InstanceLearner.java revision 0a63716ed0e44f7cd32b81a444429318d42d8f08 
1/*
2 * Copyright (C) 2008-2009 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17package android.gesture;
18
19import java.util.ArrayList;
20import java.util.Collections;
21import java.util.Comparator;
22import java.util.TreeMap;
23
24/**
25 * An implementation of an instance-based learner
26 */
27
28class InstanceLearner extends Learner {
29    @Override
30    ArrayList<Prediction> classify(int sequenceType, float[] vector) {
31        ArrayList<Prediction> predictions = new ArrayList<Prediction>();
32        ArrayList<Instance> instances = getInstances();
33        int count = instances.size();
34        TreeMap<String, Double> label2score = new TreeMap<String, Double>();
35        for (int i = 0; i < count; i++) {
36            Instance sample = instances.get(i);
37            if (sample.vector.length != vector.length) {
38                continue;
39            }
40            double distance;
41            if (sequenceType == GestureStore.SEQUENCE_SENSITIVE) {
42                distance = GestureUtilities.cosineDistance(sample.vector, vector);
43            } else {
44                distance = GestureUtilities.squaredEuclideanDistance(sample.vector, vector);
45            }
46            double weight;
47            if (distance == 0) {
48                weight = Double.MAX_VALUE;
49            } else {
50                weight = 1 / distance;
51            }
52            Double score = label2score.get(sample.label);
53            if (score == null || weight > score) {
54                label2score.put(sample.label, weight);
55            }
56        }
57
58        double sum = 0;
59        for (String name : label2score.keySet()) {
60            double score = label2score.get(name);
61            sum += score;
62            predictions.add(new Prediction(name, score));
63        }
64
65        // normalize
66        for (Prediction prediction : predictions) {
67            prediction.score /= sum;
68        }
69
70        Collections.sort(predictions, new Comparator<Prediction>() {
71            public int compare(Prediction object1, Prediction object2) {
72                double score1 = object1.score;
73                double score2 = object2.score;
74                if (score1 > score2) {
75                    return -1;
76                } else if (score1 < score2) {
77                    return 1;
78                } else {
79                    return 0;
80                }
81            }
82        });
83
84        return predictions;
85    }
86}
87