Merge pull request #4931 from breakanalysis/1.8-lp-reduce-min-batch-size

1.8 LinkPrediction reduce min batch size

Author: breakanalysis

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

@@ -75,7 +75,7 @@ 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/LinkPrediction.java

@@ -28,7 +28,7 @@
 import org.neo4j.gds.ml.linkmodels.pipeline.logisticRegression.LinkLogisticRegressionPredictor;
 
 public abstract class LinkPrediction extends Algorithm<LinkPrediction, LinkPredictionResult> {
-
+    static final int MIN_NODE_BATCH_SIZE = 10;
     private final LinkLogisticRegressionData modelData;
     private final LinkFeatureExtractor linkFeatureExtractor;
     private final Graph graph;

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(LinkPrediction.MIN_NODE_BATCH_SIZE)
             .concurrency(concurrency());
 
         topK().ifPresent(knnBuilder::topK);