Skip to content

[T-070b] FailureClusterer: HDBSCAN with min-cluster rule and small-n fallback #127

Closed

Description

@CodewithJha

Task T-070b · Milestone M6 — Analysis + risk · Size S · Priority P0 · Epic #6

Implements FR-ANA-001

Branch feat/t-070b-failureclusterer-hdbscan-with-min-cluster — one task = one branch = one PR (doc 32 §6)

Summary

Implement analysis/clustering.py: HDBSCAN(min_cluster_size=max(floor, ⌈fraction·n_fail⌉), metric='euclidean', cluster_selection_method='eom'); noise points kept as 'unclustered'; n_fail < 6 ⇒ provisional modes grouped by failure class.

Context / motivation

Unknown number of modes with arbitrary shapes and explicit noise is exactly HDBSCAN's strength (ADR-0012).

Scope

In scope

  • ClusteringResult (labels, probabilities, outlier scores).
  • HdbscanSettings model in core/settings/analysis.py with the two new keys documented in doc 27 §4.8 (generated reference via T-003c).

Implementation guidance

  • Module roboprobe/analysis/clustering.py; interface per doc 05 §7: class FailureClusterer: def __init__(self, settings: HdbscanSettings) -> None; def fit(self, F: FeatureMatrix) -> ClusteringResult.
  • min_cluster_size = max(analysis.hdbscan.min_cluster_size_floor, ceil(analysis.hdbscan.min_cluster_fraction * n_fail)); min_samples defaults to min_cluster_size (setting analysis.hdbscan.min_samples: int | None, new).
  • Use sklearn.cluster.HDBSCAN(metric="euclidean", cluster_selection_method="eom", store_centers=None) (scikit-learn ≥ 1.3; no separate hdbscan package — keeps the dependency set lean, ADR-0012).
  • ClusteringResult (frozen): labels: NDArray[int] (−1 = noise), probabilities, outlier_scores (GLOSH, computed via 1 - probabilities fallback if not exposed), method: Literal["hdbscan", "class_fallback"], min_cluster_size.
  • Small-n fallback: if n_fail < analysis.hdbscan.min_failures_for_clustering (new, default 6) group by primary_failure and mark modes provisional (ClusteringResult.provisional=True).
  • Determinism: HDBSCAN is deterministic for fixed input order — sort rows by run_id before fitting and map labels back.
  • Log analysis.clustered with n_fail, n_clusters, n_noise, method, duration_ms (doc 21 §5.1).
  • Standards: respect the layering contracts (doc 16 §2) and conventions in doc 34; no hard-coded tunables — thresholds/defaults come from typed settings (doc 27); no magic numbers (ruff PLR2004); public APIs typed (mypy strict) and documented.

Configuration keys

From the configuration reference (doc 27). Keys marked (new) must be added to the typed settings models and appear in the generated reference (T-003c).

  • analysis.hdbscan.min_cluster_fraction
  • analysis.hdbscan.min_cluster_size_floor
  • analysis.hdbscan.min_samples (new)
  • analysis.hdbscan.min_failures_for_clustering (new)

Acceptance criteria

  • Synthetic Gaussian blobs: ARI ≥ 0.9 (doc 18 §4.6).
  • Fallback path tested.
  • n_fail = 0 ⇒ empty result, no exception.
  • Fallback groups exactly by failure class and sets provisional=True.
  • Label assignment invariant under input row permutation (test).
  • Runs in < 1 s for 2 000 failures (benchmark marker).
  • Tests listed below added and green in CI; coverage of new code meets the gate in doc 18 §3.
  • Docs, config reference and CHANGELOG Unreleased updated where behaviour, keys or interfaces changed.
  • Definition of Done (doc 32 §3) satisfied.

Tests required

  • tests/unit/analysis/test_clustering.py.
  • Property test: labels length = n rows; −1 allowed only for noise.
  • Benchmark test @pytest.mark.benchmark for 2 000 rows.

Dependencies

Blocked by: #126 (T-070a)

Blocks: #128 (T-103), #129 (T-071a), #130 (T-071b)

Parallelization notes

Files / paths owned by this task — avoid editing other paths; shared files (pyproject.toml entry points, cli/main.py, api/routers/__init__.py, router.tsx, docker/compose.yaml, ci.yml) are append-only:

  • backend/src/roboprobe/analysis/clustering.py

Small.

Definition of Done

Per doc 32 §3 and AGENTS.md §5: lint, format, mypy strict and import-linter green; tests added and passing; docs updated; PR template completed with Closes #<this issue>; Conventional Commit messages referencing T-070b (doc 16 §6); commits authored only by the repository owner identity (AGENTS.md §4); CI green; reviewer checklist (doc 32 §5.1) satisfied.

References

Design docs: doc 09 §3.2 · doc 18 §4.6 · ADR-0012 · doc 21 §5.1

Requirements: FR-ANA-001 (traceability: doc 28)

External:

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:analysisFailure clustering, SHAP, rules, interactionsmilestone:M6M6 — Analysis + riskpriority:P0Critical path / blocks otherssize:S<= 1 daytaskChild task of a milestone epictype:featureNew functionality

    Projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions