source: java/ecj/cecj/eval/TDLImprovingEvaluator.java @ 193

Last change on this file since 193 was 193, checked in by Maciej Komosinski, 10 years ago

Set svn:eol-style native for all textual files

  • Property svn:eol-style set to native
File size: 3.0 KB
Line 
1/*
2  Copyright 2009 by Marcin Szubert
3  Licensed under the Academic Free License version 3.0
4 */
5
6package cecj.eval;
7
8import ec.EvolutionState;
9import ec.Individual;
10import ec.util.Parameter;
11
12/**
13 * Wrapper for <code>CoevolutionaryEvaluator</code> performing an additional learning phase before
14 * actual evaluation.
15 *
16 * The only role of this class is improving each individual of the population by running a specific
17 * temporal difference learning (TDL) algorithm before the evaluation. Since the exact
18 * implementation of this algorithm depends on the problem, evaluator delegates the learning task to
19 * the provided <code>TDLImprover</code> interface realization. At the beginning of evolutionary
20 * process individuals may require some preparation for running TDL. For this reason, appropriate
21 * interface methods are invoked before the first evaluation. Clearly, not every problem can be
22 * approached by reinforcement learning paradigm so this class has also a limited scope of
23 * applicability.
24 *
25 * Note that this evaluator realizes the Coevolutionary Reinforcement Learning idea.
26 *
27 * @author Marcin Szubert
28 * @see TDLImprover
29 *
30 */
31public class TDLImprovingEvaluator extends CoevolutionaryEvaluator {
32
33        private static final String P_INNER_EVALUATOR = "inner-evaluator";
34        private static final String P_TDL_IMPROVER = "tdl-improver";
35        private static final String P_TDL_FREQUENCY = "tdl-frequency";
36
37        private CoevolutionaryEvaluator innerEvaluator;
38        private TDLImprover temporalDifferenceImprover;
39        private boolean firstEvaluation = true;
40        private int tdlFrequency;
41
42        @Override
43        public void setup(EvolutionState state, Parameter base) {
44                super.setup(state, base);
45
46                Parameter innerEvaluatorParam = base.push(P_INNER_EVALUATOR);
47                innerEvaluator = (CoevolutionaryEvaluator) (state.parameters.getInstanceForParameter(
48                                innerEvaluatorParam, null, CoevolutionaryEvaluator.class));
49                innerEvaluator.setup(state, innerEvaluatorParam);
50
51                Parameter tdlImproverParam = base.push(P_TDL_IMPROVER);
52                temporalDifferenceImprover = (TDLImprover) (state.parameters.getInstanceForParameter(
53                                tdlImproverParam, null, TDLImprover.class));
54                temporalDifferenceImprover.setup(state, tdlImproverParam);
55
56                Parameter tdlImprovingFrequency = base.push(P_TDL_FREQUENCY);
57                tdlFrequency = state.parameters.getIntWithDefault(tdlImprovingFrequency, null, 1);
58        }
59
60        @Override
61        public void evaluatePopulation(EvolutionState state) {
62                if (firstEvaluation) {
63                        for (int subpop = 0; subpop < numSubpopulations; subpop++) {
64                                Individual[] inds = state.population.subpops[subpop].individuals;
65                                for (Individual ind : inds) {
66                                        temporalDifferenceImprover.prepareForImproving(state, ind);
67                                }
68                        }
69                        firstEvaluation = false;
70                }
71
72                if ((state.generation % tdlFrequency) == 0) {
73                        for (int subpop = 0; subpop < numSubpopulations; subpop++) {
74                                Individual[] inds = state.population.subpops[subpop].individuals;
75                                for (Individual ind : inds) {
76                                        temporalDifferenceImprover.improve(state, ind);
77                                }
78                        }
79                }
80
81                innerEvaluator.evaluatePopulation(state);
82        }
83}
Note: See TracBrowser for help on using the repository browser.