Docker is the only way to do ML. One thing I didn’t think to do before for a while was make separate docker files for each project so all of the deps are installed automatically and that has since helped tremendously.