ticclib

ticclib adapts the TICC class (see below) into an estimator compatible with the scikit-learn API.

TICC

TICC is a python solver for efficiently segmenting and clustering a multivariate time series. It is initialised with the number of clusters k, a window size w, a regularization parameter lambda and smoothness parameter beta. A TICC estimator can then be fit to a T-by-n data matrix. TICC breaks the T timestamps into segments where each segment belongs to one of the k clusters. The total number of segments is affected by the smoothness parameter beta. Segmentation and labelling is performed by running an expectation-maximisation algorithm where TICC alternately assigns points to clusters using a dynamic programming algorithm and updates the cluster parameters by solving a Toeplitz Inverse Covariance Estimation problem.

For details about the method and implementation see the paper [1].

Download & Setup

Download the source code for this implementation by running in the terminal:
```
 git clone https://github.com/grddavies/TICC.git
```

Install dependencies with pip

 pip install -r TICC/requirements.txt

or with conda

 conda install -n <envname> --file TICC/requirements.txt

Install ticclib in editable mode using pip
```
 pip install -e TICC/
```

Using TICC

The TICC-constructor takes the following parameters:

number_of_clusters: the number 'k' of underlying clusters to fit.
window_size: the size of the sliding window in samples.
lambda_parameter: sparsity of the Markov Random Field (MRF) for each of the clusters. The sparsity of the inverse covariance matrix of each cluster.
beta: The switching penalty used in the TICC algorithm. Same as the beta parameter described in the paper.
maxIters: The maximum iterations of the TICC algorithm before convergence. Default value is 100.
n_jobs: The maximum number of concurrently running jobs to be run via joblib.
cluster_reassignment: The proportion of points (0, 1) to move from a valid cluster to an empty cluster during fit.
random_state : The generator used to initialise assingments and randomize point shuffling during empty cluster reassignment.
verbose : If true print out iteration number and log any empty cluster reassignments.

Running the fit method on an array of multivariate timeseries data 'X', with rows in ascending-time order will return a fitted TICC estimator. A fitted estimator will have a list of 'k' clusters fitted to X, each with a block Toeplitz inverse covariance matrix which defines the cross-feature correlations for that cluster, and a list of points in X assigned to that cluster.

Example Usage

See example.py for usage, and comparison with a Gaussian Mixture Model. Requires matplotlib and networkx for visualisations.

References

[1] D. Hallac, S. Vare, S. Boyd, and J. Leskovec Toeplitz Inverse Covariance-Based Clustering of Multivariate Time Series Data Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 215--223

grddavies / TICC