It's a list containing lists. Each sublist is an epoch, and it follows the following order: validation loss, training loss (over 1), and the PCC.
Install docker and make, then run the command make run. This should automatically download relevant data files into a new directory titled data and then create and run a docker image named ml-nwp. This will take a couple of minutes, but after the initial setup run, it should be a lot faster.
Install docker. Run the command ./dev.sh; this may take a minute to install and setup the environment, and you should be placed into a shell; subsequent entries will be substantially faster. Here, the directory /root (or ~ since the default user is root and I didn't change that) is linked to the directory where the script is run from; all changes made in any other place will be deleted when the environment is exited.
You can try running the setup.sh script; this may or may not work. This is almost certainly outdated and will probably give you a strange environment.