Time-Based Anomaly Detection
The Anomaly Detection Operator is a low code tool for integrating anomaly detection into any enterprise application.
It uses time series constructive anomaly detection to flag anomalous moments in the data, by time and by ID.
For more information, see the Anomaly Detection section of the ADS documentation.
Input Data
- A date/time column.
- A target column.
- (Optional) One or more series columns, such that the target is indexed by date/time and series.
- (Optional) An arbitrary number of extra variables.
Besides this input data, the you can specify validation data, if available. Validation data
must have all the columns of the input data plus a binary column titled
anomaly. The anomaly column must have a value of minus 1
for anomalous rows and 1 for normal rows.
Finally, provide test_data to receive test metrics and evaluate the
Operator's performance more easily. Test data must be indexed by date and (optionally) series.
Test data must have a value of minus 1 for anomalous rows and 1 for normal rows.
kind: operator
type: anomaly
version: v1
spec:
datetime_column:
name: ds
input_data:
url: https://raw.githubusercontent.com/facebook/prophet/main/examples/example_yosemite_temps.csv
model: autots
target_column: yads operator run -f anomaly.yamlMulitvariate Anomaly Detection
If you have extra variables that you think might be related, then use multivariate anomaly detection. All the extra columns specified in the input data are used in finding if the target column is anomalous.
Auto Model Selection
Operators users don't need to know anything about the underlying models to use them. By
default we set model to auto. However, if you want more control over the
modeling parameters, set the model parameter to either autots or
automlx and then pass parameters directly into
model_kwargs.