Skip to content

diegojromerolopez/djbdd

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DJBDD

What's this?

A Java 7 BDD package with the GPL with classpath linking exception license based on GPL version 2 or later, as you prefer. See this explanation about the license. Howeer, if your use case requires other license, I can re-license this project for you, please write me to my email (at the botton of this README).

This package provides a Binary Decision Diagram library you can use to make operations with boolean logical formulas and study its properties.

This software has been developed by Diego J. Romero-López as a tool for his Master's Thesis.

Hay una versión en español de este documento aquí.

Introduction

A Binary Decision Diagram is a complete truth table of a boolean expression in a reduced graph form. For an introduction, see Binary Decision Diagram on Wikipedia.

This library provides all the operations you need to work with them.

How to use it

// These are the boolean variables used in our formulas
// They can have any alphanumeric name you want starting with a letter and
// being unique variable names.
String[] variables={"a", "b", "c", "d", "e", "f"};

// You always have to initialize the BDD system before you create any
// BDD object. Be careful with that.
BDD.init(variables);

// The functions are specified as boolean expressions with Java syntax
// adding the operators -> (implication) and <-> (double implication)
String function = "(a && b) || (c && d)";

// The variable ordering is given by the order
// of variables in the array of strings 'variables'.

// Construction of a new BDD
BDD bdd = new BDD(function);

// Printing the BDD in the standard output
// The internal table node will be shown by stdout
bdd.print();

// You can print it as a image PNG using a dot library
Printer.printBDD(bdd, "bdd_"+bdd.size());

// Creating other BDDs:
String function2 = "(a && e) || f"
BDD bdd2 = new BDD(function2);
bdd2.print();

// Operations with BDDs

// This BDD is the logical AND between bdd and bdd2
BDD bdd3 = bdd.apply("and", bdd2);

// This BDD is the logical OR between bdd and bdd2
BDD bdd4 = bdd.apply("or", bdd2);

// Destroy explicitly bdd2
// In case we need to compute sizes, reduce the global graph or print it
bdd2.delete();

// If you think you can have few free memory,
// you would have to free by calling the garbage collector
BDD.gc();

Complete tests

The main program of this library is full of examples. The all-in-one jar version is located in store/DJBDD.jar, the dependant on libraries jar one is in dist/DJBDD.jar. You should use store/DJBDD.jar for your experiments.

Easiest option is opening this project with Netbeans. There are several executing configurations:

  • Two examples of computing consistency in the real world.
  • Six benchmark of BDD-reducing algorithms.
  • Tests that you can execute to verify BDD operations.

Basic Options

Output
BDD print as image
java -jar DJBDD.jar --image --<format> <file>

Note that this option will work only if you are in a Linux/UNIX system with the dot tool to draw graphs in the path /usr/bin/dot.

BDD printing
java -jar DJBDD.jar --print --<format> <file>

Prints a BDD in the standard output.

Restriction on BDD formats allowed
  • dimacs: Dimacs CNF format. See SAT format or CNF for more information.
  • cstyle: C-style boolean expression preceded by a line with all variables separated by commas. For example:
    • a && !b
    • a -> (!b && (c || d))
    • a <-> (b && !c)
    • a != (b && !c)
Restriction on variable naming

IMPORTANT: for each variable there can no be any other that contains it as substring from the left. That is, if we have a variable with the name 'x1' we cannot use other variable with the name 'x11'. It's not in my future plans to change that, so name your variables with names like:

  • {x1}
  • {x11}
  • {x12}
Other notes

Example source data: directory data has some examples of each format (look the extension).

BDD reduction benchmarks

See below for a description of each reduction method.

# Sifting Algorithm
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks/ sifting
# Window Permutation Algorithm (window size = 2)
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks/ window_permutation window_size=2
# Random Swapper
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks/ random_swapper random_seed=121481 iterations=100
# Genetic Algorithm
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks-genetic genetic random_seed=10 population=8 generations=10 selection_percentage=0.2 mutation_probability=0.1
# Memetic Algorithm
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks-memetic memetic random_seed=121481 population=8 generations=10 selection_percentage=0.2 mutation_probability=0.1
## Iterative Sifting
java -jar ./DJBDD/store/DJBDD.jar --memory-optimization-benchmark --dimacs ./data/benchmarks/ isifting iterations=100

Features

Shared hash table

All BDDs use the same hash table, sharing the vertices and subgraphs. The goal for using this data structure is reducing the number of repeated vertices. Each BDD has one root vertex though.

