如何理解ndarray.reshape函数？

Question

The prototype of reshape() is that reshape(shape, order="C") , and the type of shape is tuple. So we should call this function with myarray.reshape((1000, 1, 32, 32)) , But I find that many use myarray.reshape(1000, 1, 32, 32) , why?

Answer 1

It's a bit of hidden flexibility built into the reshape method.

The keyword here needs to be explicit: you can't do for example:

myarray.reshape(1000, 1, 32, 32, "C")

You'll get a TypeError , saying that an integer is required.

(In fact, even using a tuple:

myarray.reshape((1000, 1, 32, 32), "C")

raises the TypeError .)

If you look at the source code (eg, at GitHub , you'll see that, after the keywords are parsed, there is a check for 0 or 1 arguments. In that case, the argument is interpreted as a tuple. If there are more arguments, each is interpreted as an integer and combined into a tuple as the new shape (the keyword arguments have already been taken out).

---

As to which one you should use: I guess there's not really a good answer.
You could stick with the documentation, and use tuples.
The multiple-integer-arguments convention, however, feels rather obvious.

Do stick to the convention that you, or the project you're working on, use(s). Don't use tuples and individual integer arguments in the same project.