NAME

id - run the decision tree learning algorithm over a training set

SYNOPSIS

id(+RelationName)

DESCRIPTION

The id function induces a decision tree from examples given as a relation in Prolog's database. The output is a frame that contains the new tree.

Like the other propositional learners, id requires a mode declaration to specify the types of attributes that the relation has.

The mode declaration defines the range of values each attribute is allowed to take. The last attribute is always taken to be the class.

EXAMPLE

mode object(
                size(small, medium, large),
                colour(red, blue, green),
                shape(wedge, brick, pillar, sphere),
                class(yes, no)
).

object(medium, blue, brick, yes).
object(small, red, sphere, yes).
object(large, green, pillar, yes).
object(large, green, sphere, yes).

object(small, red, wedge, no).
object(large, red, wedge, no).
object(large, red, pillar, no).

: id(object)?
id0
: pf id0!
id0 isa [id] with
        creator: id(object);
        date: '97/10/21 - 10:45:24';
        n_examples: 7;
        errors: 0;
        time: 0;
        rule: 
object(Size, Colour, Shape, Class) :-
        (Shape = wedge -> Class = no
        |Shape = brick -> Class = yes
        |Shape = pillar -> 
                (Colour = red -> Class = no
                |Colour = green -> Class = yes)
        |Shape = sphere -> Class = yes).
!

IMPLEMENTATION

This algorithm is a mix of the entropy function used for splitting in Quinlan's ID3 and Niblett and Bratko's pruning, which is used in Assistant.

BUGS

This implementation can't handle continuous attributes yet.

SEE ALSO

aq, induct