Upvoted the story because I am keen to see some discussion around it. I have kept an eye on Torch7 but there have been a few things that make me a little wary.

Upvoted the story because I am keen to see some discussion around it. I have kept an eye on Torch7 but there have been a few things that make me a little wary. A prominent one is the sheer number of unreplied to posts on their mailing list. Numpy/Scipy and Julia lists seem much more welcoming and prompt. Even if people there cannot help they do respond. Now if we ignore the social side of it, I would really like to know why Torch7/Lua and not say Numpy/Scipy/Theano or Julia for that matter. To me scratching an itch is a fine enough excuse, just wanted to know if there are any compelling technical advantages to do it this way. Some aspects are quite promising, one is to JIT compile away most of the FFI glue, another is interaction with Numpy/Scipy code. There have been a few projects to allow seamless back and forth between Lua and Python, so if that can be done with Numpy and Cython that is indeed quite an exciting possibility.