Torch needs QT, which takes forever to compile. And to build BLAS is also pain in the ass! Tried Amazon AWS cloud server with GPU, feels better than the reviews said. But the price at $0.6/hr is really frustrating.
Caffe is completely written in C++, but relies heavily on multiple Google produced libraries.
Cannot help pondering how much energy there is to waste for the world’s people to try these things out, just like couldn’t help thinking how much gas people had wasted learning to drive when I was learning to drive, years ago.