API Reference - DistributedTrainingStrategies - DistributedGradientsCoordinator

DistributedGradientsCoordinator is a base class for distributed gradient ascent / descent.

Constructors

new()

Create new model object. If any of the arguments are nil, default argument values for that argument will be used.

DistributedGradientsCoordinator.new(gradientChangeMode: string): DistributedGradientObject

Parameters:

  • gradientChangeMode: Set what to do with the model parameters for a given gradient. Available options are:

    • Descent (Default)

    • Ascent

Returns:

  • DistributedGradientsCoordinatorObject: The generated distributed gradient training object.

Functions

setParameters()

Set model’s parameters. When any of the arguments are nil, previous argument values for that argument will be used.

DistributedGradientsCoordinator:setParameters(gradientChangeMode: string)

Parameters:

  • gradientChangeMode: Set what to do with the model parameters for a given gradient. Available options are:

    • Descent

    • Ascent

addGradients() 

DistributedGradientsCoordinator:addGradients(Gradients: any)

Parameters:

  • Gradients: The table of matrices / matrix containing the gradient values.

setModelParameters() 

DistributedGradientsCoordinator:setMainModelParameters(ModelParameters: any, doNotDeepCopy: boolean)

Parameters:

  • ModelParameters: The model parameters for the main model.

  • doNotDeepCopy: Set whether or not to deep copy the model parameters.

getModelParameters() 

DistributedGradientsCoordinator:getModelParameters(doNotDeepCopy: boolean): any

Returns:

  • ModelParameters: The model parameters for the main model.

  • doNotDeepCopy: Set whether or not to deep copy the model parameters.

start()

Creates a new thread for real-time gradient descent / ascent.

DistributedGradientsCoordinator:start(): coroutine

Returns:

  • gradientChangeCoroutine: A coroutine that handles the modification of the model parameters.

stop()

Stops the threads for real-time training.

DistributedGradientsCoordinator:stop()

clearGradients()

Clears the stored gradients inside the DistributedGradients object.

DistributedGradientsCoordinator:clearGradients()

Inherited From