With deep learning, using conda environments and environment modules to switch between various tensorflow, pytorch, cuda, and cudnn versions for reproducing/bui

With deep learning, using conda environments and environment modules to switch between various tensorflow, pytorch, cuda, and cudnn versions for reproducing/building off of other's results is essential. It is sort of hilarious how many breaking changes there are even between minor versions of these. Thank gosh I was able to install a 2 year old version of pytorch (0.4.1) today to reproduce some code from within the last year.