Relaxing score comparisons for rank_eval tests (#68976)

The currently allowed delta in this tests is to strict to account for slight floating point
differences across different platforms, e.g. ARM.

Closes #68936

Author: Christoph Büscher

client/rest-high-level/src/test/java/org/elasticsearch/client/RankEvalIT.java

@@ -17,13 +17,13 @@
 import org.elasticsearch.index.rankeval.ExpectedReciprocalRank;
 import org.elasticsearch.index.rankeval.MeanReciprocalRank;
 import org.elasticsearch.index.rankeval.PrecisionAtK;
-import org.elasticsearch.index.rankeval.RecallAtK;
 import org.elasticsearch.index.rankeval.RankEvalRequest;
 import org.elasticsearch.index.rankeval.RankEvalResponse;
 import org.elasticsearch.index.rankeval.RankEvalSpec;
 import org.elasticsearch.index.rankeval.RatedDocument;
 import org.elasticsearch.index.rankeval.RatedRequest;
 import org.elasticsearch.index.rankeval.RatedSearchHit;
+import org.elasticsearch.index.rankeval.RecallAtK;
 import org.elasticsearch.search.builder.SearchSourceBuilder;
 import org.junit.Before;
 
@@ -68,7 +68,7 @@ public void testRankEvalRequest() throws IOException {
         RankEvalResponse response = execute(rankEvalRequest, highLevelClient()::rankEval, highLevelClient()::rankEvalAsync);
         // the expected Prec@ for the first query is 5/7 and the expected Prec@ for the second is 1/7, divided by 2 to get the average
         double expectedPrecision = (1.0 / 7.0 + 5.0 / 7.0) / 2.0;
-        assertEquals(expectedPrecision, response.getMetricScore(), Double.MIN_VALUE);
+        assertEquals(expectedPrecision, response.getMetricScore(), 0.00000000001);
         Map<String, EvalQueryQuality> partialResults = response.getPartialResults();
         assertEquals(2, partialResults.size());
         EvalQueryQuality amsterdamQueryQuality = partialResults.get("amsterdam_query");
@@ -129,7 +129,7 @@ public void testMetrics() throws IOException {
 
             RankEvalRequest rankEvalRequest = new RankEvalRequest(spec, new String[] { "index", "index2" });
             RankEvalResponse response = execute(rankEvalRequest, highLevelClient()::rankEval, highLevelClient()::rankEvalAsync);
-            assertEquals(expectedScores[i], response.getMetricScore(), Double.MIN_VALUE);
+            assertEquals(expectedScores[i], response.getMetricScore(), 0.00000000001);
             i++;
         }
     }

modules/rank-eval/src/internalClusterTest/java/org/elasticsearch/index/rankeval/RankEvalRequestIT.java

@@ -116,7 +116,7 @@ public void testPrecisionAtRequest() {
         // the expected Prec@ for the first query is 4/6 and the expected Prec@ for the
         // second is 1/6, divided by 2 to get the average
         double expectedPrecision = (1.0 / 6.0 + 4.0 / 6.0) / 2.0;
-        assertEquals(expectedPrecision, response.getMetricScore(), Double.MIN_VALUE);
+        assertEquals(expectedPrecision, response.getMetricScore(), 0.0000000001);
         Set<Entry<String, EvalQueryQuality>> entrySet = response.getPartialResults().entrySet();
         assertEquals(2, entrySet.size());
         for (Entry<String, EvalQueryQuality> entry : entrySet) {
@@ -159,7 +159,7 @@ public void testPrecisionAtRequest() {
         // if we look only at top 3 documente, the expected P@3 for the first query is
         // 2/3 and the expected Prec@ for the second is 1/3, divided by 2 to get the average
         expectedPrecision = (1.0 / 3.0 + 2.0 / 3.0) / 2.0;
-        assertEquals(expectedPrecision, response.getMetricScore(), Double.MIN_VALUE);
+        assertEquals(expectedPrecision, response.getMetricScore(), 0.0000000001);
     }
 
     /**

modules/rank-eval/src/test/java/org/elasticsearch/index/rankeval/RankEvalResponseTests.java

@@ -93,7 +93,7 @@ public void testSerialization() throws IOException {
             randomResponse.writeTo(output);
             try (StreamInput in = output.bytes().streamInput()) {
                 RankEvalResponse deserializedResponse = new RankEvalResponse(in);
-                assertEquals(randomResponse.getMetricScore(), deserializedResponse.getMetricScore(), Double.MIN_VALUE);
+                assertEquals(randomResponse.getMetricScore(), deserializedResponse.getMetricScore(), 0.0000000001);
                 assertEquals(randomResponse.getPartialResults(), deserializedResponse.getPartialResults());
                 assertEquals(randomResponse.getFailures().keySet(), deserializedResponse.getFailures().keySet());
                 assertNotSame(randomResponse, deserializedResponse);