The goal of this training is to step a user through becoming familiar with the capabilities of MOEAFramework, a free and open source Java library created by Dave Hadka, that allows the user to design, execute, and test out a variety of popular multi-objective evolutionary algorithms (MOEAs). In this series, we will demonstrate the capabilities of MOEAFramework in 4 parts. Part 1 demonstrates how to hook up an external optimization problem to MOEAFramework. Part 2 will cover the optimization of the problem using a variety of algorithms. Part 3 will illustrate how to calculate metrics to assess the performance of the algorithms. Finally, Part 4 will step through generation of relevant figures from the metrics that convey the effectiveness, efficiency, reliability, and controllability of the algorithms.

The example test case that will be used throughout this tutorial is the DPS version of the Lake Problem. The code is written and adapted by Dr. Julianne Quinn and can be found here. However, the tutorial will be set up so that the user can easily swap in their own problem formulation.

Finally, this guide is specifically built to connect an external problem that is written in C++ but Dave Hadka’s Beginner Guide to the MOEA Framework, has examples of how to create the Java executable for problems written in a language other than C++.

Connecting Your Problem

Before starting this training, it is recommended to set up a directory in a Linux environment with Java installed. If you are a student in Reed Research Group, create a folder in your Cube directory called “MOEA_Diagnostics_Tutorial”. Throughout the training, we will be adding various subdirectories and files to this folder. For this part of the tutorial, you will need the following in your directory:

1. MOEAFramework Demo .jar file which can be found by clicking on “Demo Application” on the far right.
2. The C++ version of the Lake Problem
3. SOWs_Type6.txt (natural inflows read in by the .cpp file)
4. moeaframework.c and moeaframework.h found here

Once you have the .cpp file, the next step is to get it set up to be recognized by MOEAFramework. I have made these changes in this file in my GitHub repo. If you compare with the file in 2, you will see some changes.

First, I have commented out all parts of the code related to Borg, which the problem was originally set up to be optimized with. For this tutorial, we will be optimizing with multiple algorithms and comparing their performance. Secondly, in lines 55-57, I have use the #define directive to declare the number of variables, objectives, and constraints to be constants. Lines 300-309 is where the direct connection to MOEAFramework comes in. MOEA_Init initializes the communication between C++ and MOEAFramework and takes the number of objectives and constraints as arguments. Then we can start reading and evaluating solutions. MOEA_Read_doubles extracts the decision variables and stores them in an array, vars. Line 304 calls the lake_problem function that we would like to evaluate, along with the variables, and constraints. This results in the objs array being filled. Finally, MOEA_Write writes the objectives back into the framework. When all solutions are done being read and written, MOEA_Terminate closes the connection. The important thing here is just to make sure that you are passing the names of the arguments of your lake_problem function correctly.

Next, we must make an executable of the C++ file that Java can read. The makefile is located here and takes lake.cpp, moeaframework.c, moeaframework.h, and some relevant libraries and compiles them into an executable called “lake”. In order to run this file, simply type “make”.

Finally, we must turn our executable into a java class. The relevant java file can be found here. This file can be found in the MOEAFramework documentation, but must be tailored to an external problem. Lines 21-25 import in the relevant tools from MOEAFramework that help to configure and solve the problem. Lines 40, 42, 43, and 48 show the first change from the original file. Here we insert the name of our executable that was generated in the last step, “lake”. In lines 53, 58, and 63, we state the number of decision variables, objectives, and constraints. Finally, in lines 67-75, we create a newSolution ( ) method to specify that our solution should have 6 real-valued decision variables between 0 and 1.

In order to create a class file, simply type:

 

javac -classpath MOEAFramework-2.12-Demo.jar lake.java 

A file called lake.class will be created.

The first part of training is done! Next time, we will set up some scripts, call these executables, and optimize the lake problem with a variety of different algorithms.