anomaly-detection builds on Facebook's fbprophet library, enabling you to identify unusual outliers and trends within hierarchical time series data in only a few lines of code. This library:
- Flags and prioritizes anomalies based on configurable Prophet forecasts
- Identifies changepoints in your data to help you spot sudden trend shifts
- Enables you to plot and measure trend differences between hierarchical groups
What makes this package different from other anomaly detection libs?
- Leverages Facebook's Prophet algorithm, rather than older, classical approaches (e.g., KNN, smoothing algorithms, etc.)
- Explores differences in parameters derived from generative models, rather than focusing only on discrimant boundaries
- Overrides Prophet's methods to provide an easier usage and debugging experience
Start by installing pystan and and fbprophet, then install this repo using git clone https://github.com/ntlind/anomaly-detection.
Check out the two .ipynb examples in /examples