BooleanPerceptronClassifier with batchSize > 1 can silently lose the weight updates of misclassified training documents: the weight FST is rebuilt only when a misclassified document sits at a multiple of batchSize and never after the last batch, and updates inside a batch are computed from the last FST build and overwrite each other, so the trained model is wrong (released since 5.0.0)

Finding LFV-680Area: classification / BooleanPerceptronClassifier (batched training)Priority: mediumImpact: silent wrong resultsNovelty: new

Finding page for LFV-680

AI-assisted finding, confirmed by automated checks, not yet triaged by a human Lucene committer. This site files nothing upstream.

What it shows

You index two training documents labelled false, 'bb' and 'aa aa', train BooleanPerceptronClassifier with bias 1.5 and batchSize 2, and expect assignClass('aa aa') = false, which is what batchSize 1 gives on the same data and what both an online and a mini-batch perceptron give. Instead it returns true (score 0.81): the update for the misclassified 'aa aa' (batch position 1) is written only to the pending weight map, because the FST that holds the weights is rebuilt only when a misclassified document sits at a position that is a multiple of batchSize, and the constructor ends by clearing that map without a final rebuild. Updates inside a batch are also computed from the last FST build and overwrite each other, so with batchSize 10 and three training documents 'aa' labelled false only the first update is applied and the model still answers true.

The test shows that BooleanPerceptronClassifier with batchSize 2, trained on 'bb' and 'aa aa' (both labelled false, bias 1.5), never applies the weight update for the misclassified 'aa aa' (batch position 1), because the FST is rebuilt only when a misclassified document is at a multiple of batchSize and never after the last batch.

Expected vs actual

Expected: assignClass("aa aa") = false, as with batchSize 1 on the same data. Actual on main with -ea and with -da: true (expected:<false> but was:<true>).

Files

Run it

# Lucene checkout at the tested commit, built with: ./gradlew --no-daemon assemble   (JDK 25)
LUCENE=/path/to/lucene; V=12.0.0-SNAPSHOT
CP=$LUCENE/lucene/core/build/libs/lucene-core-$V.jar
CP=$CP:$LUCENE/lucene/codecs/build/libs/lucene-codecs-$V.jar
CP=$CP:$LUCENE/lucene/test-framework/build/libs/lucene-test-framework-$V.jar
CP=$CP:$LUCENE/lucene/analysis/common/build/libs/lucene-analysis-common-$V.jar
CP=$CP:$LUCENE/lucene/classification/build/libs/lucene-classification-$V.jar
CP=$CP:$LUCENE/lucene/grouping/build/libs/lucene-grouping-$V.jar
CP=$CP:$LUCENE/lucene/queries/build/libs/lucene-queries-$V.jar
CP=$CP:randomizedtesting-runner-2.9.2.jar:junit-4.13.2.jar:hamcrest-3.0.jar   # test-framework dependencies (Maven Central)
CP=$CP:junit-platform-console-standalone-6.1.3.jar   # JUnit Platform console launcher (Maven Central)

javac -d classes -cp "$CP" $(find src -name '*.java')

# run 1: assertions on, seed DEADBEEF
java -ea -esa --add-modules jdk.incubator.vector --enable-native-access=ALL-UNNAMED -Dtests.seed=DEADBEEF \
     -cp "classes:$CP" org.junit.platform.console.ConsoleLauncher \
     execute --select-class org.apache.lucene.classification.TestBooleanPerceptronBatchSize

Gradle alternative: copy the test into the matching module's src/test/ tree of the checkout and run ./gradlew -p lucene/<module> test --tests <class> -Ptests.seed=<seed> (add -Ptests.asserts=false for a run without assertions).

Feedback

Open an issue in lucene-ai-bug-hunt. Please do not open issues or pull requests in apache/lucene about this finding unless a Lucene committer agrees it should be filed.

Snapshot of 2026-11-04. AI-assisted; not yet triaged by a human Lucene committer. Feedback: open an issue in lucene-ai-bug-hunt.