View() and reshape()

I learned something about the difference between view() and reshape().

The view() does not change the original data stored. But reshape() may change the original data (when the original data is not continuous),reshape() may create a new memory space for the data

My doubt is whether the use of reshape() in RNN, CNN or other networks will affect the back propagation of errors, and affecting the final result?

It just make the computations less efficient.
For example, RNN usually take temporal dimensions 1st because blocks are contiguous from left to right.
If you slice through other dimensions, you are taking parts of different memory blocks.

1 Like

thank you so much. :grinning: