CropGym

Introduction

This is the code base for the paper “CropGym: a Reinforcement Learning Environment for Crop Management” by Hiske Overweg, Herman N.C. Berghuijs and Ioannis N. Athanasiadis.

Installation

The code has been tested using python 3.8.5. To install all required packages, do the following:

Clone this repository

Install the crop gym environment with the following command

pip install -e gym_crop

Install required packages for the training script by running:

pip install -r requirements.txt

This patch to the PCSE package is required to be able to run the code.

Training an agent

Agents can be trained using the following command:

python scripts/training_script.py --name repr --beta 10 --tensorboard /path/to/tensorboard/save/dir --log /path/to/model/save/dir --n_steps=10000

The results in the paper have been obtained with a yet unpublished branch of the PCSE package, which contains a recent calibration of crop growth parameters.