[7.11][ML] Ensure feature importance is not empty in RegressionIT (#6…

…8074) (#68104)

This commit addresses the possibility that feature importance
is empty for the row with values 2, 20, 20. The reason this may
happen is because those rows' values are the average of the
other two and the model computes feature importance is close to
zero for that reason.

The test adjustes the values in a way that no row has values
that are the average of the other rows.

Fixes #59413
Fixes #59664

Backport of #68074

x-pack/plugin/ml/qa/native-multi-node-tests/src/javaRestTest/java/org/elasticsearch/xpack/ml/integration/RegressionIT.java

@@ -64,9 +64,14 @@ public class RegressionIT extends MlNativeDataFrameAnalyticsIntegTestCase {
     private static final String NUMERICAL_FEATURE_FIELD = "feature";
     private static final String DISCRETE_NUMERICAL_FEATURE_FIELD = "discrete-feature";
     static final String DEPENDENT_VARIABLE_FIELD = "variable";
-    private static final List<Double> NUMERICAL_FEATURE_VALUES = org.elasticsearch.common.collect.List.of(1.0, 2.0, 3.0);
-    private static final List<Long> DISCRETE_NUMERICAL_FEATURE_VALUES = org.elasticsearch.common.collect.List.of(10L, 20L, 30L);
-    private static final List<Double> DEPENDENT_VARIABLE_VALUES = org.elasticsearch.common.collect.List.of(10.0, 20.0, 30.0);
+
+    // It's important that the values here do not work in a way where
+    // one of the feature is the average of the other features as it may
+    // result in empty feature importance and we want to assert it gets
+    // written out correctly.
+    private static final List<Double> NUMERICAL_FEATURE_VALUES = org.elasticsearch.common.collect.List.of(5.0, 2.0, 3.0);
+    private static final List<Long> DISCRETE_NUMERICAL_FEATURE_VALUES = org.elasticsearch.common.collect.List.of(50L, 20L, 30L);
+    private static final List<Double> DEPENDENT_VARIABLE_VALUES = org.elasticsearch.common.collect.List.of(500.0, 200.0, 300.0);
 
     private String jobId;
     private String sourceIndex;
@@ -141,6 +146,7 @@ public void testSingleNumericFeatureAndMixedTrainingAndNonTrainingRows() throws
             assertThat(resultsObject.get("is_training"), is(destDoc.containsKey(DEPENDENT_VARIABLE_FIELD)));
             @SuppressWarnings("unchecked")
             List<Map<String, Object>> importanceArray = (List<Map<String, Object>>)resultsObject.get("feature_importance");
+
             assertThat(importanceArray, hasSize(greaterThan(0)));
             assertThat(
                 importanceArray.stream().filter(m -> NUMERICAL_FEATURE_FIELD.equals(m.get("feature_name"))
@@ -434,7 +440,6 @@ public void testDependentVariableIsLong() throws Exception {
         assertMlResultsFieldMappings(destIndex, predictedClassField, "double");
     }
 
-    @AwaitsFix(bugUrl = "https://github.com/elastic/elasticsearch/issues/59664")
     public void testWithDatastream() throws Exception {
         initialize("regression_with_datastream");
         String predictedClassField = DEPENDENT_VARIABLE_FIELD + "_prediction";