Algorithms

• Core algorithms
• Custom algorithms
  • Available custom algorithms
  • Implementing a custom algorithm
    • settings.CUSTOM_ALGORITHMS
    • The custom algorithm file
    • General purpose algorithms only
    • JSON friendly results ONLY
    • int and floats ONLY and no nans
  • anomalyScore
  • Test for unreliable results
  • Custom algorithm requirements
  • Additional results in the return
  • Error handling
    • The use of numba optimisations
    • flux/tornado
  • Example custom algorithms
  • Running a Mirage only custom algorithm on a metric all the time
  • Some things to consider
• matrixprofile
  • Considerations when implementing matrixprofile
  • HOWEVER - trigger_history_override
  • Enabling matrixprofile to run in Mirage
• m66
  • m66 TCPDBench results
  • Testing m66 with TCPDBench
    • Build datasets on CentOS 8
    • results
• sigma
• spectral_residual
• DBSCAN
• Local Outlier Factor
• PCA
• isolation_forest
• prophet
• one_class_svm
• MSTL
• spectral_entropy
• macd
• LAD
• probabilistic_forecasts_generalized_pareto_distribution_ets
• grafana_promql_anomaly_detection