This repository contains an extended version of CropGym; the code used in the paper "Adaptive fertilizer management for optimizing nitrogen use efficiency with constrained reinforcement learning".

How to install:

Requires python 3.9+

1. Clone the PCSE repo and install
2. Clone this repo
3. Download this file and unzip the single csv file under the folder pcse_gym/utils/weather_utils/random_weather_csv/ (create the non-existing folders). It is the generated random weather used to train the RL agents.
4. Install stable-baselines3, sb3contrib, scipy, lib_programname, rllte-core and tqdm with pip

How to use:

Example to train a model using the NUE reward function with the LagrangianPPO agent and the E3B intrinsic reward:

python train_winterwheat.py --reward NUE --environment 2 --agent LagPPO --seed 4 --nsteps 3000000 --random-weather --random-init --irs E3B