Changeset 2130 for trunk

Timestamp:

08/24/16 10:06:39 (8 years ago)

Author:

pharms

Message:

• improved partial trie training to make it even more partial by not training instances of tasks that occur less often then the currently detected sequence also in other sessions

File:

• trunk/autoquest-core-tasktrees/src/main/java/de/ugoe/cs/autoquest/tasktrees/temporalrelation/TaskInstanceTrie.java (modified) (4 diffs)

trunk/autoquest-core-tasktrees/src/main/java/de/ugoe/cs/autoquest/tasktrees/temporalrelation/TaskInstanceTrie.java

@@ -83 +83 @@
         
         Map<ITask, Counter> instanceCountMap = new HashMap<ITask, Counter>();
-        
         Console.traceln(Level.FINEST, "preparing training");
         int noOfTaskInstances = 0;
@@ -106 +105 @@
         
         Counter processedTaskInstances = new Counter();
-        int counterRecheckAt = noOfTaskInstances / 10; // recheck the maximum count after each
+        Counter currentSequenceMaxCount = new Counter();
+        int counterRecheckAt = noOfTaskInstances / 20; // recheck the maximum count after each
                                                        // 10% of processed task instances
         for (IUserSession session : userSessions) {
-            train(session, maxOrder, instanceCountMap, processedTaskInstances, counterRecheckAt);
+            train(session, maxOrder, instanceCountMap, processedTaskInstances,
+                  currentSequenceMaxCount, counterRecheckAt);
         }      
         
@@ -148 +149 @@
                        Map<ITask, Counter> taskInstanceCountMap,
                        Counter             processedTaskInstances,
+                       Counter             currentSequenceMaxCount,
                        int                 counterRecheckAt)
     {
         List<ITask> subsequence = new LinkedList<ITask>();
         
-        int sequenceMaxCount = 0;
-        
         for (ITaskInstance currentTaskInstance : userSession) {
             
@@ -159 +159 @@
             
             if (processedTaskInstances.count >= counterRecheckAt) {
-                sequenceMaxCount = getCurrentSequenceMaxCount();
+                currentSequenceMaxCount.count = getCurrentSequenceMaxCount();
                 processedTaskInstances.count = 0;
             }
             
-            if (occurrenceCount < sequenceMaxCount) {
+            if (occurrenceCount < currentSequenceMaxCount.count) {
                 // this task instance does not need to be considered, as it occurs not often enough
                 // to be part of a sequence, that occurs most often. Therefore train all remaining