WCC.CG: WCC.CG Constrained descent optimizer

Description Usage Format Methods

Description

WCC.CG Constrained descent optimizer

Usage

1

Format

An object of class R6ClassGenerator of length 24.

Methods

initialize(weights.initial)

Creates a new computer for determining the best weighted combination of the ML libraries based on stochastic gradient descent and simplex projection. @param weights.initial vector (default = NULL) the initial vector of weights to use. Can be NULL if the number_of_algorithms is specified.

@param number_of_algorithms integer (default = NULL) the number of algorithms to calculate the weights for. Is used to initialize the weights. Can be NULL if the weights are provided

compute(Z, Y, libraryNames, ...)

Method to compute the best weighted combination for a set of estimators. In this implementation we use a two-step approach, in which we first estimate the weights based on a gradient descent procedure, and then project these weights back to the L1 simplex, scaling them between 0-1.

@param Z matrix containing the outcomes of each of the estimators

@param Y vector containing the actual observed outcome.

@param libraryNames vector containing the names of the optimizer.

@param ... other parameters to pass to the underlying combination computers.

@return vector of the trained / updated weights.


frbl/OnlineSuperLearner documentation built on Feb. 9, 2020, 9:28 p.m.