Skip to content
Learn Finite State Automata that generalize over a set of training instances
Latest commit 029d56e Nov 13, 2010 @DrDub first commit
Failed to load latest commit information.
run_config first commit Nov 13, 2010
src/net/duboue/pablo first commit Nov 13, 2010
COPYING first commit Nov 13, 2010
README first commit Nov 13, 2010 first commit Nov 13, 2010
best-instance.png first commit Nov 13, 2010
toy-training.txt first commit Nov 13, 2010



This is an (incomplete) implementation of GAL, the Genetic Automaton Learner as 
defined in Chapter 5 of Belz (2000).

The point of this code is to learn a FSA that *generalizes* from a set of positive
examples. (If you just want to cover exactly the input, you can use any existing
package for doing FSA minimization.)

The input file is in text format, one sequence per line, the alphabet will be induced
by tokens separated by white-spaces.

This project currently builds under Eclipse, feel free to contribute an ant script.

To run an example, use the LearnFSA run configuration in run_config/

To create a graph from the output, install the package graphviz and use

  dot -Tpng > best-instance.png
Take a look at the PNG in the repository for an example of the learned automata.

Edit the to get better goodness of fit. Read chapter 5 of Belz (2000)
for more details and for further ideas about how to improve it. Patches are welcomed.

Pablo Duboue, November 2010


Belz, Anja (2000) "Computational Learning of Finite-State Models for 
Natural Language Processing", PhD Thesis, University of Sussex.

Something went wrong with that request. Please try again.