API Reference - Others - OnlineLearning

Online learning allows models to update continuously as it receives new data, making it capable of real-time training.

Constructors

new()

Creates a new online learning object

 OnlineLearning.new(Model: ModelObject, isOutputRequired: boolean, batchSize: integer): OnlineLearningObject

Parameters:

• Model: The model to be trained.

• isOutputRequired: Set whether or not the model requires labels / token output sequence arrays.

• batchSize: The size of data needed before training the model.

Functions

start()

Creates new threads for real-time training.

OnlineLearning:start(showFinalCost: boolean, showWaitWarning: boolean): coroutine

Parameters:

• showFinalCost: Set whether or not the final cost is displayed when training is complete.

• showWaitWarning: Set whether or not to show that the model have been waiting for data for more than 30 seconds.

Returns:

• trainCoroutine: A coroutine that trains the model.

stop()

Stops the threads for real-time training.

OnlineLearning:stop()

addInput()

Adds feature vector / token input sequence array to to queue.

OnlineLearning:addInput(input: matrix / tokenSequenceArray[])

Parameters:

• input: A (1 x n) matrix / a token input sequence array to be added to the reinforcement queue.

addOutput()

Adds label / token output sequence array to queue.

OnlineLearning:addOutput(output: integer / tokenSequenceArray[])

Parameters:

• output: The actual label related to the previous feature vector / a token output sequence array.

returnCostArray()

Returns cost array from the queue.

OnlineLearning:returnCostArray(): number[]

Returns:

• costArray: The cost array related to actual value and the predicted value.

Notes:

• Be aware that the model may suffer concept drift from training over long periods of time.