This is the official implementation of paper: Estimating Mutual Information between Time Series and
Temporal Event Sequences Across Diverse Analysis Tasks. KDD 2026.

If you use the code, please cite our paper.

All the programs can be run directly to reproduce the results in the paper.

The implementation of the proposed measurement is in:
+ normalize_clustered_mi.py

Experiment code:

Time-Delayed Mutual Information Experiments:
+ tdmi_controlled_digit_ground_truth_exp.py

Repeated Temporal Pattern:
1) Seasonality:
+ Autocorrelation.py
+ Fourier_transform.py
+ seasonality_real_data.py

2) Point Repeated Pattern:
+ tempurature_exp.py

Continuous Feature Selection:
+ feature_selection.py
+ variants_for_ablation_study.py

Discrete Covariates Identification:
+ Covariate_selection_rossmann.py
+ Covariate_selection_M5.py
It requires a dependency on https://github.com/google-research/timesfm and https://huggingface.co/amazon/chronos-2

As Continuous Feature Selection and Discrete Covariates Identification experiments take time to run, we also saved the detailed results in the following files for quick checking:

Continuous Feature Selection
+ feature_selection_threshold90.txt

Discrete Covariates Identification
+ M5_threshold90.txt
+ rossmann_threshold90.txt