Skip to content

[T-070a] FailureFeatureBuilder: weighted parameter + class one-hot + trajectory features #126

Closed

Description

@CodewithJha

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

Implements FR-ANA-001

Branch feat/t-070a-failurefeaturebuilder-weighted-parameter-class-one — one task = one branch = one PR (doc 32 §6)

Summary

Implement analysis/features.py building the per-failing-run feature matrix of doc 09 §3.1: normalised params (w_param), failure-class one-hot × w_class, trajectory features (s_fail/L, t_fail/T_max, min_clearance/d_norm, max_loc_error/e_max) × w_traj.

Context / motivation

'Same conditions + same kind of failure at the same place ⇒ same mode' — the feature design defines what a failure mode is.

Scope

In scope

  • FeatureMatrix (X, run_ids, column names) — internal analysis type (DataFrames allowed inside analysis, doc 34 §7).
  • FeatureWeights typed settings model (param, class_, traj) with validation > 0.
  • FeatureDiagnostics (n_input, n_fail, n_excluded_invalid, missing_traj).

Implementation guidance

  • Module roboprobe/analysis/features.py (layer: domain — numpy/pandas only, no SQLAlchemy/FastAPI imports; import-linter enforces).
  • Interface (doc 05 §7): class FailureFeatureBuilder: def __init__(self, weights: FeatureWeights) -> None; def build(self, runs: Sequence[RunRecord], space: ScenarioSpace) -> FeatureMatrix.
  • FeatureMatrix (frozen dataclass in analysis/types.py): X: npt.NDArray[np.float64] shape (n_fail, d + n_classes + 4), run_ids: tuple[UUID, ...], columns: tuple[str, ...], space_version: int.
  • Column blocks (doc 09 §3.1): (1) space.normalize(values) for every enabled, unpinned parameter × w_param; (2) one-hot of primary_failure over FailureClass members in enum order × w_class; (3) trajectory block × w_traj: s_fail / L_nominal, t_fail / T_max, min_clearance / clear.d_norm, max_loc_error / loc.e_max — each clipped to [0, 1].
  • Normalisers come from the run's resolved oracle thresholds snapshot (stored with the campaign) — never re-read global defaults, otherwise re-analysis after a config change silently changes modes.
  • Only verdict == FAIL runs from phases other than confirm are included; INVALID runs are excluded and counted in the returned diagnostics.
  • Weights come from AnalysisSettings.feature_weights (core/settings); missing trajectory metrics (e.g. analytic backend) ⇒ block filled with 0 and diagnostics.missing_traj=True.
  • 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.feature_weights

Acceptance criteria

  • Hand-computed feature rows for 3 synthetic runs.
  • Weights honoured from config.
  • Pinned parameters are excluded from the parameter block (test with a pinned space).
  • Changing analysis.feature_weights via env (ROBOPROBE_ANALYSIS__FEATURE_WEIGHTS__TRAJ=0) scales the block — verified by test.
  • Builder is pure: same inputs ⇒ byte-identical X (hash test).
  • mypy strict + import-linter green (no infrastructure imports).
  • 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_features.py.
  • Property test (Hypothesis): all features ∈ [0, max(weight)]; row count equals FAIL count.
  • Factory-built RunRecords (tests/factories.py) — no DB.

Dependencies

Blocked by: #61 (T-011a), #34 (T-021)

Blocks: #127 (T-070b)

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/features.py

Parallel with T-072/T-076.

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-070a (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 §3.1 · doc 05 §7 · ADR-0012 · doc 16 §2 · doc 27 §4.8

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

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