API Reference - Others - TrainingModifier

Modifies the model’s batch training to other modes.

Notes

  • When using “Stochastic” mode, make sure you set the model’s max number of iterations to 1.

  • batchSize is only applicable for “Minibatch” mode.

Constructors

new()

Creates a new training modifier object. If any of the arguments are not given, default argument values for that argument will be used.

TrainingModifier.new(Model: ModelObject, gradientDescentType: string, batchSize: integer, showOutput: boolean): GradientDescentModifierObject

Parameters:

  • Model: The model object to modify its training capabilities.

  • gradientDescentType: The type of gradient descent to be used when train() function is called. Available modes are “Batch”, “MiniBatch” and “Stochastic”.

  • batchSize: The batch size to split the featureMatrix and labelVector into multiple parts.

  • showOutput: Set whether or not to show the final cost for each epoch (MiniBatch) or data (Stochastic).

Returns:

  • TrainingModifierObject: A training modifier object that uses the model’s train(), predict() and reinforce() functions so that it behaves like a regular model.

Functions

train()

Trains the machine/deep learning model under specific gradient descent mode.

TrainingModifier:train(...): number[]

Parameters:

  • …: The parameters are the same to the original model’s train() function.

Returns:

  • costArray: An array containing cost values.

predict()

Predict the values for given data.

TrainingModifier:predict(...): ...

Parameters:

…: The parameters are the same to the original model’s predict() function.

Returns:

…: The outputs are the same to the original model’s predict() function.

reinforce()

Reward or punish model based on the current state of the environment.

TrainingModifier:reinforce(currentFeatureVector: Matrix, rewardValue: number, returnOriginalOutput: boolean): integer, number -OR- Matrix

Parameters:

  • currentFeatureVector: Matrix containing data from the current state.

  • rewardValue: The reward value added/subtracted from the current state (recommended value between -1 and 1, but can be larger than these values).

  • returnOriginalOutput: Set whether or not to return predicted vector instead of value with highest probability.

Inherited From