DataPredict

API Reference - Others - OnlineLearning

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

Creates a new online learning object

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

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.

Creates new threads for real-time training.

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

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.

Stops the threads for real-time training.

OnlineLearning:stop()

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

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

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

Adds label / token output sequence array to queue.

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

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

Returns cost array from the queue.

OnlineLearning:returnCostArray(): number[]

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