Counterfactual Reasoning with Probabilistic Graphical Models for Analyzing Socioecological Systems

This bundle contains the manuscript submitted to the Environmetrics and entitled "Counterfactual Reasoning with Probabilistic Graphical Models for Analyzing Socioecological Systems". The organisation is the following:

• code: Java files for replicating the case of study.
• data: datasets and models used in the case of study. This includes the original data and the processed data and model.
• lib: packages needed for running the code.
• output: output of learning (set of precise SCMs) and inference (csv files with the query results).

Setup

The code has been tested with Java openjdk 12.0.1. For checking The Java version running in your system use the following command:

$ java --version

openjdk 12.0.1 2019-04-16
OpenJDK Runtime Environment (build 12.0.1+12)
OpenJDK 64-Bit Server VM (build 12.0.1+12, mixed mode, sharing)

Running the code

Here we illustrate how to do all the tasks for counterfactual reasoning. First code/inference.java allows to preprocess the original dataset and to build the partially-defined SCM used in the paper. This basically creates files final_model.uai and final_data.csv at data folder. For running it:

java -cp ./lib/credici_socioeco.jar ./code/preprocess.java

Afterwards, a java terminal script for learning is provided at code/LearnSingleRun.java. Note that this allows to run a single EM run and it can be used for any model. For details about its usage, run:

java -cp ./lib/credici_socioeco.jar ./code/LearnSingleRun.java --help

Usage: <main class> [-hq] [--debug] [-d=<dataPath>] [-l=<logfile>]
                    [-m=<maxIter>] [-o=<output>] [-s=<seed>] <modelPath>
      <modelPath>           Model path in UAI format.
      --debug               Debug flag. Defaults to false
  -d, --data=<dataPath>     Data path in CSV format.
  -h, --help                Display a help message
  -l, --logfile=<logfile>   Output file for the logs.
  -m, --maxiter=<maxIter>   Maximum EM internal iterations. Default to 500
  -o, --output=<output>     Output folder for the results. Default working dir.
  -q, --quiet               Controls if log messages are printed to standard output.
  -s, --seed=<seed>         Random seed. If not specified, it is randomly selected.

For instance, with the data and model from the case of study run:

java -cp ./lib/credici_socioeco.jar ./code/LearnSingleRun.java --data ./data/final_data.csv --maxiter 300 --seed 1234 ./data/final_model.uai

[2023-11-26T12:28:34.551755][INFO][seed1234] Set up logging
[2023-11-26T12:28:34.554098][INFO][seed1234] args: --data;./data/final_data.csv;--maxiter;300;--seed;1234;./data/final_model.uai
[2023-11-26T12:28:34.554560][INFO][seed1234] Starting logger with seed 1234
[2023-11-26T12:28:34.595258][INFO][seed1234] Loaded data from: ././data/final_data.csv
[2023-11-26T12:28:34.621941][INFO][seed1234] Loaded model from: ././data/final_model.uai
 [...]

Finally, the last task involves running the desired queries. For those in the case of study, run:

java -cp ./lib/credici_socioeco.jar ./code/inference.java