We train a self-supervised siamese model that enables querying for similar behavior on spatiotemporal volumes. Here we provide the code and data needed to reproduce the representative figures of the paper. See the notes and the included readme file for details.
We train a self-supervised siamese model that enables querying for similar behavior on spatiotemporal volumes. Here we provide the code and data needed to reproduce the representative figures of the paper.
To run it, you will need:
Any modern Linux distribution. Tested on CentOS 7.0
Singularity
An HPC-focused containerization tool.
Available as a package or can be built from source.
You will need root to build the S4 container image (see the README).
NVIDIA drivers.
The code uses Tensorflow and CUDA, so an NVIDIA GPU is needed.
The driver cannot be containerized with Singularity, so it should be installed on the host.
Everything else is installed automatically inside the container.
To build and run the code, you will run the build-and-reproduce.sh
script:
cd S4
chmod u+x build-and-reproduce.sh && ./build-and-reproduce.sh
You might need to comment out the lines calling git
if you downloaded the code without using git.
If you run into problems, follow the detailed instructions in the README.md