Tensor Broadcasting

xtensors.broadcast(X: XTensor, Y: XTensor, dimcast: Dimcaster, dimmerge: DimMerger, coordmerge: CoordMerger, shapecheck: bool = True) → Tuple[np.ndarray, np.ndarray, Dims, Coords]

Given two named tensors, return two numpy ND arrays of the same rank that can be broadcast together, along with the dimension names and coordinates after broadcasting.

Parameters:

• X,Y – xtensors.XTensor objects to be broadcast together

• dimcast – An object implementing the xtensors.Dimcaster protocol

• dimmerge – An object implmenting the xtensors.DimMerger protocol

• coordmerge – An object implementing the xtensors.CoordMerger protocol

xtensors.unilateral_broadcaster(X: XTensor, Y: XTensor) → Tuple[np.ndarray, np.ndarray, Dims, Coords]

This broadcaster takes the second tensor and permutes its dimensions to match the first one. The algorithm for finding the permutation is defined by unilateral_dimcast()

xtensors.vanilla_broadcaster(X: XTensor, Y: XTensor) → Tuple[np.ndarray, np.ndarray, Dims, Coords]

Vanilla broadcaster: the same as what’s used in torch’s named tensors. Dimension names (or None) have to match at the same axis position.

xtensors.cast(broadcaster: Broadcaster, X: XTensor) → Callable[[XTensor], XTensor]

xtensors.cast(broadcaster: Broadcaster, X: XTensor, Y: XTensor) → XTensor

Broadcast the second tensor with broadcaster(X, Y). If Y is None or not provided, a casting function is returned instead.