1 | package de.ugoe.cs.quest.usageprofiles;
|
---|
2 |
|
---|
3 | import java.util.LinkedList;
|
---|
4 | import java.util.List;
|
---|
5 | import java.util.Random;
|
---|
6 |
|
---|
7 | import de.ugoe.cs.quest.eventcore.Event;
|
---|
8 | import de.ugoe.cs.quest.eventcore.StringEventType;
|
---|
9 |
|
---|
10 | /**
|
---|
11 | * <p>
|
---|
12 | * This class provides functions to create flattened first-order Markov models from
|
---|
13 | * {@link HighOrderMarkovModel}s and {@link PredictionByPartialMatch} models through state
|
---|
14 | * splitting.
|
---|
15 | * </p>
|
---|
16 | * <p>
|
---|
17 | * If possible, the normal high-order markov model should be used, as the Events may be broken by
|
---|
18 | * the flattener, as, e.g., the information {@link ReplayableEvent}s contain is not preserved.
|
---|
19 | * </p>
|
---|
20 | *
|
---|
21 | * @author Steffen Herbold
|
---|
22 | * @version 1.0
|
---|
23 | */
|
---|
24 | public class ModelFlattener {
|
---|
25 |
|
---|
26 | private static final String EVENT_SEPARATOR = "-=-";
|
---|
27 |
|
---|
28 | Trie<Event> firstOrderTrie;
|
---|
29 |
|
---|
30 | /**
|
---|
31 | * <p>
|
---|
32 | * Takes a {@link HighOrderMarkovModel} and returns a {@link FirstOrderMarkovModel} that
|
---|
33 | * conserves the high-order memory through state splitting.
|
---|
34 | * </p>
|
---|
35 | *
|
---|
36 | * @param model
|
---|
37 | * model that is flattened
|
---|
38 | * @return flattened first-order Markov model
|
---|
39 | */
|
---|
40 | public FirstOrderMarkovModel flattenHighOrderMarkovModel(HighOrderMarkovModel model) {
|
---|
41 | int markovOrder = model.trieOrder - 1;
|
---|
42 | FirstOrderMarkovModel firstOrderModel = new FirstOrderMarkovModel(new Random());
|
---|
43 | firstOrderModel.trieOrder = 2;
|
---|
44 | if (markovOrder == 1) {
|
---|
45 | firstOrderModel.trie = new Trie<Event>(model.trie);
|
---|
46 | firstOrderModel.trieOrder = 2;
|
---|
47 | }
|
---|
48 | else {
|
---|
49 | firstOrderTrie = new Trie<Event>();
|
---|
50 | TrieNode<Event> rootNode = model.trie.find(null);
|
---|
51 | generateFirstOrderTrie(rootNode, new LinkedList<String>(), markovOrder);
|
---|
52 | firstOrderTrie.updateKnownSymbols();
|
---|
53 | firstOrderModel.trie = firstOrderTrie;
|
---|
54 | }
|
---|
55 |
|
---|
56 | return firstOrderModel;
|
---|
57 | }
|
---|
58 |
|
---|
59 | /**
|
---|
60 | * <p>
|
---|
61 | * <b>This method is not available yet and always return null!</b>
|
---|
62 | * </p>
|
---|
63 | * <p>
|
---|
64 | * Takes a {@link PredictionByPartialMatch} model and returns a {@link FirstOrderMarkovModel}
|
---|
65 | * that conserves the high-order memory through state splitting.
|
---|
66 | * </p>
|
---|
67 | *
|
---|
68 | * @param model
|
---|
69 | * model that is flattened
|
---|
70 | * @return flattened first-order Markov model
|
---|
71 | */
|
---|
72 | public FirstOrderMarkovModel flattenPredictionByPartialMatch(PredictionByPartialMatch model) {
|
---|
73 | // TODO implement method
|
---|
74 | return null;
|
---|
75 | }
|
---|
76 |
|
---|
77 | /**
|
---|
78 | * <p>
|
---|
79 | * Converts all nodes of the given depth into first-order node through state-splitting. For each
|
---|
80 | * node at the given depth a new node is created and appropriate transitions will be added.
|
---|
81 | * </p>
|
---|
82 | * <p>
|
---|
83 | * This method traverses through the tree recursively. If the recursion reaches the desired
|
---|
84 | * depth in the tree, node are added.
|
---|
85 | * </p>
|
---|
86 | *
|
---|
87 | * @param currentNode
|
---|
88 | * node whose sub-trie is currently traversed
|
---|
89 | * @param parentIDs
|
---|
90 | * ID strings of the ancestors of the currentNode
|
---|
91 | * @param depth
|
---|
92 | * depth to go - NOT the current depth.
|
---|
93 | */
|
---|
94 | private void generateFirstOrderTrie(TrieNode<Event> currentNode,
|
---|
95 | List<String> parentIDs,
|
---|
96 | int depth)
|
---|
97 | {
|
---|
98 | for (TrieNode<Event> child : currentNode.getChildren()) {
|
---|
99 | String currentId = child.getSymbol().getId();
|
---|
100 | if (depth > 1) {
|
---|
101 | List<String> childParentIDs = new LinkedList<String>(parentIDs);
|
---|
102 | childParentIDs.add(currentId);
|
---|
103 | generateFirstOrderTrie(child, childParentIDs, depth - 1);
|
---|
104 |
|
---|
105 | }
|
---|
106 | else {
|
---|
107 | StringBuilder firstOrderID = new StringBuilder();
|
---|
108 | for (String parentID : parentIDs) {
|
---|
109 | firstOrderID.append(parentID + EVENT_SEPARATOR);
|
---|
110 | }
|
---|
111 | firstOrderID.append(currentId);
|
---|
112 | TrieNode<Event> firstOrderNode =
|
---|
113 | firstOrderTrie.getChildCreate(new Event(new StringEventType(firstOrderID
|
---|
114 | .toString())));
|
---|
115 | firstOrderNode.setCount(child.getCount());
|
---|
116 | for (TrieNode<Event> transitionChild : child.getChildren()) {
|
---|
117 | StringBuilder transitionID = new StringBuilder();
|
---|
118 | for (String parentID : parentIDs.subList(1, parentIDs.size())) {
|
---|
119 | transitionID.append(parentID + EVENT_SEPARATOR);
|
---|
120 | }
|
---|
121 | transitionID.append(currentId + EVENT_SEPARATOR);
|
---|
122 | transitionID.append(transitionChild.getSymbol().getId());
|
---|
123 | TrieNode<Event> firstOrderTransitionChild =
|
---|
124 | firstOrderNode.getChildCreate(new Event(new StringEventType(transitionID
|
---|
125 | .toString())));
|
---|
126 | firstOrderTransitionChild.setCount(transitionChild.getCount());
|
---|
127 | }
|
---|
128 | }
|
---|
129 | }
|
---|
130 | }
|
---|
131 | }
|
---|