GARP Caveat

This output is produced by an experimental modelling tool which is still under development. It is supplied for comment, and without warranty, although reasonable efforts will be made to act on suggestions.

In any modelling system the data included is very important in determining the quality of the results. The GARP system finds the best model it can, using minimal assumptions, to maximise the predictive accuracy and explanatory capacity of the set of output rules. Where information is not available, and due to chance, good rules are not found, conflicts between rules and less that 100% accuracy results.

While some capacity to deal with biases in the distribution of the data is claimed, no claims are made as to reliability of the output under extreme biases.

For a number of reasons genetic algorithms can provide different solutions on successive runs.

• The genetic algorithm is a non-deterministic algorithm, i.e. the time it takes to iteratively convergence to a stable state cannot be determined prior to running the algorithm. Thus a stopping condition is used when the model is judged "stable enough".
• While the algorithm may converge to a stable state, a (large) number of final states may be possible.
• The data provided to GARP is randomly sampled from a larger population and is therefore different every time.