You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This resolvesdata-apis#471.
The existing spec provides conflicting guidance saying both that the
axes over which to compute the dot product must be equal, while also
saying that input arrays must be broadcast compatible without
qualification, thus implying that the contracted axis could also
broadcast.
This commit explicitly defines broadcast behavior for only the
contracted axes, thus bringing vecdot inline with broadcasting
behavior in tensordot.
This resolves#471.
The existing spec provides conflicting guidance saying both that the
axes over which to compute the dot product must be equal, while also
saying that input arrays must be broadcast compatible without
qualification, thus implying that the contracted axis could also
broadcast.
This commit explicitly defines broadcast behavior for only the
contracted axes, thus bringing vecdot inline with broadcasting
behavior in tensordot.
The spec implies that it should https://data-apis.org/array-api/latest/API_specification/generated/signatures.linear_algebra_functions.vecdot.html#signatures.linear_algebra_functions.vecdot. However, other functions like matmul do not broadcast along the contracted axis.
This came up in the numpy implementation, which is currently wrong, but is being fixed at numpy/numpy#21928.
The text was updated successfully, but these errors were encountered: