Generalising Data Description for Machine Learning
Home Downloads Email

Abstract

The main aim of Inductive Programming is to make it possible to create solutions to new inductive inference problems with as little modification as possible to solutions for existing problems. This has initially been done through the creation of type classes which represent general statistical models. However, these models alone only provide a degree of generalisation, as the data-set being operated on must still be described through the creation of variable types to represent columns and the instantiation of these types in the appropriate type classes. The creation of these types and the creation of functions to operate on them can be extremely tedious and is error prone when done manually. This project presents a method for overcoming this problem by using the Haskell meta-programming extension Template Haskell to automatically generate appropriate types by examining the contents of the data file being operated on. It also provides various tools for generating utility functions to help with handling the newly generated types.




A llama evolves to a lamda and promptly invents Lisp

Last updated 29th November 2006.
A version which will (hopefully) continue to be updated can be found [here].