001package es.ucm.fdi.gaia.jcolibri.evaluation.evaluators; 002 003import java.io.BufferedReader; 004import java.io.BufferedWriter; 005import java.io.FileReader; 006import java.io.FileWriter; 007import java.util.ArrayList; 008import java.util.Collection; 009import java.util.Date; 010import java.util.Iterator; 011import java.util.List; 012 013import org.apache.logging.log4j.LogManager; 014 015import es.ucm.fdi.gaia.jcolibri.casebase.CachedLinealCaseBase; 016import es.ucm.fdi.gaia.jcolibri.cbraplications.StandardCBRApplication; 017import es.ucm.fdi.gaia.jcolibri.cbrcore.CBRCase; 018import es.ucm.fdi.gaia.jcolibri.cbrcore.CBRCaseBase; 019import es.ucm.fdi.gaia.jcolibri.evaluation.EvaluationReport; 020import es.ucm.fdi.gaia.jcolibri.evaluation.Evaluator; 021import es.ucm.fdi.gaia.jcolibri.exception.ExecutionException; 022import es.ucm.fdi.gaia.jcolibri.util.FileIO; 023import es.ucm.fdi.gaia.jcolibri.util.ProgressController; 024 025/** 026 * This method splits the case base in two sets: one used for testing where each 027 * case is used as query, and another that acts as normal case base. 028 * <br> 029 * This method is different of the other evaluators beacuse the split is stored in a file that can be used in following evaluations. 030 * This way, the same set is used as queries for each evaluation. 031 * <br> 032 * The generateSplit() method does the initial random split and saves the query set in a file. 033 * Later, the HoldOutfromFile() method uses that file to load the queries set and perform the evaluation. 034 * 035 * @author Juan A. Recio GarcĂa & Lisa Cummins 036 * @version 2.0 037 */ 038 039public class SameSplitEvaluator extends Evaluator { 040 041 protected StandardCBRApplication app; 042 043 public void init(StandardCBRApplication cbrApp) { 044 045 report = new EvaluationReport(); 046 app = cbrApp; 047 try { 048 app.configure(); 049 } catch (ExecutionException e) { 050 LogManager.getLogger().error(e); 051 } 052 } 053 054 public void generateSplit(int testPercent, String filename) 055 { 056 //Run the precycle to load the case base 057 LogManager.getLogger().info("Running precycle()"); 058 CBRCaseBase caseBase = null; 059 try { 060 caseBase = app.preCycle(); 061 } catch (ExecutionException e) { 062 LogManager.getLogger().error(e); 063 } 064 065 if (!(caseBase instanceof CachedLinealCaseBase)) 066 LogManager.getLogger().warn( 067 "Evaluation should be executed using a cached case base"); 068 069 ArrayList<CBRCase> originalCases = new ArrayList<CBRCase>( 070 caseBase.getCases()); 071 072 ArrayList<CBRCase> querySet = new ArrayList<CBRCase>(); 073 ArrayList<CBRCase> caseBaseSet = new ArrayList<CBRCase>(); 074 075 // Split the case base 076 splitCaseBase(originalCases, querySet, caseBaseSet, testPercent); 077 078 save(querySet, filename); 079 } 080 081 protected void save(Collection<CBRCase> queries, String filename) 082 { 083 try { 084 BufferedWriter br = null; 085 br = new BufferedWriter(new FileWriter(filename)); 086 //if (br == null) 087 // throw new Exception("Error opening file for writing: "+ filename); 088 089 for (CBRCase _case : queries) { 090 br.write(_case.getID().toString()); 091 br.newLine(); 092 } 093 br.close(); 094 } catch (Exception e) { 095 LogManager.getLogger().error(e); 096 } 097 098 099 } 100 101 public void HoldOutfromFile(String file) { 102 try { 103 // Obtain the time 104 long t = (new Date()).getTime(); 105 int numberOfCycles = 0; 106 107 // Run the precycle to load the case base 108 LogManager.getLogger().info("Running precycle()"); 109 CBRCaseBase caseBase = app.preCycle(); 110 111 if (!