Information Theory Inspired Pattern Analysis for Time-series Data

This repository includes the code required to reproduce the experiments and figures in the pre-print paper "Information Theory Inspired Pattern Analysis for Time-series Data".

1. Requirements

This directory contains the code required to run the experiments to produce the results presented in the paper "Information Theory Inspired Pattern Analysis for Time-series Data"

To get started and download all dependencies, run:

pip install -r requirements.txt 

2. Minder dataset

2.1 Raw Data

The Minder dataset is privacy. To apply and download the dataset, please contact with Prof. Payam Barnaghi.

2.2 Visualisation of the pre-processed data

2.3 Feature generation and selection

Generate entropy features, shown as EntropyFeatures.

Entropy features for the Minder dataset includes: Entropy of Markov chains: ./MINDER/EntropyFeatures/activity_daytime_per_week_mk_entropy.ipynb, ./MINDER/EntropyFeatures/activity_night_per_week_mk_entropy.ipynb

Entropy rate of Markov chains: ./MINDER/EntropyFeatures/activity_daytime_per_week_mk_entropy_rate.ipynb, ./MINDER/EntropyFeatures/activity_night_per_week_mk_entropy_rate.ipynb

Entropy production of Markov chains: ./MINDER/EntropyFeatures/activity_daytime_per_week_mk_entropy_production.ipynb, ./MINDER/EntropyFeatures/activity_night_per_week_mk_entropy_production.ipynb

Von Neumann Entropy of Markov chains (activity frequency): ./MINDER/EntropyFeatures/activity_daytime_per_week_mk_vn_entropy_frequency.ipynb, ./MINDER/EntropyFeatures/activity_night_per_week_mk_vn_entropy_frequency.ipynb

Von Neumann Entropy of Markov chains (activity duration): ./MINDER/EntropyFeatures/activity_daytime_per_week_mk_vn_entropy_duration.ipynb, ./MINDER/EntropyFeatures/activity_night_per_week_mk_vn_entropy_duration.ipynb

The baseline features: ./MINDER/EntropyFeatures/activity_daytime_night_per_week_frequency.ipynb

2.4 Modelling and results

The evaluation results are:

3. Epileptic Seizure Recognition Dataset (ESRD)

This repository includes the code for Epileptic Seizure Recognition Dataset (ESRD).

3.1 Raw Data

Download the orginal data[1].

3.2 Visualisation of the pre-processed data

The code is in ./ESRD/plot_raw.ipynb.

3.3 Feature generation and selection

Generate entropy features, shown as ./ESRD/generate_entropy.ipynb.

Select entropy features by Pearson relationship matrix and mutual information.

3.4 Modelling and results

(1) Baseline-CNN

The code is shown in ./ESRD/model_baseline_CNN.ipynb.

(2) Baseline-LSTM

The code is shown in ./ESRD/model_baseline_LSTM.ipynb.

(3) Entropy-MLP

The code is shown in ./ESRD/network_pytorch.

The evaluation results are:

[1]Andrzejak, R.G., Lehnertz, K., Mormann, F., Rieke, C., David, P. and Elger, C.E., 2001. Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: Dependence on recording region and brain state. Physical Review E, 64(6), p.061907.

4. PTB Diagnostic ECG Database (PTBDB)

This repository includes the code for PTB Diagnostic ECG Database.

4.1 Raw Data

Download the orginal data[2].

Download the pre-processed data.

Here we utilize the pre-processed data.

4.2 Visualisation of the pre-processed data

For the normal participants:

For the abnormal participants:

The code is in ./PTBDB/generate_entropy.ipynb.

4.3 Feature generation and selection

Generate entropy features, shown as ./PTBDB/generate_entropy.ipynb.

Select entropy features by Pearson relationship matrix and mutual information.

4.4 Modelling and results

(1) Baseline-CNN

The code is shown in ./PTBDB/model_baseline_CNN.ipynb.

(2) Baseline-MLP

The code is shown in ./PTBDB/model_baseline_MLP.ipynb.

(3) Entropy-MLP

The code is shown in ./PTBDB/network_pytorch.

The evaluation results are:

The code is shown in ./PTBDB/plot_result.ipynb.

[1]Andrzejak, R.G., Lehnertz, K., Mormann, F., Rieke, C., David, P. and Elger, C.E., 2001. Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: Dependence on recording region and brain state. Physical Review E, 64(6), p.061907.

[2]Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., ... & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online]. 101 (23), pp. e215 - e220.