Abstract:Time-series causal discovery (TSCD) is a fundamental problem of machine learning. However, existing synthetic datasets cannot properly evaluate or predict the algorithms' performance on real data. This study introduces the CausalTime pipeline to generate time-series that highly resemble the real data and with ground truth causal graphs for quantitative performance evaluation. The pipeline starts from real observations in a specific scenario and produces a matching benchmark dataset. Firstly, we harness deep neural networks along with normalizing flow to accurately capture realistic dynamics. Secondly, we extract hypothesized causal graphs by performing importance analysis on the neural network or leveraging prior knowledge. Thirdly, we derive the ground truth causal graphs by splitting the causal model into causal term, residual term, and noise term. Lastly, using the fitted network and the derived causal graph, we generate corresponding versatile time-series proper for algorithm assessment. In the experiments, we validate the fidelity of the generated data through qualitative and quantitative experiments, followed by a benchmarking of existing TSCD algorithms using these generated datasets. CausalTime offers a feasible solution to evaluating TSCD algorithms in real applications and can be generalized to a wide range of fields. For easy use of the proposed approach, we also provide a user-friendly website, hosted on <a class="link-external link-http" href="http://www.causaltime.cc" rel="external noopener nofollow">this http URL</a>.

Learning Temporal Causal Sequence Relationships from Real-Time Time-Series

Causal Discovery from Temporal Data: An Overview and New Perspectives

Amortized Causal Discovery: Learning to Infer Causal Graphs from Time-Series Data

Inferring extended summary causal graphs from observational time series

Survey and Evaluation of Causal Discovery Methods for Time Series

TS-CausalNN: Learning Temporal Causal Relations from Non-linear Non-stationary Time Series Data

Learning Causal Relations from Subsampled Time Series with Two Time-Slices

Causal Discovery in Semi-Stationary Time Series

Temporal Causal Inference with Time Lag

Learning domain-specific causal discovery from time series

Nonlinear Causal Discovery in Time Series

Causality Learning from Time Series Data for the Industrial Finance Analysis Via the Multi-dimensional Point Process

Causal Discovery from Temporally Aggregated Time Series

Entropy-based Discovery of Summary Causal Graphs in Time Series

Towards Learning Segmented Temporal Sequences: A Decision Tree Approach

CausalTime: Realistically Generated Time-series for Benchmarking of Causal Discovery

Causal Temporal Regime Structure Learning

Causal inference for time series

CUTS: Neural Causal Discovery from Unstructured Time-Series Data

Detecting and Explaining Causes From Text For a Time Series Event

Mining Interpretable Spatio-temporal Logic Properties for Spatially Distributed Systems