Does anyone know how this compares to Twitter's recently released torch autograd? https://blog.twitter.com/2015/autograd-for-torch https://blog.twitter.com/201

Does anyone know how this compares to Twitter's recently released torch autograd? https://blog.twitter.com/2015/autograd-for-torch https://blog.twitter.com/2015/autograd-for-torch EDIT: To be a bit more specific, autograd just requires that one writes the forward model, and it can involve any program logic like conditionals, etc, so long as the end result is differentiable. It then takes care of all the gradient calculations.