scitex-seizure-metrics

Unified evaluation library for seizure detection and forecasting.

Contents:

• Quick start
  • Detection (per-window classification)
  • Forecasting (continuous stream with explicit AlarmPolicy)
  • Cross-paper bridge
• policy
  • AlarmPolicy
• detection
  • evaluate()
• forecasting
  • bootstrap_ci()
  • evaluate()
  • evaluate_stream()
  • sweep_policies()
  • sweep_thresholds()
• classification (forecasting confusion matrix)
  • Confusion-matrix convention (alarm / prediction-opportunity basis)
  • Caveats a reviewer should confirm
  • Observed lead time
  • AlarmClassification
  • alarm_classification()
  • lead_time_summary()
  • observed_lead_times()
• sensitivity_tiw
  • SensitivityTiWCurve
  • TiWSignificance
  • binomial_above_chance()
  • chance_sensitivity()
  • monotone_upper_envelope()
  • seizures_from_labels()
  • sensitivity_tiw_curve()
  • surrogate_above_chance()
• bridge
  • Symbols
  • Windows per SOP (K)
  • Alarm sensitivity (per-seizure detection probability)
  • FP/hr (alarms per hour)
  • SampleToAlarmBounds
  • alarm_to_sample()
  • sample_to_alarm()
• calibration
  • CalibrationReport
  • calibration_report()
• surrogates
  • circadian()
  • from_history()
  • get()
  • multidien()
  • periodic()
  • persistence()
  • poisson()
  • register()
• papers
  • metrics()
  • metrics()
  • metrics()
  • metrics()
  • metrics()
  • metrics()
  • metrics()
• plots
  • cadence_ablation()
  • ioc_vs_surrogate()
  • metric_correlation_heatmap()
  • reliability_diagram()
  • sample_vs_alarm_scatter()
  • sensitivity_tiw()
  • sensitivity_vs_fp_per_hour()
• report (MetricsReport)
  • MetricsReport
• Full API
  • AlarmPolicy
  • MetricsReport

Concepts & math:

• Sample-to-Alarm Bridge: Mathematical Derivation
  • Setup and notation
  • Sample → Alarm
  • Alarm → Sample (inverse)
  • Properties and edge cases
  • Worked example
  • References
• The Alarm Confusion Matrix: Definitions and the SPH/SOP Framework
  • The SPH/SOP alarm framework
  • The four cells
  • Derived scores
  • Observed lead time
  • When to use detection vs forecasting metrics
  • Worked example
  • References
• Sensitivity vs Time-in-Warning: Mathematical Derivation
  • Setup and notation
  • Time-in-warning
  • Sensitivity (seizure-level, SOP-aware)
  • Chance baseline: the diagonal
  • Significance: is the curve above chance?
  • Properties and edge cases
  • Worked example
  • References

Decisions:

• ADR-0001: True negative for an alarm-based seizure-warning system — SOP-length interictal opportunities (2026-06-29)

Indices

• Index

• Module Index

• Search Page