9:03 p.m.