Welcome to the documentation for Fast TICC.
Contents:
This library implements the clustering algorithm described in “Toeplitz Inverse Covariance-Based Clustering of Multivariate Time Series Data” by D. Hallac, S. Vare, S. Boyd, and J. Leskovec. Its purpose is to take multivariate time-series data where each data point has a timestamp and two or more data values, then assign labels to those data points based on how their values change (or don’t) together.
Here are a few examples of the kinds of data you could use with Fast TICC.
Car telemetry. Cars are continuously collecting data about the state of the vehicle, including the positions of the steering wheel, brake pedal, accelerator pedal; vehicle speed; and engine statistics such as RPM, temperature, and fluid levels. You can use TICC to automatically identify segments of data where the car is driving straight, braking to come to a stop, braking into a turn, or accelerating out of a turn.
Stock prices. TICC can identify periods of behavior when collections of stocks exhibit coordinated behavior – rising or falling together, being sold as a group or not.
Health/fitness telemetry. Suppose that a fitness tracker collects heart rate and velocity and acceleration data – the kind of data used to count steps, for example. TICC can label segments of data that may correspond to low- or high-intensity exercise, walking, running, or climbing steps.
Like any clustering algorithm, TICC is inherently unaware of the meaning of the data – it’s all just numbers. Understanding the meaning of the clusters it creates is up to you.
Our library was originally based on the reference TICC implementation by David Hallac and colleagues. We thank them for making their work available to the community.
If you’d like to contribute, please visit our GitHub repository and jump in. We welcome discussions, feature requests, and pull requests.