This repository provides an improved version of https://github.com/zxq623/FedHydro that does not require Docker.
- Download the CAMELS dataset from https://ral.ucar.edu/solutions/products/camels
- Choose one region and select seven basins from it as data-rich watersheds, and one as a data-scarce watershed.
- Place related files into the directories ./dataset/series_data/discharge_data and ./dataset/series_data/forcing_data
- Execute ./dataset/series_data/utils/generate_data.py to generate the merged dataset
- Place the merged dataset into ./dataset/series_data/
- Create a class for loading the merged dataset in ./dataset/
To achieve the best results, we suggest using PyTorch == 1.11.0.
You can adjust the parameters in utils/options.py, and then run main_fedhydro.py.
We provide two pretrained models on basin18,the code used to be test the model is under folder /utils