Dear all,
A kind contributor to the list was able to provide me with some source
code that showed how an integration of MASON and ECJ might be
implemented. However, this was for earlier versions of MASON and ECJ
and they, especially MASON, have changed in the interim. I've have 95%
of the implementation working, however, I'm seeking suggestions for the
last 5% as the solution still escapes me.
Although I have been able to get MASON and ECJ to integrate well in
"headless" mode, I'm having a great deal of difficulty getting them to
work in "head" mode. I'm not certain if there are Swing threading
issues involved or if I simply don't understand the lifecycle
sufficiently. Basically, I'd like the option of the GP running with or
without graphics. This is set in the parameters with a boolean value,
"graphics". What I would like is to be able to simply update the
GUIState with each change in the SimState. There doesn't seem to be an
straight forward way to do this so I've been "cheating" by adding a
Console and controlling in programmatically. The problem is that it
doesn't always work (the display does not change either because state
is not changing or it is not being updated).
Basically, I have a GPProblem subclass from ECJ that uses a SimState
subclass from MASON as its simulator. The key method for the GPProblem
is the evaluate() method. A scaled down version is presented here.
Basically, there are 3 loops. The number of trials, the number of
games/contests and the number of steps. The simulation ends either with
a time out of 100 steps or with a "kill" at some time step 0 < t < 100.
A new Simulator is created for each game (although it would be nice if
this could be reset to keep windows from popping up and disappearing
constantly...one thing at a time).
I'm completely unsure of my strategy for implementing and using the
GUIState when graphics == true. Line 127 appears to be a hack. The main
observed problem is that when the simulation initializes, the windows
appear but nothing happens. Other times, it runs fine and still other
times it runs fine for *some* of the games. I've also noticed the
positions of the agents (which are set randomly) change from when the
window appears and the simulation starts to run (when the simulation
does indeed run).
Any comments or suggestions would be greatly appreciated.
Cordially,
Steve Butcher
Johns Hopkins University
96 public void evaluate( final EvolutionState state, final
Individual individual, final int threadnum) {
...
107 for ( int i = 0; i < numTrials; i++ ) {
...
113 for ( int j = 0; j < games; j++) {
...
115 Individual[] brains = assignBrains( j,
individual);
116 simulator = new Simulator( state.random[ 0],
brains, state, threadnum, data, stack, this);
117
118 simulator.start(); // is this problematic if
graphics == true?
119
120 SimulatorDisplay display = null;
121 Console console = null;
122
123 if ( graphics) {
124 display = new SimulatorDisplay( simulator);
125 console = new Console( display);
126 console.setVisible( true);
127 try { Thread.sleep( 10000); } catch (
InterruptedException e) {}
128 console.pressPause(); // in order to step
through the simulation using console.pressPlay()
129 display.start();
130 }
131
132 int step = 0;
133 while ( !simulator.finished && step < steps) {
// terminates with kill or 100 steps whichever comes first
...
135 if ( graphics) {
136 console.pressPlay();
137 } else {
138 simulator.schedule.step( simulator);
139 }
140 try { Thread.sleep( 100); } catch (
InterruptedException e) {}
141 step++;
142 }
143
144 if ( graphics) {
145 display.quit();
146 display = null;
147 console = null;
148 }
...
152 }
...
155 }
...
173 }
|