(caseBase instanceof CachedLinealCaseBase)) 112 LogManager.getLogger().warn( 113 "Evaluation should be executed using a cached case base"); 114 115 ArrayList<CBRCase> originalCases = new ArrayList<CBRCase>(caseBase.getCases()); 116 ArrayList<CBRCase> querySet = new ArrayList<CBRCase>(); 117 ArrayList<CBRCase> caseBaseSet = new ArrayList<CBRCase>(); 118 119 // Split the case base 120 splitCaseBaseFromFile(originalCases, querySet, caseBaseSet, file); 121 122 int totalSteps = querySet.size(); 123 ProgressController.init(getClass(), 124 "Same Split - Hold Out Evaluation", totalSteps); 125 126 127 // Clear the caseBase 128 caseBase.forgetCases(originalCases); 129 130 // Set the cases that acts as case base in this repetition 131 caseBase.learnCases(caseBaseSet); 132 133 // Run cycle for each case in querySet 134 for (CBRCase c : querySet) { 135 // Run the cycle 136 LogManager.getLogger().info( 137 "Running cycle() " + numberOfCycles); 138 app.cycle(c); 139 140 ProgressController.step(getClass()); 141 numberOfCycles++; 142 } 143 144 ProgressController.finish(getClass()); 145 146 // Revert case base to original state 147 caseBase.forgetCases(originalCases); 148 caseBase.learnCases(originalCases); 149 150 // Run the poscycle to finish the application 151 LogManager.getLogger().info("Running postcycle()"); 152 app.postCycle(); 153 154 t = (new Date()).getTime() - t; 155 156 // Obtain and complete the evaluation result 157 report.setTotalTime(t); 158 report.setNumberOfCycles(numberOfCycles); 159 160 } catch (Exception e) { 161 LogManager.getLogger().error(e); 162 } 163 164 } 165 166 /** 167 * Splits the case base in two sets: queries and case base 168 * 169 * @param wholeCaseBase 170 * Complete original case base 171 * @param querySet 172 * Output param where queries are stored 173 * @param casebaseSet 174 * Output param where case base is stored 175 * @param testPercent 176 * Percentage of cases used as queries 177 */ 178 protected void splitCaseBase(Collection<CBRCase> wholeCaseBase, 179 List<CBRCase> querySet, List<CBRCase> casebaseSet, int testPercent) { 180 querySet.clear(); 181 casebaseSet.clear(); 182 183 int querySetSize = (wholeCaseBase.size() * testPercent) / 100; 184 casebaseSet.addAll(wholeCaseBase); 185 186 for (int i = 0; i < querySetSize; i++) { 187 int random = (int) (Math.random() * casebaseSet.size()); 188 CBRCase _case = casebaseSet.get(random); 189 casebaseSet.remove(random); 190 querySet.add(_case); 191 } 192 } 193 194 protected void splitCaseBaseFromFile(Collection<CBRCase> wholeCaseBase, 195 List<CBRCase> querySet, List<CBRCase> casebaseSet, String filename) 196 { 197 querySet.clear(); 198 casebaseSet.clear(); 199 200 casebaseSet.addAll(wholeCaseBase); 201 202 try { 203 BufferedReader br = null; 204 br = new BufferedReader(new FileReader(FileIO.findFile(filename).getFile())); 205 //if (br == null) 206 // throw new Exception("Error opening file: " + filename); 207 208 String line = ""; 209 while ((line = br.readLine()) != null) { 210 CBRCase c = null; 211 int pos=0; 212 boolean found = false; 213 for(Iterator<CBRCase> iter = casebaseSet.iterator(); iter.hasNext() && (!found); ) 214 { 215 c = iter.next(); 216 if(c.getID().toString().equals(line)) 217 found = true; 218 else 219 pos++; 220 } 221 if(c==null) 222 { 223 LogManager.getLogger().error("Case "+line+" not found into case base"); 224 continue; 225 } 226 227 casebaseSet.remove(pos); 228 querySet.add(c); 229 } 230 br.close(); 231 } catch (Exception e) { 232 LogManager.getLogger().error(e); 233 } 234 } 235 236}