Hacker News Reader

Parent Full thread

quickthrower2

transpose? In the sense that

[[1,2,3,4,5],[6,7,8,9,10]]

Becomes

[[1,6],[2,7],[3,8],[4,9],[5,10]]

View on HN

Woa! Of course!

And if you use PyTorch, its O(1) operation. You just change how you index into the data.

I believe numpy introduced ndarrays with user-specified strides, that made this a O(1) operation. Even if not, it certainly predates pytorch in this regard. Pytorch (and other tensor libraries) just used the ndarray concept from numpy.

Thanks yeah I didn’t mean to imply that Pytorch invented the idea

Reply on news.ycombinator.com