feat: Add RobustKMeans, tests for new components, and bug fixes

New Features:
- RobustKMeans estimator with trim, noise_cluster, and m_estimator modes
  for outlier-resistant clustering with outlier score output
- Comprehensive test suites for previously untested components

Test Suites Added (108 new tests, 716 total):
- OutlierDetectionSuite: 16 tests for distance-based/trimmed detection
- SparseBregmanKernelSuite: 28 tests for sparse SE, KL, L1, Spherical kernels
- ConstraintsSuite: 30 tests for must-link/cannot-link constraints
- ConstrainedKMeansSuite: 17 tests for semi-supervised clustering
- RobustKMeansSuite: 17 tests for robust clustering with persistence
- ExtendedPersistenceSuite: 5 tests for new model persistence

Bug Fixes:
- SparseSEKernel.divergenceSparse missing 0.5 factor
- AgglomerativeBregmanModel persistence serializing IntParam instead of value

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: derrickburns

.gitignore

@@ -15,6 +15,7 @@ project/target/
 project/project/
 project/metals.sbt
 .bsp/
+.sbt/
 
 # Scala-IDE specific
 .scala_dependencies

CHANGELOG.md

@@ -13,12 +13,39 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - SECURITY.md with vulnerability reporting guidelines
 - CONTRIBUTING.md with development guidelines
 - Test suite fixes for Scala 2.12/2.13 compatibility
+- Spherical K-Means / cosine divergence support across estimators and models
+- New estimators: Mini-Batch K-Means, DP-Means, Balanced K-Means, Constrained (semi-supervised) K-Means, Kernel K-Means, Agglomerative Bregman, Bregman mixture models (EM), and CoClustering following the Spark ML Estimator/Model pattern
+- Bregman-native k-means++ seeding plus executable examples for spherical k-means
+- Outlier detection scaffolding with distance- and trim-based detectors
+- Property-based kernel accuracy suites and a performance benchmark suite with JSON outputs
+- **RobustKMeans estimator** for outlier-resistant clustering with trim, noise_cluster, and m_estimator modes
+- **Test suites for new components** (108 new tests, 716 total):
+  - OutlierDetectionSuite: 16 tests for distance-based and trimmed outlier detection
+  - SparseBregmanKernelSuite: 28 tests for sparse-optimized SE, KL, L1, Spherical kernels
+  - ConstraintsSuite: 30 tests for must-link/cannot-link constraints and penalty computation
+  - ConstrainedKMeansSuite: 17 tests for semi-supervised clustering with soft/hard constraints
+  - RobustKMeansSuite: 17 tests for robust clustering with outlier handling and persistence
 
 ### Fixed
 - Package name conflicts in StreamingKMeans and XMeans test suites
 - Scala 2.12 compatibility issues with `isFinite` method
+- SparseSEKernel divergenceSparse missing 0.5 factor (now matches SquaredEuclideanKernel)
+- AgglomerativeBregmanModel persistence serializing IntParam object instead of value
 - Spark 3.4 compatibility issues with `model.summary` API
 - CollectionConverters imports for cross-version support
+- BLAS `doMax` comparison, division-by-zero guards in strategies and co-clustering initializer, and invalid javac option in `build.sbt`
+
+### Changed
+- Unified divergence math via `BregmanFunction` and refactored kernel factory for consistency
+- Added Bregman-native initialization path and enriched Scaladoc across major estimators
+- Enhanced clustering iterator and constraint frameworks to support new variants
+
+### Performance
+- Accelerated squared-Euclidean assignment and Elkan-style cross-iteration bounds for Lloyd's iterations
+- Vectorized BLAS helpers for common linear algebra operations
+
+### Removed
+- Legacy RDD API and associated coreset/transform modules (DataFrame/ML API is now the sole surface)
 
 ## [0.6.0] - 2025-10-18
 

CLAUDE.md

@@ -21,7 +21,8 @@
 3. **Mathematical fidelity first.** Correct Bregman formulations beat micro-perf. Perf changes must not alter semantics.
 4. **Determinism matters.** Same seed ⇒ identical results. Avoid nondeterministic ops in core loops.
 5. **Tight PRs.** Small, test-backed, CI-friendly. No speculative abstractions.
-6. **Maintain the roadmap.** When making changes, update `ROADMAP.md` to reflect completed work, new issues discovered, or priority changes.
+6. **Maintain the roadmap.** Keep `ROADMAP.md` forward-looking; move completed work into `CHANGELOG.md` and leave the roadmap focused on upcoming items and priorities.
+7. **Use shared model helpers.** Clustering models should mix in `HasTrainingSummary` and `CentroidModelHelpers` for consistent summaries/metadata; avoid reintroducing ad-hoc summary fields.
 
 ---
 
@@ -37,12 +38,11 @@
 **Claude must:**
 1. **Inspect `ROADMAP.md`** at the start of significant work to understand current priorities and context.
 2. **Update `ROADMAP.md`** when:
-   - Completing a bug fix → mark as ✅ FIXED with date
-   - Discovering a new bug → add to Bug Fixes section with priority
-   - Completing a feature → move to Completed Items section
-   - Identifying technical debt → add to appropriate section
+   - Discovering a new bug or opportunity → add with priority
+   - Changing priorities → update ordering/sections
    - Making architectural decisions → add to Decision Log
-3. **Reference roadmap items** in commit messages and PR descriptions where applicable.
+3. **Move completed work** (features, fixes, docs, perf) into `CHANGELOG.md` with dates and drop it from `ROADMAP.md` so the roadmap stays forward-looking.
+4. **Reference roadmap items** in commit messages and PR descriptions where applicable.
 
 ---