Fix NPE when scoring CoxPH mojo from h2o 3.32.x.x

Author: Mathanraj-Sharma

h2o-genmodel/src/main/java/hex/genmodel/algos/coxph/CoxPHMojoModel.java

@@ -78,13 +78,27 @@ public double[] score0(double[] row, double[] predictions) {
   @Override
   public double[] score0(double[] row, double offset, double[] predictions) {
     int[] enumOffset = null;
-    if (_interaction_targets != null) {
-      enumOffset = evaluateInteractions(row);
+    
+    if (_nums == -1) {
+      predictions[0] = forCategories(row) + forOtherColumns(row) - forStrata(row) + offset;
+    } else {
+      if (_interaction_targets != null) {
+        enumOffset = evaluateInteractions(row);
+      }
+      predictions[0] = forCategories(row) + forOtherColumns(row, enumOffset) - forStrata(row) + offset; 
     }
-    predictions[0] = forCategories(row) + forOtherColumns(row, enumOffset) - forStrata(row) + offset;
     return predictions;
   }
 
+  private double forOtherColumns(double[] row) {
+    double result = 0.0;
+    int catOffsetDiff = _cat_offsets[_cats] - _cats;
+    for(int i = _cats ; i + catOffsetDiff < _coef.length; i++) {
+      result += _coef[catOffsetDiff + i] * featureValue(row, i);
+    }
+    return result;
+  }
+  
   private double forOtherColumns(double[] row, int[] enumOffset) {
     double result = 0.0;
     int coefLen = _coef.length;

h2o-genmodel/src/main/java/hex/genmodel/algos/coxph/CoxPHMojoReader.java

@@ -28,7 +28,7 @@ protected void readModelData() throws IOException {
     _model._interaction_targets = readkv("interaction_targets");
     _model._interaction_column_index = new HashSet<>();
     _model._interaction_column_domains = new HashMap<>();
-    _model._nums = readkv("num_numerical_columns");
+    _model._nums = readkv("num_numerical_columns", -1);
     _model._num_offsets = readkv("num_offsets");
 
     if (_model._interaction_targets != null) {

h2o-genmodel/src/test/java/hex/genmodel/algos/coxph/CoxPHMojoModelTest.java

@@ -1,6 +1,12 @@
 package hex.genmodel.algos.coxph;
 
 import org.junit.Test;
+import java.nio.file.Paths;
+
+import hex.genmodel.MojoModel;
+import hex.genmodel.easy.RowData;
+import hex.genmodel.easy.EasyPredictModelWrapper;
+
 import static org.junit.Assert.*;
 
 public class CoxPHMojoModelTest {
@@ -33,4 +39,46 @@ public void testForOneCategory() {
         assertEquals(0.6, mojo.forOneCategory(row, 1, 0), 0);
     }
 
+    @Test
+    public void testCoxPHBackwardCompatibility() throws Exception {
+        String mojofile = String.valueOf(
+            Paths.get(
+                CoxPHMojoModelTest.class.getClassLoader().getResource("hex/genmodel/algos/coxph/CoxPH_bc.zip").toURI()
+            ).toFile()
+        );
+
+        EasyPredictModelWrapper.Config config = new EasyPredictModelWrapper.Config()
+                .setModel(MojoModel.load(mojofile));
+        EasyPredictModelWrapper model = new EasyPredictModelWrapper(config);
+
+        String [][] inputs = new String[][] {
+            {"0", "50", "1", "-17.1553730321697", "0.123203285420945","0",
+                "0", "1", "-2.3004572363122033", "-23.840464872142775", "c0.l2", "c1.l2"},
+            {"0", "6", "1", "3.83572895277207", "0.254620123203285", "0", 
+                "0", "2", "33.01605679101838", "-12.944002705270874" ,"c0.l1","c1.l2"}, 
+            {"0", "1", "0", "6.29705681040383", "0.265571526351814", "0", 
+                "0", "3", "-22.54601829241052", "77.61631885563669", "c0.l2", "c1.l0"}
+        };
+        double [] expected = {0.0902431, 0.422815, 0.663896};
+        double[] preds = new double[inputs.length];
+        for (int i = 0; i < inputs.length; i++) {
+            RowData row = new RowData();
+            row.put("start", inputs[i][0]);
+            row.put("stop", inputs[i][1]);
+            row.put("event", inputs[i][2]);
+            row.put("age", inputs[i][3]);
+            row.put("year", inputs[i][4]);
+            row.put("surgery", inputs[i][5]);
+            row.put("transplant", inputs[i][6]);
+            row.put("id", inputs[i][7]);
+            row.put("C1", inputs[i][8]);
+            row.put("C2", inputs[i][9]);
+            row.put("C3", inputs[i][10]);
+            row.put("C4", inputs[i][11]);
+            preds[i] = model.predictCoxPH(row).value;
+        }
+
+        assertArrayEquals(expected, preds, 0.000001);
+    }
+
 }