Python version of EcohNet

What

EcohNet is a method for computing predictability-based relationships among variables in a multivariate time series. EcohNet is implemented by a type of neural network called echo state network and the progressive selection of its input variables to identify the smallest set of predictors that minimize the prediction error for a given target variable, and then evaluate the unique contribution of each variable. Based on the concept of Granger causality, the network obtained by EcohNet can be interpreted as representative of the causal relationships inherent in a given time series.

Reference: https://www.pnas.org/doi/10.1073/pnas.2204405119

How to run

Prerequisits

• Python 3.8
• pipenv
• numpy

Run

On Google Colab

See ecohnet.ipynb.

On your environment

pipenv install
pipenv shell
$ python --version
> Python 3.8.*
python sctipts/run_ecohnet.py

Initially, the code executes echonet on rdata_011322_2.csv in data folder. To run it for your own data, the following steps are required:

1. Place your data (csv file) in data folder
2. Edit run_ecohnet.py as follows:

DATA_FILE = "YOURDATA.csv"

See ecohnet_script.ipynb for an example of execution and visualization.

Folder structure

• src/ecohnet.

  • Folder containing the main implementation.

• data

  • Folder containing the observation data.

• out

  • Folder where the experimental results are stored.

• ecohnet.ipynb.

  • The notebook to run the experiment and visualize it. It also contains the implementation needed for visualization.

• external/wolfram/LakeColors.txt

  • Exported mathematica colormaps. Used for visualization.

• scripts/

  • run_ecohnet.py
    • Script to run experiments on csv data from the console. Faster than running on a notebook.