API Reference - Others - DistributedModelParameters

DistributedTraining is a base class for distributed learning. The individual child models will calculate their own model parameters and these will create a new main model parameters using average.

Notes:

  • The child models must be created separately. Then use addModel() to put it inside the DistributedLearning object.

Constructors

new()

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

DistributedModelParameters.new(totalNumberOfChildModelUpdatesToUpdateMainModel: number): DistributedLearningObject

Parameters:

  • totalNumberOfChildModelUpdatesToUpdateMainModel: The required total number of reinforce() and train() function calls from all child models to update the main model.

Returns:

  • DistributedLearningObject: The generated distributed learning object.

Functions

setParameters()

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

DistributedModelParameters:setParameters(totalNumberOfChildModelUpdatesToUpdateMainModel: number)

Parameters:

  • totalNumberOfChildModelUpdatesToUpdateMainModel: The required total number of reinforce() and train() function calls from all child models to update the main model.

addModelParameters() 

DistributedModelParameters:addModelParameters(ModelParameters: Matrix/TableOfMatrices)

Parameters:

  • ModelParameters: The model parameters to be received by the DistributedModelParameters.

setModelParametersMerger()

Sets the ModelParametersMerger into the DistributedModelParameters.

DistributedModelParameters:setModelParametersMerger(ModelParametersMerger: ModelParametersMergerObject)

ModelParametersMerger: A ModelParametersMerger object to be used by the DistributedModelParameters object

setMainModelParameters() 

DistributedModelParameters:setMainModelParameters(MainModelParameters: any)

Parameters:

  • MainModelParameters: The model parameters for the main model.

getMainModelParameters() 

DistributedModelParameters:getMainModelParameters(): any

Returns:

  • MainModelParameters: The model parameters for the main model.

getCurrentTotalNumberOfChildModelUpdatesToUpdateMainModel() 

DistributedModelParameters:getCurrentTotalNumberOfChildModelUpdatesToUpdateMainModel(): number

Returns:

  • getCurrentTotalNumberOfChildModelUpdatesToUpdateMainModel: The current total number of reinforcements from all child models.

start()

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

DistributedModelParameters:start(): coroutine

Returns:

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

reset()

Reset the main model’s stored values (excluding the parameters).

DistributedModelParameters:reset()

destroy()

Destroys the model object.

DistributedModelParameters:destroy()