API Reference - Containers - BaseContainer

Constructors

new()

Creates a new container object.


BaseContainer.new({ClassesList: {any}, cutOffValue: number, timeDependent: boolean, parallelGradientDescent: boolean, WeightBlockArray: {WeightBlock}, OutputBlockArray: {FunctionBlock}}): ContainerObject

Parameters:

  • timeDependent: An indicator that the container uses datasets that are time dependent. Enabling this will cause the container to always use slower backward propagation method. [Default: False]

  • parallelGradientDescent: An indicator that container will perform parallel gradient descent. See fastBackwardPropagate() function for more details. [Default: True]

  • WeightBlockArray: An array containing all the weight blocks that will be loaded to the container. [Default: {}]

  • OutputBlockArray: An array containing all the function blocks that will be loaded to the container. [Default: {}]

Returns:

  • Container: The generated container object.

Functions

forwardPropagate() 


Sequential:forwardPropagate(featureTensor: tensor): tensor

Parameters:

  • featureTensor: The feature tensor to be used as an input.

Returns:

  • generatedLabelTensor: The generated label tensor produced from passing across the function blocks inside the container object.

calculateWeightLossTensorArray() 


Sequential:calculateWeightLossTensorArray(lossTensor: tensor): {tensor}

Parameters:

  • lossTensor: The loss tensor to be used for calculating the weights of the function blocks.

Returns:

  • weightLossTensorArray: An array containing the weight loss tensors. The first tensor in the array represents the weight loss tensor for the first weight block.

gradientDescent() 


Sequential:gradientDescent(weightLossTensorArray: {tensor}, numberOfData: number)

Parameters:

  • weightLossTensorArray: An array containing the weight loss tensors. The first tensor in the array represents the weight loss tensor for the first weight block.

  • numberOfData: The value to divide with the weight loss tensors.

fastBackwardPropagate()

Performs the gradient descent after immediately receiving the weight loss tensor for that particular weight block.


Sequential:fastBackwardPropagate(lossTensor: tensor)

Parameters:

  • lossTensor: The loss tensor to be used for calculating the weights of the function blocks.

slowBackwardPropagate()

Performs the gradient descent after all weight loss tensors are received from all weight blocks.


Sequential:slowBackwardPropagate(lossTensor: tensor)

Parameters:

  • lossTensor: The loss tensor to be used for calculating the weights of the function blocks.

backwardPropagate()

A wrapper function that switches between fastBackwardPropagate() and slowBackwardPropagate().


Sequential:backwardPropagate(lossTensor: tensor)

Parameters:

  • lossTensor: The loss tensor to be used for calculating the weights of the function blocks.

setWeightTensorArray() 


Sequential:setWeightTensorArray(weightTensorArray: {tensor}, doNotDeepCopy: boolean)

Parameters:

  • weightTensorArray: An array containing all the weight tensors to be set to individual weight blocks. The first tensor in the array represents the weight tensor for the first weight block.

  • doNotDeepCopy: Set whether or not to deep copy the weight tensors in the weight array.

getWeightTensorArray() 


Sequential:getWeightTensorArray(doNotDeepCopy: boolean): {tensor}

Parameters:

  • doNotDeepCopy: Set whether or not to deep copy the weight tensors from the weight blocks.

setMultipleInputBlocks() 


Sequential:setMultipleInputBlocks(...: FunctionBlockObject)

Parameters:

  • TrainableBlock: The weight blocks to be added to the graph container. Order matters.

setMultipleInputBlocks() 


Sequential:setMultipleInputBlocks(...: FunctionBlockObject)

Parameters:

  • InputBlock: The function blocks to be added to the graph container. Order matters.

setMultipleOutputBlocks() 


Sequential:setMultipleOutputBlocks(...: FunctionBlockObject)

Parameters:

  • Function: The function blocks to be added to the graph container. Order matters.

getWeightBlockByIndex() 


ComputationalGraph:getWeightBlockByIndex(index): FunctionBlockObject

Parameters:

  • index: The index of the weight block.

Returns:

  • FunctionBlock: A function block from the specified index.

getWeightBlockArray() 


Sequential:getWeightBlockArray(): {FunctionBlockObject}

Returns:

  • WeightBlockArray: An array containing all the weight blocks. The first function block in the array represents the first weight block.

getInputBlockByIndex() 


ComputationalGraph:getInputBlockByIndex(index): FunctionBlockObject

Parameters:

  • index: The index of the input block.

Returns:

  • FunctionBlock: A function block from the specified index.

getInputBlockArray() 


ComputationalGraph:getInputBlockArray(): {FunctionBlockObject}

Returns:

  • FunctionBlockArray: An array containing all the function blocks. The first function block in the array represents the first function block.

getOutputBlockByIndex() 


ComputationalGraph:getOutputBlockByIndex(index): FunctionBlockObject

Parameters:

  • index: The index of the output block.

Returns:

  • FunctionBlock: A function block from the specified index.

getOutputBlockArray() 


ComputationalGraph:getOutputBlockArray(): {FunctionBlockObject}

Returns:

  • FunctionBlockArray: An array containing all the function blocks. The first function block in the array represents the first function block.

Returns

  • weightTensorArray: An array containing all the weight tensors from the weight blocks. The first tensor in the array represents the weight tensor for the first weight block.

Inherited From