Rich I/O API

This library provides method to load logical clausules in DIMACS format and Java native, includen implication and double implication operators.

Memory efficient

This library shares vertices between different BDDs.

Vertices grouped by levels

You can access the vertices that has each variable in an efficient way. This will be used for BDD-reducing algorithms.

Operations implemented

  • Apply [6] [3].
  • Restrict. [2].
  • Swapping of two variables. [5].

Implemented Reduction Algorithms

This package contains many reduction algoritms. They are implemented in the djbdd.reductors package.

These reduction algorithms are implemented as children classes of djbdd.reductors.ReductionAlgorithm, so they share the same API. This API contains a execute and run method. The first one must be overriden for each particular reduction method while the second will act as a façade of execute. The run method also measures the elapsed time in the reduction method for comparing the reduction algorithms

// Some logic function used to build a BDD
String function1 = "(a && d) || (b && c)";
BDD bdd1 = new BDD(function1);

// Call to garbage collector & print the
// node table 
BDD.gc();
BDD.T.print();
        
Printer.printBDD(bdd1, "test15_bdd1_BEFORE_"+bdd1.size());

// Construct a reduction algorithm. For example SiftingReductor:
SiftingReductor reductor = new SiftingReductor();
// Start the reduction process
reductor.run();

// Call to garbage collector & print the
// node table (to compare with the first node table)
BDD.gc();
BDD.T.print();

It is important to note that these reduction methods reduce all the BDDs created in the program because they reduce the vertex mesh by swapping vertex levels.

They are listed below:

Window Permutation

Developed in the class djbdd.reductors.WindowPermutation, this algorithm was proposed by Fujita et al. & Ishiura et al. Richard L. Rudell described it and compared with its own reduction method in [7]. This implementation is based on his description.

// Window size
int windowSize = 2;
WindowPermutationReductor reductor = new WindowPermutationReductor(windowSize);
reductor.run();

Rudell's Variable Sifting

This package contains a basic implementation of the variable reordering proposed by Richard L. Rudell in [7]. This method try to find the best position for each variable keeping fixed in their position the rest.

Use example:

SiftingReductor reductor = new SiftingReductor();
reductor.run();

Random Swapper Reduction

We wanted to compare a pure random algorithm with our evolutionary algorithms to test if our results are due to randomness or because this method truly works. This reduction method is based in swapping two variables chosen at random in each iteration in the BDD tree.

int iterations = 1000;
RandomSwapperReductor reductor = new RandomSwapperReductor(iterations);
reductor.run();

Note that this algorithm, the genetic algorithm and the memetic algorithm need a random seed to be set before their execution.

Genetic Reduction

W. Lenders & C. Baier defined the genetic operators for developing a Genetic Algorithm for this BDD reduction problem in [8]. We have implemeted a version of their approach in the GeneticReductor class.

// Number of chromosomes
int populationSize = 10;
// Number of generations of the algorithm
int generations = 1000;
// % of population selected
double selectionPercentage = 10;
// Probability of mutating a gene in a chromosome
double mutationProbability = 0.1;
GeneticReductor reductor = new GeneticReductor(populationSize, generations, selectionPercentage, mutationProbability);
reductor.run();

Memetic Reduction

Based on [8], we propose another algorithm that combines the Genetic Algorithm optimizing each chromosome using the Rudell's Sifting Algorithm. Its name is MemeticReductor and its parameters are the same than the GeneticReductor.

