An toy experiment with Netflix's Metaflow framework.
Building graphs from RAISE's Open Source data. Each data file contains three H5 datasets of grid shape (65, 33, 33)
. For the toy example, to produce the dataset, we need to:
- Convert the H5 file to a Numpy file.
- Build the graph with a PyTorch-Geometric Data structure.
- Serialize the graph to a PyTorch object.
To download the data from RAISE's resources, run the download-data.sh
.
run.sh
to run the entire pipeline.