Reduce minimum batch size in link prediction predict

Author: breakanalysis

alpha/alpha-proc/src/main/java/org/neo4j/gds/ml/linkmodels/pipeline/predict/ApproximateLinkPrediction.java

@@ -39,6 +39,7 @@
 import java.util.stream.Stream;
 
 public class ApproximateLinkPrediction extends LinkPrediction {
+    static final int MIN_NODE_BATCH_SIZE = 10;
     private final KnnBaseConfig knnConfig;
 
     public ApproximateLinkPrediction(

alpha/alpha-proc/src/main/java/org/neo4j/gds/ml/linkmodels/pipeline/predict/ExhaustiveLinkPrediction.java

@@ -35,6 +35,7 @@
 import java.util.stream.LongStream;
 
 public class ExhaustiveLinkPrediction extends LinkPrediction {
+    private static final int MIN_NODE_BATCH_SIZE = 10;
     private final int topN;
     private final double threshold;
 
@@ -75,7 +76,7 @@ ExhaustiveLinkPredictionResult predictLinks(
                 partition,
                 progressTracker
             ),
-            Optional.empty()
+            Optional.of(MIN_NODE_BATCH_SIZE)
         );
 
         ParallelUtil.runWithConcurrency(concurrency, tasks, Pools.DEFAULT);

alpha/alpha-proc/src/main/java/org/neo4j/gds/ml/linkmodels/pipeline/predict/LinkPredictionPredictPipelineBaseConfig.java

@@ -116,6 +116,7 @@ default KnnBaseConfig approximateConfig() {
         var knnBuilder = ImmutableKnnBaseConfig.builder()
             .sampleRate(sampleRate())
             .nodeWeightProperty("NotUsedInLP")
+            .minBatchSize(ApproximateLinkPrediction.MIN_NODE_BATCH_SIZE)
             .concurrency(concurrency());
 
         topK().ifPresent(knnBuilder::topK);