This repo contains code to evaluate spatiotemporal predictions with Optimal Transport (OT). In contrast to standard evaluation metrics (MSE, MAE etc), the OT error is a spatial metric that evaluates the spatial distribution of the errors.
Install the package via pypi:
pip install geot
You might have to install a suitable torch version first. Then, i. If you want to use Optimal Transport as a loss function for training a DL model, make sure to install torch for cuda.
Alternatively, install from source:
git clone https://github.com/mie-lab/geospatialOT.git
cd geospatialOT
conda create -n geot_env
conda activate geot_env
pip install .
Check out our tutorial to get started with a simple example.
Explanation of usage:
- Assume we want to predict some observations in several locations, e.g., bike sharing demand at bike sharing stations, over time
- Usually, the error is just averaged over locations (MSE between GT and prediction)
- However, in real-world applications, there are distance-based costs involved with prediction errors. For example, prediction errors cause costs for relocating bikes
- Assue we can define a cost matrix with the pairwise costs between locations, indicating the cost to account for errors
- With Optimal Transport, we can compute the minimum costs for transforming the predictions to the ground truth - a better indicator for the real-world costs of prediction errors than just the MSE
With this code, you can compute the OT error for your predictions. The input is usually just the observations at a set of location, the predictions and the cost matrix. The output is the OT error (a single number) or the optimal transport matrix T.