[API][partial]
[Search][no]
[UI][no]
Anomaly detection engines assign a confidence value to specific entries in time-series objects to predict which events are anomalous.
Engine input
Anomaly detection engines can specify supportedInputFormats in their manifest for mime types they can support natively (e.g. text/plain, text/csv). Engines receive the entire file as its input and is responsible for outputting insights from the entire file in the .aion output.
Engine output
Anomaly detection engine output conforms to the anomaly validationContract and writes results into the object array as objects of type anomaly. The objectCategory array specifies one or more anomaly objects and their respective confidence value. Alternatively, you can use a series array instead of the object array to represent time-based anomaly objects.
- The engine provider specifies the text written to the
class key based on taxonomy. - The engine provider specifies the text written to the
@id key based on taxonomy. Make sure that the @id field is unique. - The weighting of or confidence in the anomalies is expressed with the
confidence key.
aiWARE doesn't require engines to conform to a master anomaly taxonomy. To map to external taxonomies, aiwARE uses class names (and @id if appropriate).
For more information, see the official anomaly validation contract json-schema.
Examples
The minimal example of an anomaly detection engine output using only the required keys:
{
"validationContracts": ["anomaly"],
"object": [
{
"type": "anomaly"
}
]
}
A typical engine output example that classifies anomalies in a time-based form:
{
"validationContracts": ["anomaly"],
"series": [{
"startTimeMs": 1243,
"stopTimeMs": 2120,
"object": {
"type": "anomaly",
"objectCategory": [
{
"class": "anomaly",
"@id": "12345",
"confidence": 0.468
}
]
}
}]
}
