[SYSTEMDS-3777] Additional adasyn real data tests

This generalizes the adasyn test for additional real data set. On the
titantic dataset, adasyn gives a 1.6% improvement of test accuracy
(for a basic logreg model, 0.781 -> 0.797).

Author: mboehm7

src/main/java/org/apache/sysds/parser/DMLTranslator.java

@@ -1610,7 +1610,7 @@ else if (source instanceof ExpressionList){
 				return currBuiltinOp;
 			}
 			else{
-				throw new ParseException("Unhandled instance of source type: " + source.getClass());
+				throw new ParseException("Unhandled instance of source type: " + source);
 			}
 		} 
 		catch(ParseException e ){

src/test/java/org/apache/sysds/test/functions/builtin/part1/BuiltinAdasynRealDataTest.java

@@ -35,6 +35,8 @@ public class BuiltinAdasynRealDataTest extends AutomatedTestBase {
 
 	private final static String DIABETES_DATA = DATASET_DIR + "diabetes/diabetes.csv";
 	private final static String DIABETES_TFSPEC = DATASET_DIR + "diabetes/tfspec.json";
+	private final static String TITANIC_DATA = DATASET_DIR + "titanic/titanic.csv";
+	private final static String TITANIC_TFSPEC = DATASET_DIR + "titanic/tfspec.json";
 
 	@Override
 	public void setUp() {
@@ -56,15 +58,30 @@ public void testDiabetesAdasynK6() {
 		runAdasynTest(DIABETES_DATA, DIABETES_TFSPEC, true, 0.787, 6, ExecType.CP);
 	}
 
+	@Test
+	public void testTitanicNoAdasyn() {
+		runAdasynTest(TITANIC_DATA, TITANIC_TFSPEC, false, 0.781, -1, ExecType.CP);
+	}
+	
+	@Test
+	public void testTitanicAdasynK4() {
+		runAdasynTest(TITANIC_DATA, TITANIC_TFSPEC, true, 0.797, 4, ExecType.CP);
+	}
+	
+	@Test
+	public void testTitanicAdasynK5() {
+		runAdasynTest(TITANIC_DATA, TITANIC_TFSPEC, true, 0.797, 5, ExecType.CP);
+	}
+	
 	private void runAdasynTest(String data, String tfspec, boolean adasyn, double minAcc, int k, ExecType instType) {
 		Types.ExecMode platformOld = setExecMode(instType);
 		try {
 			loadTestConfiguration(getTestConfiguration(TEST_NAME));
 
 			String HOME = SCRIPT_DIR + TEST_DIR;
 			fullDMLScriptName = HOME + TEST_NAME + ".dml";
-			programArgs = new String[] {"-stats",
-				"-args", data, String.valueOf(adasyn), String.valueOf(k), output("R")};
+			programArgs = new String[] {"-stats", "-args",
+				data, tfspec, String.valueOf(adasyn), String.valueOf(k), output("R")};
 
 			runTest(true, false, null, -1);
 

src/test/resources/datasets/diabetes/tfspec.json

@@ -0,0 +1 @@
+ 

src/test/scripts/functions/builtin/adasynRealData.dml

@@ -20,22 +20,36 @@
 #-------------------------------------------------------------
 
 
-M = read($1, data_type="matrix", format="csv", header=TRUE);
-Y = M[, ncol(M)] + 1
-X = M[, 1:ncol(M)-1]
-upsample = as.logical($2)
+M = read($1, data_type="frame", format="csv", header=TRUE,
+        naStrings= ["NA", "null","  ","NaN", "nan", "", " ", "_nan_", "inf", "?", "NAN", "99999", "99999.00"]);
+Y = as.matrix(M[, ncol(M)]) + 1
+F = M[, 1:ncol(M)-1]
+tfspec = read($2, data_type="scalar", value_type="string")
+upsample = as.logical($3)
+
+if( tfspec != " " ) {
+  F = M[, 1:ncol(M)] # FIXME
+  [X,meta] = transformencode(target=F, spec=tfspec);
+  X = X[,1:ncol(X)-1];
+  X = imputeByMode(X);
+}
+else {
+  X = as.matrix(F);
+}
+
+[X,C,S] = scale(X=X, scale=TRUE, center=TRUE);
 
 [Xtrain, Xtest, Ytrain, Ytest] = split(X=X, Y=Y, f=0.7, seed=3);
 
 if( upsample ) {
   # oversampling all classes other than majority
-  [Xtrain,Ytrain] = adasyn(X=Xtrain, Y=Ytrain, k=$3, seed=7);
+  [Xtrain,Ytrain] = adasyn(X=Xtrain, Y=Ytrain, k=$4, seed=7);
 }
 
 B = multiLogReg(X=Xtrain, Y=Ytrain, icpt=2);
 [P,yhat,acc] = multiLogRegPredict(X=Xtest, Y=Ytest, B=B);
 print("accuracy: "+acc)
 
 R = as.matrix(acc/100);
-write(R, $4);
+write(R, $5);