AFRL Reseach

Research for drone trajectory prediction.

TODO:

• Implement a "true" test of model accuracy that integrates the velocity prediction to find position.
• Implement model hyperparameter configuration.

Dependencies

The dependencies of this project are managed with uv and python virtual environments.

Install uv (if not already installed):

curl -LsSf https://astral.sh/uv/install.sh | sh

Install dependencies and activate virtual environment.

uv sync
source .venv/bin/activate

Getting the data

All the data is stored in the data directory.

Think of this as a "staging area" where we can convert all the data into a consistent format.

It also by default stores all intermediate steps.

TLDR: Run the get data script to get data.

python src/get_data.py

Inside the data/scripts directory, you will find 4 scripts, each corresponding to a different dataset.

Preprocessing

The data preprocessing is run in the notebook at the root of the project directory.

Create folds

When creating folds, we must be careful to not incure data leakage between folds. Thus, we split the data into folds by files.

python src/datasets.py data/velocity/max_norm --shuffle --seed 42

Replace data/velocity/max_norm with data/position/max_norm for position data.

Train

To train, call the main python file.

python -m src.main <name of run> -s <json file of split> -f <0-indexed split> -c <json configuration>