The BasisEmbedding template does not allow for all-ones or all-zero basis states

[josh146]

There is a bug in the input validation for the BasisEmbedding template. It should be checking that the basis state array only contains 0 and 1 integers, but the way it is currently coded, it also disallows all-ones and all-zero basis states:

https://github.com/PennyLaneAI/pennylane/blob/master/pennylane/templates/embeddings/basis.py#L54

[antalszava]

Indeed, the following raises an error when using BasisEmbedding:

import pennylane as qml
import numpy as np

dev = qml.device('default.qubit', wires=[0,1,2])

@qml.qnode(dev)
def circuit():
    qml.templates.embeddings.BasisEmbedding(np.array([1,1,1]), wires=[0,1,2])
    return qml.expval(qml.PauliZ(0))

circuit()

ValueError: Basis state must only consist of 0s and 1s; got [1, 1, 1]

However, this got me wondering: isn't BasisEmbedding identical to BasisStatePreparation? 🤔

For the latter, the check is correct and it seems as though it had the same functionality. Maybe @mariaschuld could also help here. If they are identical, we could maybe consider having BasisEmbedding depend on BasisStatePreparation and eventually deprecating it.

[mariaschuld]

Yes, there is no real reason why we couldn't use BasisStatePreparation. I think I always thought it hacks the initial state, while I wanted a hardware-compatible solution for the embedding, but I guess this argument is not valid (any more). But this is a general decision, I propose to decouple it from this bug.

I made a fix in PR #1114...