Installation.md

Guide to Run Battle of Networks 2024 (BoN 2024)

Prerequisites

• Python Version: The models for BoN 2024 were trained using Python 3.11.6. Ensure you have this version installed. You can download it from Python 3.11.6 Download (https://www.python.org/downloads/release/python-3116/).
• For Mac Users: Homebrew installed (https://brew.sh)

Setting Up a Virtual Environment

To manage dependencies effectively, it's recommended to set up a virtual environment (venv). This isolates the project dependencies from your global Python installation. Follow these steps:

Windows (with Visual Studio>2015 installed)

1. Create a Virtual Environment:

• On Windows open Command Prompt (press Windows Key + R, then type cmd and press Enter).
• Create a new virtual environment by running: python -m venv /path/to/new/virtual/environment_name Replace /path/to/new/virtual/environment_name with your desired directory path.

2. Activate the Virtual Environment:

• Activate your virtual environment by running: /path/to/new/virtual/environment_name/Scripts/activate. Ensure you use the correct path where your virtual environment is located.

3. Install Required Modules:

• Browse to the code directory of the Water-Futures BoN2024 codebase cd /path/to/the/code/BoN2024
• Install all necessary modules by running: pip install -r requirements_windows.txt

Mac with Apple Silicon

1. Create a Virtual Environment:

• Open a terminal.
• Create a new virtual environment by running: python -m venv /path/to/new/virtual/environment_name Replace /path/to/new/virtual/environment_name with your desired directory path.

2. Activate the Virtual Environment:

• Activate your virtual environment by running: source /path/to/new/virtual/environment_name/bin/activate. Ensure you use the correct path where your virtual environment is located.

3. Install OpenMP library

• brew install libomp

4. Install Required Modules:

• Browse to the code directory of the Water-Futures BoN2024 codebase cd /path/to/the/code/BoN2024
• Install all necessary modules by running: pip install -r requirements_mac.txt

Manual installation

1. Follow steps 1 and 2 from the previous section based on your system.

2. Install the following libraries using the instructions for your machine on each website:

• pytorch (https://pytorch.org)
• tensorflow (https://www.tensorflow.org/install/pip)
• lightbm (https://pypi.org/project/lightgbm/) -> pip install lightgbm
• darts (https://github.com/unit8co/darts/blob/master/INSTALL.md) -> pip install "u8darts[notorch]"
• jupyter (https://jupyter.org/install) -> pip install jupyter
• plotly (https://plotly.com/python/getting-started/) -> pip install plotly==5.18.0
• dash (https://dash.plotly.com/installation) -> pip install dash
• openpyxl (https://pypi.org/project/openpyxl/) -> pip insall openpyxl
• adtk (https://pypi.org/project/adtk/) -> pip install adtk
• metpy (https://pypi.org/project/MetPy/) -> pip install metpy
• dm-tree (https://pypi.org/project/dm-tree/) -> pip install dm-tree
• torchmetrics (https://pypi.org/project/torchmetrics/) -> pip install torchmetrics

Download the data

If you want to produce new models and results for reproducibility, you can skip this part: the code is already set up to use the data directory in this project.

On the other hand, if you downloaded this repo to investiagte the results produced by our team, follow the following steps:

1. download the data folder from Zenodo (#TBD).
2. change the value of the environmental variable BON2024_DATA_FOLDER in the .env file present in this repository to the full path of the downloaded data folder.

Running the Code

After setting up the virtual environment and installing the dependencies, you're ready to run the code. Run the water_futures.py file or using Jupyter Notebook, select the correct Kernel and run water_futures.ipynb.

Additional Notes

### GPU Acceleration BoN 2024 can leverage NVIDIA GPU for enhanced performance. If your system does not have an NVIDIA GPU, modify the configuration in the 'water_futures.xx' file:

Change Line 265 from: cfg['device'] = 'cuda' to: cfg['device'] = 'cpu'

Note: the Metal acceleration on Mac with M1 or M2 is not working at the time of the submission. Use 'cpu' to ensure a correct running on Apple devices, too.

Important Note

The virtual environment for this project includes both TensorFlow and PyTorch. Be aware that this can consume significant storage space.

Results

The forecast results can be found at: data/results/strategies/avg_top5/avg_top5__iter_1__eval__.xlsx or: data/solutions/WaterFutures_SolutionTemplate_W1.xlsx

Model Run Times

Each models takes the following times during training, if you aim to run the entire project account for long times (more than two days on a single machine). We have 52 iterations during training, 4 in testing and 1 for the final evaluation. All models are run for multiple seeds (for the solution we used n_train_seed=1 and n_test_seed =3 ), but a more appropriate and complete solution should use 5/10 seed.

The reported times are from a Mac with an 8 core (4+4) Apple silicon M1 processor and 16 GB of RAM.

Model Time(s/iteration)

LightGBM Simple | 285 LightGB Robust | 555 Light GBM last week | 283 XGBM | 281 Wavenet | 3250 (55 min)

All other non deep-learning models take negligible times (less then a minute per iteration).