-
Notifications
You must be signed in to change notification settings - Fork 47
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
RFC: clarify broadcast_to
semantics
#823
Comments
I agree it should be updated. The unidirectional broadcasting is important for in-place operators (see https://data-apis.org/array-api/latest/API_specification/broadcasting.html#in-place-semantics). |
@asmeurer While I'm not aware of a read-only concept in the array API, I just wanted to point out that NumPy's |
Read-only isn't a concept that's in the array API. Not all libraries might implement it. The array API leaves all mutation with views undefined so this isn't an issue. |
It is a slight issue, since Not sure I like the "bidirectional" term, but happy if Aaron is. It might be good to just say that broadcastable arrays means that there is a common broadcast shape that both arrays can be "broadcast to" (the algorithm describing how to find said broadcast shape). |
Agreed that it's a slight issue at the moment - more for NumPy than for the array API standard though. The ideal solution would be something like copy-on-write for NumPy, which could be introduced in a backwards compatible way since |
I'll have to disagree until I see a clearer plan on how NumPy could introduce CoW (for read-only arrays?) while not generally breaking view semantics in a way that the whole world gets wrong results. (And NumPy isn't the only array library that uses view semantics!) |
I can't say that I find the names "bidirectional" and "unidirectional" particularly appealing. I personally think of broadcasting as a (non-closed) binary operation on shape tuples. "Unidirectional" broadcasting is a special case where the result of the broadcast has to be the same as the second shape. |
I'm finding the
broadcast_to
specification a little underspecified. In the docs we see the following for the shape parameter:The broadcasting link goes on to specify bidirectional broadcasting. That would imply to me that
np.broadcast_to(np.asarray([[-1, -1], [-1, -1]]), (2, 1, 2))
should work since shapes (2, 2) and (2, 1, 2) are bidirectionally compatible. Somewhat reasonably in my opinion, NumPy did not interpret this in that way and raises an exception.Since
np.broadcast_to(np.asarray([[-1, -1], [-1, -1]]), (1, 2, 2))
does work, it seems that broadcasting compatibility is unidirectional. i.e.x.shape
must be broadcastable toshape
. Is it worth spelling out explicitly the difference in how this works, like ONNX does? I couldn't find any explanation in the standard itself.It does say the following, although I read the "a specified shape" part as "any shape" rather than simply the
shape
parameter.If this ambiguity is shared I am happy to contribute a clarification.
The text was updated successfully, but these errors were encountered: