public static synchronized TetrisWorldResponse Execute() {
String theRequest =
AbstractMessage.makeMessage(
MessageUser.kEnv.id(),
MessageUser.kBenchmark.id(),
EnvMessageType.kEnvCustom.id(),
MessageValueType.kString.id(),
"GETTETRLAISWORLD");
String responseMessage = RLGlue.RL_env_message(theRequest);
TetrisWorldResponse theResponse;
try {
theResponse = new TetrisWorldResponse(responseMessage);
} catch (NotAnRLVizMessageException ex) {
System.out.println("Not a valid RL Viz Message in Tetrlais World Response" + ex);
return null;
}
return theResponse;
}
public void runExperiment() {
System.out.println("\n\nExperiment starting up!");
String taskSpec = RLGlue.RL_init();
System.out.println("RL_init called, the environment sent task spec: " + taskSpec);
System.out.println("\n\n----------Sending some sample messages----------");
/* Talk to the agent and environment a bit... */
String responseMessage = RLGlue.RL_agent_message("what is your name?");
System.out.println("Agent responded to \"what is your name?\" with: " + responseMessage);
responseMessage = RLGlue.RL_agent_message("If at first you don't succeed; call it version 1.0");
System.out.println(
"Agent responded to \"If at first you don't succeed; call it version 1.0 \" with: "
+ responseMessage
+ "\n");
responseMessage = RLGlue.RL_env_message("what is your name?");
System.out.println("Environment responded to \"what is your name?\" with: " + responseMessage);
responseMessage = RLGlue.RL_env_message("If at first you don't succeed; call it version 1.0");
System.out.println(
"Environment responded to \"If at first you don't succeed; call it version 1.0 \" with: "
+ responseMessage);
System.out.println("\n\n----------Running a few episodes----------");
for (int i = 0; i < 1; i++) runEpisode(20000);
runEpisode(1);
/* Remember that stepLimit of 0 means there is no limit at all! */
// runEpisode(0);
RLGlue.RL_cleanup();
System.out.println("\n\n----------Stepping through an episode----------");
// taskSpec = RLGlue.RL_init();
// for(int i = 0; i < 2; i++) {
// /*We could also start over and do another experiment */
// /*We could run one step at a time instead of one episode at a time */
// /*Start the episode */
// Observation_action startResponse = RLGlue.RL_start();
// int firstObservation = startResponse.o.intArray[0];
// int firstAction = startResponse.a.intArray[0];
// System.out.println("First observation and action were: " +
// firstObservation + " and: " + firstAction);
//
// /*Run one step */
// Reward_observation_action_terminal stepResponse = RLGlue.RL_step();
//
// /*Run until the episode ends*/
// while (stepResponse.terminal != 1) {
// stepResponse = RLGlue.RL_step();
// if (stepResponse.terminal != 1) {
// /*Could optionally print state,action pairs */
// System.out.println(
// ("(state,action)=(" +
// stepResponse.o.intArray[0] +
// "," +
// stepResponse.a.intArray[0] + ")"));
// }
// }
//
// System.out.println("\n\n----------Summary----------");
//
// int totalSteps = RLGlue.RL_num_steps();
// double totalReward = RLGlue.RL_return();
// System.out.println("It ran for " + totalSteps +
// " steps, total reward was: " + totalReward);
// }
// RLGlue.RL_cleanup();
}
