RUL-84, RUL-85: Regression and survival contrast sets fixed (results agree with the original ones).

Author: agudys

adaa.analytics.rules/build.gradle

@@ -27,7 +27,7 @@ codeQuality {
 }
 
 sourceCompatibility = 1.8
-version = '1.7.18'
+version = '1.7.19'
 
 
 jar {

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/ContrastSurvivalFinder.java

@@ -34,7 +34,7 @@ public double calculate(ExampleSet dataset, ContingencyTable ct) {
             }
 
             Covering cov = (Covering)ct;
-            Set<Integer> examples = new HashSet<>();
+            Set<Integer> examples = new IntegerBitSet(dataset.size());
             examples.addAll(cov.positives);
             examples.addAll(cov.negatives);
             KaplanMeierEstimator entireEstimator = new KaplanMeierEstimator(dataset, examples);
@@ -129,6 +129,7 @@ public void postprocess(
         notifyRuleReady(rule);
     }
 
+    @Override
     protected boolean checkCandidate(
             ExampleSet dataset,
             Rule rule,
@@ -146,7 +147,6 @@ protected boolean checkCandidate(
             Rule newRule = (Rule) rule.clone();
             newRule.setPremise(newPremise);
 
-
             Covering cov = new Covering();
             newRule.covers(dataset, cov, cov.positives, cov.negatives);
 
@@ -166,7 +166,7 @@ protected boolean checkCandidate(
                 }
             }
 
-            if (checkCoverage(cov.weighted_p, cov.weighted_n, new_p, new_n, dataset.size(), 0, uncovered.size(), rule.getRuleOrderNum())) {
+            if (checkCoverage(cov.weighted_p, cov.weighted_n, new_p, new_n, cov.weighted_P, cov.weighted_N, uncovered.size(), rule.getRuleOrderNum())) {
 
                 double quality = params.getInductionMeasure().calculate(dataset, cov);
 
@@ -178,12 +178,14 @@ protected boolean checkCandidate(
                 if (quality > currentBest.quality ||
                         (quality == currentBest.quality && (new_p > currentBest.covered || currentBest.opposite))) {
 
-                    Logger.log("\t\tCurrent best: " + candidate + " (p=" + cov.weighted_p +
-                            ", new_p=" + (double) new_p +
+                    /*
+                    Logger.log("\t\tCurrent best: " + candidate +
+                            " (p=" + cov.weighted_p + ", new_p=" + (double) new_p +
+                            ", n=" + cov.weighted_n + ", new_n=" + (double) new_n +
                             ", P=" + cov.weighted_P +
                             ", mean_y=" + cov.mean_y + ", mean_y2=" + cov.mean_y2 + ", stddev_y=" + cov.stddev_y +
                             ", quality=" + quality + "\n", Level.FINEST);
-
+                    */
                     currentBest.quality = quality;
                     currentBest.condition = candidate;
                     currentBest.covered = new_p;

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/RegressionFinder.java

@@ -340,7 +340,6 @@ protected ElementaryCondition induceCondition(
 					Double [] keys = values2ids.keySet().toArray(new Double[values2ids.size()]);
 					Logger.log("\tChecking attribute " + attr.getName() + ", unique values: " + keys.length + "\n", Level.FINEST);
 
-
 					// check all possible midpoints
 					for (int keyId = 0; keyId < keys.length - 1; ++keyId) {
 						double key = keys[keyId];