^inductive programming^

Some comparisons

other inductive programming
Weka:
System / library.
Implementation language: Java.
Types of models: Some, e.g. 'classifier'.
Implementation: Haskell.
 
Strong, compile-time, polymorphic types for data & models (thanks to Haskell).
ILP, Inductive Logic Programming:
Searches for a hypothesis in the form of a logic program.
V. large search space, unless given "hints".
Some work on complexity of hypotheses and/or proof.
User closely involved in the selection of the search space and design of search process.
 
Strong notion of complexity / information-content of hypothesis and of data|hypothesis.
users.monash.edu/~lloyd/Seminars/2005-II/Compare/index.shtml  

Weka's BN

Weka's Bayesian networks "assume that all variables are discrete"[Weka] p.22 and "a limitation of the current classes is that they assume that there are no missing values"[Weka] p.23.

In Weka, continuous variables must be discretised first and the way this is done may affect the outcome. This is unnecessary for modelling and, for splitting, is part of the network optimisation when using our [IP] classification trees.


[Weka] R. R. Bouckaert. Bayesian networks in Weka. TR 14/2004, Comp. Sci. Dept.. U. of Waikato, Sept. 2004.

[IP] L. Allison. Types and classes of machine learning and data mining. 26th Australasian Comp. Sci. Conf. (ACSC), Adelaide, pp.207--215, Feb. 2003.





as
this is to that  
subject of this talk inductive inference (arbitrary types of data)
list prelude list processing (arbitrary element types)
{parser combinator} parsing (chars, strings, symbols, parse trees)
embedded language for X X  
denotational semantics of L L  
?inductive programming? statistical model  
functional programming function  
Created with "vi (Linux & Solaris)",   charset=iso-8859-1