timeToTarget v1.2.1

by thomas

prediction of boundary crossing

Properties
  • resource
  • metric name of the metric configuration
  • target threshold value that is checked
  • window size of prediction window in seconds (default 3600). Alternatively you can use an iso8601 duration expression such as ‘PT2H’
  • trend ‘up’ if checking that predictions will be higher than target, ‘down’ if lower than target.

    When not set, direction depends on the last observation (‘up’ if target > last_observation)

  • confidenceLevel confidence level of the prediction: indicates the minimal proportion of the prediction distribution that must cross the target.

    confidenceLevel=0.5 (default) can be used with all algorithms, and checks whether the (most likely) prediction crosses the target.

    Other values for confidenceLevel check that target is crossed with that confidence (according to the confidence interval produced by the model!).

    Currently only Arima supports confidence intervals for predictions.

    By default (e.g. confidenceLevel=0.5)

    • Crossed means that, according to the model, there is above 50% chance that the target might be crossed;
    • Not Crossed hence means that this chance is below 50% (or likewise an above 50% chance that target is NOT crossed).

    When low (e.g. confidenceLevel=0.1),

    • Crossed means that the target might be attained (at least 1 in 10);
    • Not Crossed means that there is above 90% chance that the target will NOT be attained

    When high (e.g. confidenceLevel=0.9),

    • Crossed means that the target most probably will be attained (more than 9 out of 10, according to the model);
    • Not Crossed that there is above 10% chance that the target will NOT be attained

  • params overrides for the analytics option configuration (see below)
States
  • Crossed if it is predicted that values above (up) or below (down) the target will be attained ( with above confidenceLevel probability )
  • Not Crossed otherwise
Raw data returned
  • target the target value
  • crossed_at observation at which the target is crossed first
  • confidence_level confidence level used to decide whether the target might be attained
  • prediction_windowthe size of the prediction window in which target crossings are checked
  • trend up or down, depending whether predictions above or below the target were checked
  • extremal maximal lower (‘trend=up’) or minimal upper (‘trend=down’) boundary of the (one-sided) confidence intervals of the predictions.

    For c=0.5, this is the maximal (‘trend=up’) or minimal (‘trend=down’) prediction.

  • last_observation last observation before the predictions
  • first_prediction first prediction
  • last_prediction last prediction

you must have TS Analytics package to run this sensor.

Before using this sensor, configure the analytics options for this metric using the analytics dashboard. Alternatively you should give a complete option configuration in the params section. See apidoc on how to specify options.

Required properties:

  • resource
  • metric
  • target
  • window
  • trend
  • confidenceLevel
  • params

External Settings:

  • waylay_analytics

States:

  • Crossed
  • Not Crossed

Raw data:

  • target (double)
  • crossed_at (object)
  • prediction_window (string)
  • trend (string)
  • extremal (double)
  • last_observation (object)
  • first_prediction (object)
  • last_prediction (object)