[aehrc#104] Implement regression tests (aehrc#109)

* Fixed the issue with boostrap sample, but using the actual sample indexes (with repeats) rather than the distinct set of indexes (aehrc#101)

* Tech/104/0.2 (aehrc#105)

* Added regression test cases generation scripts

* Added regression cases and unit test

* Moved execution of regression tests to 'regression-test' profile

* Enabled all regression test cases

* Update command line for regression tests

* Fixed chr22 regression cmd line

* Clean up: removed all splitting classes

* Refactored regression tests to use Parameterized

* Update the regression cases generation scripts

.gitignore

@@ -58,3 +58,4 @@ build
 dist
 _build
 spark-warehouse
+.*.crc

dev/test-gen-regression-cases.sh

@@ -32,7 +32,7 @@ PREFIX="CNAE-9"
 "${FWDIR}/bin/variant-spark" --spark --master local[2] -- importance -if "${DATA_DIR}/${PREFIX}-wide.csv" -ff "${DATA_DIR}/${PREFIX}-labels.csv" \
          -fc "${RESP}" \
          -on 100 -of "${OUTPUT_DIR}/${PREFIX}-imp_${RESP}.csv" \
-         -ivo 10 \
+         -io "{\"defVariableType\":\"ORDINAL(10)\"}" \
          -it csv -v -ro -rn 100 -rbs 50 -sp 4 -sr  17
 
 #
@@ -49,7 +49,7 @@ for CASE in ${FWDIR}/src/test/data/synth/*-meta.txt; do
 	"${FWDIR}/bin/variant-spark" --spark --master local[2] -- importance -if "${DATA_DIR}/${PREFIX}-wide.csv" -ff "${DATA_DIR}/${PREFIX}-labels.csv" \
 	 -fc "${RESP}" \
 	 -on 100 -of "${OUTPUT_DIR}/${PREFIX}-imp_${RESP}.csv" \
-	 -ivo ${IVO} \
+	 -io "{\"defVariableType\":\"ORDINAL(${IVO})\"}" \
  	 -it csv -v -ro -rn 100 -rbs 50 -sp 4 -sr 17
 	done
 done

src/main/java/au/csiro/variantspark/algo/split/JOrderedFastIndexedSplitter.java

@@ -46,10 +46,12 @@ protected SplitInfo doFindSplit(int[] splitIndices) {
 			if (!thisAggregator.isEmpty()) {
 				// only consider value that appeared at least once in the split
 				impurityCalc.update(thisAggregator);
-				double thisImpurity = impurityCalc.getValue(leftRightImpurity);
-				if (thisImpurity < minImpurity) {
-					result = new SplitInfo(sp, thisImpurity, leftRightImpurity.left(), leftRightImpurity.right());
-					minImpurity = thisImpurity;
+				if (impurityCalc.hasProperSplit()) {
+					double thisImpurity = impurityCalc.getValue(leftRightImpurity);
+					if (thisImpurity < minImpurity) {
+						result = new SplitInfo(sp, thisImpurity, leftRightImpurity.left(), leftRightImpurity.right());
+						minImpurity = thisImpurity;
+					}
 				}
 			}
 		}

src/main/java/au/csiro/variantspark/algo/split/JOrderedIndexedSplitter.java

@@ -49,10 +49,12 @@ public SplitInfo doFindSplit(int[] splitIndices) {
 					impurityCalc.update(i);
 				} 
 			}
-			double g = impurityCalc.getValue(leftRightImpurity);
-			if (g < minImpurity ) {
-				result = new SplitInfo(sp, g, leftRightImpurity.left(), leftRightImpurity.right());
-				minImpurity = g;
+			if (impurityCalc.hasProperSplit()) {
+				double g = impurityCalc.getValue(leftRightImpurity);
+				if (g < minImpurity ) {
+					result = new SplitInfo(sp, g, leftRightImpurity.left(), leftRightImpurity.right());
+					minImpurity = g;
+				}
 			}
 		}
 		return result;	

src/main/scala/au/csiro/variantspark/algo/Split.scala

@@ -28,6 +28,16 @@ trait IndexedSplitAggregator {
 	  left.add(agg)
 	  right.sub(agg)	  
 	}
+
+	/** 
+	 *  Is this a valid split that is one that does not put 
+	 *  all elements to one side
+	 */
+	def hasProperSplit:Boolean = !left.isEmpty() && !right.isEmpty()
+
+	/**
+	 * Get split impurity value
+	 */
 	def getValue(outSplitImp:SplitImpurity):Double = {
 	  left.splitValue(right, outSplitImp)
 	}