Package net.siefkes.nlstego.predict

The classes contained in this package provide functionality for building statistical prediction models and for predicting items based on these models.

See:
Description

Interface Summary

StatelessFilter

This interface filters which predictions should be stored in the stateless (global) model.

Class Summary

FetchStats	Abstract class for collecting statistics about the fetching of items that can be used for predicting.
Metrics	Implementations of this abstract class handle performance measuring and logging.
PPMFetchStats	Implementation of the FetchStats class using a Prediction-by-Partial-Match strategy.
PredictedToken	Wrapper for a predicted token.
PredictionManager	Manages the predictions of items, combining different prediction models.

Package net.siefkes.nlstego.predict Description

The classes contained in this package provide functionality for building statistical prediction models and for predicting items based on these models.

The implemented prediction strategy is the PPM (Prediction by Partial Match) strategy, which utilizes higher-order Markov models. A blending algorithm is used to combine the results of Markov predictors of different orders. The pluggable architecture permits complementing or replacing the PPM algorithm by other strategies.

The diploma thesis of Christian Siefkes contains a complete description of the toolkit.