If you wish to create your own models and optimizers from our library, we already have set a standard for our API design.
All our models have train() and predict() functions. They will be called when using some parts in “Others” section of library.
train() function takes in featureMatrix / tableOfTokenSequenceArray (mandatory for all models) and labelVector / tableOfTokenSequenceArray (not available for some models) in order.
predict() function takes in featureMatrix or featureVector or tableOfTokenSequenceArray.
The code for the models are object-oriented.
All our optimizers have calculate() and reset() functions. They will be called inside called inside our models.
calculate() function takes in learningRate (number) and costFunctionDerivatives (matrix) in order. It returns the adjusted costFunctionDerivatives.
reset() does not take in any parameters.
The code for the optimizers are object-oriented.
You can get more optimizer formulas here.
All our regularization objects have calculateRegularization() and calculateRegularizationDerivatives() functions. They will be called inside called inside our models.
Both takes in modelParameters (matrix) and numberOfData (integer) in order.
calculateRegularization() returns regularization values for modelParameters (matrix).
calculateRegularizationDerivatives() returns regularization values for costFunctionDerivatives (matrix).
The code for the regularization objects are object-oriented.