ml-ol-simulation

This repository contains all code for running a simulation which examines the effects of ML adoption on organizational learning (OL). All models are based on the agent-based model developed by James March (1991) and use the Mesa framework (Python) as technical foundation.

Repository structure

The following files are included in this repository:

• utils/: Utility functions for batch runners, configuration and data preprocessing
• models/: Actual simulation models (separated into agent and model definitions), along with utility functions for agent initialization and metric calculation
• nbs/: Notebooks used for testing model implementations, speedups (e.g., matrix-based metric calculations)
• data/: Collected data from simulation runs will be saved here
• /: Run script and environment configuration

Setup

All simulations should be run in a virtual environment in order to create a standardized and reproducible environment.

The following assumptions are made about your system:

• Working installation of Git
• Working installation of Python (at least version 3.7)
• Working shell for running command-line instructions (e.g., Git Bash, VS Code or PyCharm terminal)

The following setup steps have to be completed in order to run simulations:

1. Clone repository to target machine: git clone git@github.com:felixpeters/ml-ol-simulation.git or git clone https://github.com/felixpeters/ml-ol-simulation.git
2. Navigate to repository directory: cd ml-ol-simulation (on UNIX-based systems)
3. Create data folder in repository root: mkdir data (on UNIX-based systems)
4. Setup a virtual environment: python -m venv venv
5. Activate virtual environment: source venv/bin/activate (on UNIX-based systems)
6. Install required packages: pip install -r requirements.txt

Running simulations

The logic for running simulations is included in the run script (run.py) in the project root. The script is meant to be run in the virtual environment. Model runs can be configured using constants (see line starting # define constants). In detail, the following parameters can be set:

• MODEL_NAME: Determine the model to be run, currently base or alt
• DATA_PATH: Leave this untouched if you followed the setup instructions above
• CPU_COUNT: Set this to a fixed number if you wish, otherwise all available CPU cores will be used
• CONFIG: Determines which parameter set to use for simulation run, see utils/config.py for details

Conduct the following steps to run the simulation:

1. Activate the virtual environment (if it isn't already): source venv/bin/activate
2. Run the simulation: python run.py

The runtime can be estimated from the output of the run script (iteration speed is updated continuously).

Contact

Email: peters@is.tu-darmstadt.de