Analyzers

Analyzers: observers of a run, fed each settlement as it lands.

The backtrader idea — one evaluation loop, many pluggable observers. Analyzers never influence the run; they accumulate diagnostics and report in Result.analysis.

class emflow.run.analyzers.Analyzer

Bases: ABC

Base analyzer. setup receives the run context; on_settlement each scored origin; finalize returns the diagnostics dict.

property name: str

setup(feed, target_field, target_column, objective) → None

on_settlement(record) → None

finalize() → dict

class emflow.run.analyzers.PersistenceSkill

Bases: Analyzer

Scores a persistence baseline (last knowable value at each origin, broadcast over its targets) and reports the model’s skill against it. A model that can’t beat this hasn’t learned anything.

setup(feed, target_field, target_column, objective) → None

on_settlement(record) → None

finalize() → dict

class emflow.run.analyzers.QuantileCalibration

Bases: Analyzer

Empirical coverage per predicted quantile. A calibrated q10 should have ~10% of actuals below it; large gaps mean the intervals are lying.

on_settlement(record) → None

finalize() → dict

emflow.run.analyzers.default_analyzers(model=None) → List[Analyzer]