Use new training containers directly

I ran the Notebook until job.run(), since I didn't have an atomicrex executable in my environment and I couldn't install it with conda (something about incompatible requirements...). Do we need to add that to the env or did I mess something up?