Code

Lasso Granger

Lasso-Granger is an efficient algorithm for learning the temporal dependency among multiple time series based on variable selection using Lasso.

Reference: A. Arnold, Y. Liu, and N. Abe. Temporal causal modeling with graphical Granger methods. In KDD, 2007.

Copula-Granger

Copula-Granger extends the power of Lasso-Granger to non-linear datasets. It uses the copula technique to separate the marginal properties of the joint distribution from its dependency structure.

Reference: Y. Liu, M. T. Bahadori, and H. Li, “Sparse-GEV: Sparse Latent Space Model for Multivariate Extreme Value Time Series Modeling“, ICML 2012.

Granger Causality for Irregular Time Series

The Generalized Lasso Granger is designed to discover the Granger causality relationship among irregular time series; times series whose samples are not recorded on regularly spaced timestamps.

Reference: M. T. Bahadori and Yan Liu, “Granger Causality Analysis in Irregular Time Series”, SDM 2012.

Forward-Backward Granger Causality

Forward-backward Granger causality utilizes both the original time series and the time-reversed (backward) time series for temporal dependence discovery. It provides more robust temporal dependence structure estimation than lasso Granger when the length of time series is short.

Reference: D. Cheng, M. T. Bahadori, and Y. Liu, “FBLG: A Simple and Effective Approach for Temporal Dependence Discovery from Time Series Data”. In KDD, 2014.

Granger Graphical Models for Anomaly Detection in Multivariate Time Series

Extensions of Granger graphical models are developed to detect anomalies in temporal dependence in multivariate time series data.

Reference: H. Qiu, Y. Liu,  N. Subrahmanya, W. Li. Granger Graphical Models for Time-Series Anomaly Detection. In International Conference on Data Mining  (ICDM’ 2012), 2012.

Transfer-accelerated, importance weighted consistent active learning (TIWCAL)

Extension of a theoretically sound online active learner to transfer learning and domain adaptation settings.

Reference: D. Kale and Y. Liu. Accelerating Active Learning with Transfer Learning, ICDM 2013. Slides.

Hierarchical Active Transfer Learning (HATL)

Active transfer learning framework that leverages shared cluster structure and feedback from active label queries to perform effective adaptive transfer learning.

Reference: D. Kale, M. Ghazvininejad, A. Ramakrishna, J. He, and Y. Liu. Hierarchical Active Transfer Learning, SDM 2015.

Greedy Low-Rank Tensor Learning

Greedy forward and orthogonal low rank tensor learning algorithms for multivariate spatiotemporal analysis tasks, including cokring and forecasting tasks.

Reference: T. Bahadori, R. Yu and Y. Liu. Fast Multivariate Spatiotemporal Analysis via Low Rank Tensor Learning , NIPS 2014.

SPALS: Leverage Scores Sampling Tensor ALS

Apply the leverage score sampling to greatly accelerate each step of the ALS algorithm for Tensor CP decomposition.

Reference: Dehua Cheng, Richard Peng, Ioakeim Perros, and Yan Liu, SPALS: Fast Alternating Least Squares via Implicit Leverage Scores Sampling, NIPS 2016.