Insert SC2LE into the base image, preferably the latest version available, but if you can put an argument to set the version it will be better. Alternatively copy the files directly from the alvarofpp/s2client image.
Create example images configuring Tensorflow or PyTorch (other machine learning libs are welcome) based on the URNAI image.
We need a new Dockerfile for our development environment. I recommend using NVIDIA CUDA as a basis, but feel free to use any other Docker image.