// Number of chromosomes
int populationSize = 10;
// Number of generations of the algorithm
int generations = 1000;
// % of population selected
double selectionPercentage = 10;
// Probability of mutating a gene in a chromosome
double mutationProbability = 0.1;
MemeticReductor reductor = new MemeticReductor(populationSize, generations, selectionPercentage, mutationProbability);
reductor.run();

Iterative Sifting

This is our original method [9] that applies a serious of sifting of variables with hopes of finding the best position for each one.

Note that this approach seems to be almost equal to Converging Sifting made by R. Rudell in Dynamic variable ordering for ordered binary decision diagrams.

IterativeSiftingReductor contains this reduction method and can be used this way:

int iterations = 100;
IterativeSiftingReductor reductor = new IterativeSiftingReductor(iterations);
reductor.run();

A paper with the description of this method is pending evalution.

TODOs

  • Include a parallel apply.
  • Parallelization of reduction methods.

Examples

Code examples

Look the class Tester class in package djbdd.test, its full of examples.

Documentation

There is a pdf file called manual.pdf with some remarks about the implementation and how to use this library. Currently it is a work in progress, but you can read the tests, the code or ask me any question. If you are interested, there is also the Master's Thesis Memoria del Trabajo Fin de Máster (only in Spanish).

Bibliography

[1] Symbolic Boolean Manipulation with Ordered Binary Decision Diagrams, Randal E. Bryant. Carnegie Mellon University.

[2] Binary Decision Diagrams. Fabio Somenzi.

[3] Efficient implementation of a BDD package, Karl S. Brace, Richard L. Rudell & Randal E. Bryant.

[4] Implementation of an Efficient Parallel BDD Package. Tony Stornetta & Forrest Brewer.

[5] Incremental Reduction of Binary Decision Diagrams. R. Jacobi, N. Calazans & C. Trullemans.

[6] An Introduction to Binary Decision Diagrams. Henrik Reif Andersen.

[7] Dynamic variable ordering for ordered binary decision diagrams, Richard L. Rudell.

[8] Genetic Algorithms for the Variable Ordering Problem of Binary Decision Diagrams, W. Lenders & C. Baier.

[9] Iterative Sifting: A new approach to reduce BDD size. Diego J. Romero-López & Elena Ruiz-Larrocha. TBA. Algorithm already published by other author.

Cites

Please cite the author as Diego J. Romero López (Diego J. is author's name while Romero López is author's surname). If you want to write surname first, write Romero López, Diego J. In case you are interested, read more about Spanish naming customs.

If you use DJBDD and like to appear here, please send me an email.

FAQ

I don't understand what is a BDD

OK. The concept is not easy, read the Wikipedia page about them Binary Decision Diagram at Wikipedia and later the reference [1] and [2].

Do you use complement arcs?

No. I don't use them. I understand they can reduce the size of the graph, but I wanted to make a pure BDD library. Maybe I could include them in some future version.

How can I import this project to my IDE?

I've used Netbeans (yeah, I know it is ancient but works for me). You should be able to open with whatever IDE you use.

Can I use this library in Java5 or Java6?

I suppose. I have not used it in Java 6, but it should work.

Are there some code examples?

Look in the tests (djbdd.test.Tester class) and in the benchmarks.

I'm running the examples (or my custom code) and throws a exception

It depends on the exception type, but my money is on a memory-related exception. This library is memory greedy, use BDD.gc or help me to implement some reduction method on the vertex table :)

Why don't you use dynamic variable ordering when creating the BDD?

I prefer building the BDD "as is" and relying in the developer using some variable reordering algorithm, like Variable Sifting, implemented in this package.

Your code is inefficient/wrong/could be improved

I accept suggestions, critiques and comments you want to communicate me.

Why don't you use method?

Maybe I don't know it, please send me an email asking me and pointing me to a paper where it is explained. I will take a look to that.

What is the license of this code?

This code is GPL3 with classpath linking exception. That's the same license than the OpenJDK7 one. If you are unsure about this kind of license, read about it in.

Do you answer emails?

Of course I will answer questions, suggestions and comments about this library.

Who are you?

I'm Diego J., my email is die_gojr_om-erol-op_ez AT g-m-a-i-l.com (erase - and _ and replace AT with @). I'm from Spain and this is part of my master's final thesis.