decision-tree

Reasonably fast discrete decision tree. Useful for:

• Diagnosing under what circumstances X happens.

Installation:

npm require fast-decision-tree

Usage:

var dt = require('fast-decision-tree');
var data =
[ {a:1,b:2,z="pink"}
, {a:3,b:1,z="violet"}
, {a:9,b:2,z="pink"}
];
var tree = dt.create_decision_tree(data,'z');
console.log(JSON.stringify(tree,null,2));

API

create_decision_tree(data, target_attribute, options={})

data is an array of records. Each may be a simple dictionary of attribute-value pairs: {whatever:12, else:99} or it may be a complete record of the type used by the decision tree internally: {weight:100, attr:{whatever:12, else:99}}. If you use the latter form you must tell dt by setting option.weighted=true.

target_attribute is the attribute we wish to understand.

options:

• attributes are the terms in which we wish to understand the target attribute.

• fitness_function(data, attribute, target_attribute)==get_gain This is a function that returns the improvement or gain, e.g. the reduction in entropy, that would be provided if we branch on attribute. Here data comes in the internal format described previously. The default fitness function is require('fast-decision-tree').fitness_functions.get_gain.

• min_gain==-100 Don't bother branching if the gain is below this value.