apply code improvements suggested by intelliJ
This commit is contained in:
@@ -39,6 +39,7 @@ public class RLController<A extends Enum> implements LearningListener {
|
||||
protected int prevDelay;
|
||||
protected volatile boolean printNextEpisode;
|
||||
|
||||
@SafeVarargs
|
||||
public RLController(Environment<A> env, Method method, A... actions) {
|
||||
setEnvironment(env);
|
||||
setMethod(method);
|
||||
@@ -102,9 +103,7 @@ public class RLController<A extends Enum> implements LearningListener {
|
||||
if(learning.isCurrentlyLearning()){
|
||||
((EpisodicLearning<A>) learning).learnMoreEpisodes(nrOfEpisodes);
|
||||
}else{
|
||||
new Thread(() -> {
|
||||
((EpisodicLearning<A>) learning).learn(nrOfEpisodes);
|
||||
}).start();
|
||||
new Thread(() -> ((EpisodicLearning<A>) learning).learn(nrOfEpisodes)).start();
|
||||
}
|
||||
} else {
|
||||
throw new RuntimeException("Triggering onLearnMoreEpisodes on non-episodic learning!");
|
||||
@@ -179,7 +178,7 @@ public class RLController<A extends Enum> implements LearningListener {
|
||||
public void onEpisodeEnd(List<Double> rewardHistory) {
|
||||
latestRewardsHistory = rewardHistory;
|
||||
if(printNextEpisode) {
|
||||
System.out.println("Episode " + ((EpisodicLearning<A>) learning).getCurrentEpisode() + " Latest Reward: " + rewardHistory.get(rewardHistory.size() - 1));
|
||||
System.out.println("Episode " + learning.getCurrentEpisode() + " Latest Reward: " + rewardHistory.get(rewardHistory.size() - 1));
|
||||
System.out.println("Eps/sec: " + ((EpisodicLearning<A>) learning).getEpisodePerSecond());
|
||||
printNextEpisode = false;
|
||||
}
|
||||
|
||||
@@ -13,6 +13,7 @@ import java.util.List;
|
||||
public class RLControllerGUI<A extends Enum> extends RLController<A> implements ViewListener {
|
||||
private LearningView learningView;
|
||||
|
||||
@SafeVarargs
|
||||
public RLControllerGUI(Environment<A> env, Method method, A... actions) {
|
||||
super(env, method, actions);
|
||||
}
|
||||
@@ -102,7 +103,7 @@ public class RLControllerGUI<A extends Enum> extends RLController<A> implements
|
||||
@Override
|
||||
public void onLearningEnd() {
|
||||
super.onLearningEnd();
|
||||
onSaveState(method.toString() + System.currentTimeMillis() / 1000 + (learning instanceof EpisodicLearning ? "e" + ((EpisodicLearning) learning).getCurrentEpisode() : ""));
|
||||
onSaveState(method.toString() + System.currentTimeMillis() / 1000 + (learning instanceof EpisodicLearning ? "e" + learning.getCurrentEpisode() : ""));
|
||||
SwingUtilities.invokeLater(() -> learningView.updateRewardGraph(latestRewardsHistory));
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user