This project aims to find similarities in a set of formatted data and propose one or more characteristic(s) description(s) that carry the most information.

A presentation of the project is available in English and in French.

Project

The application is based on the concept of model, examples (and counter-examples), objects, attributes and relations:

A model is the definition of the formatted data on which the program works. This is where the attribute(s) and the relation(s), that describe it, are defined. (ex: a geometric shape, described by its shape, color and size)
An object is an instantiation of the model (ex: a square, red of size 8)
An example (or counter-example) is a set of objects. Each example is independant of the others.
An attribute describes an aspect of an object and can be of three type:
- numerical, an integer interval (ex: a size)
- discrete, an enumeration (ex: a color)
- hierarchical, a tree (ex: a shape)
A relation describes a directionnal connection between two objects in the same example (ex: A is next-to B)

This project is part of an enginnering school programming project, here the original subject (french).

How to populate input files

The programs use two different types of file to generate the solutions of a problem. These files are located in the ressources directory.

Model

The information on the model (ie: the definition of the object's attributes, the possible relations, the bounds of the values, etc...) is defined in a file with the cfg extension. It follows this syntax:

 attribute-name : attribute-value
 relation : relation-name[, relation-name]
 # each value can be of one of the three following types:
 # integer, in interval, defined like this:
 integer-attribute : <min> - <max>
 # an enumeration item, from an enumeration, defined like this:
 enum-attribute : <item> [, <item>]
 # a tree node, from a tree, defined like this:
 tree-attribute : (<node-name>, (<child-name>[, ...]), (<child-name>))
 # a tree definition can be split on more than one line to allow a greater readability
 
 # to define the possible relations between the objects, you must use the 'relation' keyword
 relation: relation-name[, relation-name]

See config-commented.cfg for further explanations.

Examples

The model defines the rules behind the examples, which are defined in a file with the exp extension. It follows this syntax:

 include <path-to-model-file>
 
 Example-name
     object-name:attribute-name(attribute-value)[, attribute-name(attribute-value)][,relation-name(object-linked-name)]
     [...]
 !Counter-example-name
     # same syntax as of the examples

See examples-commented.exp for further explanations.

Documentation

A documentation is also available, with graph to a better understanding of our process (see Files tabs).

To update the documentation, make sure you have installed Doxygen and Graphviz.

Then browse to the root of the project folder and run: doxygen doc/doxygen.cfg

Usage and options

Compile with make or make learning.

See makefile for further explanations.

To run the program only need the path to the example file as a parameter: learning <path-to-example-file> [options]

It accept some options:

--help or -h Print some help on basic options
--expand-relations When the solution have relations, print the whole object linked instead of its name
--no-generalization Skip the generalization step (print all the solutions, even duplicates and non-important solutions)
--no-counter-examples Skip the counter-examples handling. Prevent the suppression of solutions, even if they match the counter-examples in the example file
--no-color Remove all colors from the output
-v Verbosity level, you can use up to 4 v to add information