1 | package de.uka.ipd.sdq.pipesandfilters.framework.filters; |
2 | |
3 | import java.util.List; |
4 | |
5 | /** |
6 | * Implements the "Marginal Confidence Rule" (MCR) for filtering the warm-up |
7 | * period of a steady state simulation. |
8 | * |
9 | * The Filter is still experimental! |
10 | * |
11 | * @author Philipp Merkle |
12 | * |
13 | */ |
14 | public class MCRWarmUpFilter extends Filter { |
15 | |
16 | private int minIndex = 0; |
17 | |
18 | public List<Double> filter(List<Double> samples) { |
19 | |
20 | if (samples.size() <= 150){ |
21 | System.out.println("MCRWarmUpFilter Warning: Too few samples to get a meaningful result."); |
22 | } |
23 | |
24 | int truncatedSamplesSize = samples.size(); |
25 | double truncatedSamplesSum = 0; |
26 | for (Double d : samples) { |
27 | truncatedSamplesSum += d; |
28 | } |
29 | |
30 | double minValue = Double.MAX_VALUE; |
31 | |
32 | for (int i = 0; i < samples.size() - 1; i++) { |
33 | int remaining = samples.size() - i; |
34 | double factor = 1 / Math.pow(remaining, 3.0); |
35 | |
36 | double truncatedSampleMean = truncatedSamplesSum |
37 | / truncatedSamplesSize; |
38 | double sum = 0; |
39 | for (int j = i + 1; j < samples.size(); j++) { |
40 | sum += Math.pow(samples.get(j) - truncatedSampleMean, 2.0); |
41 | } |
42 | double d = factor * sum; |
43 | |
44 | if (d < minValue) { |
45 | // System.out.println(i + ": " + d); |
46 | minIndex = i; |
47 | minValue = d; |
48 | } |
49 | |
50 | truncatedSamplesSize--; |
51 | truncatedSamplesSum -= samples.get(0); |
52 | } |
53 | |
54 | if (minIndex > samples.size() / 3){ |
55 | //TODO: Kriterium nachschauen und logger |
56 | System.out.println("MCRWarmUpFilter Warning: Truncation point is in the last two thirds of the samples, so the confidence in this result is low."); |
57 | } |
58 | |
59 | // TODO Create new list? |
60 | return samples.subList(minIndex, samples.size() - 1); |
61 | } |
62 | |
63 | public int getTruncationIndex() { |
64 | return minIndex; |
65 | } |
66 | } |