This is an old idea: the widget that would take a learner and a parameter that defines the model complexity. The widget would vary the parameter and display error for training and test set.
This may however be impossible to implement in general because models do not expose parameters (let alone with any information about which of them affect the complexity) and their sensible ranges.
This is an old idea: the widget that would take a learner and a parameter that defines the model complexity. The widget would vary the parameter and display error for training and test set.
This may however be impossible to implement in general because models do not expose parameters (let alone with any information about which of them affect the complexity) and their sensible ranges.