Given a function which defines a tape transform, convert the function into one that applies the tape transform to quantum functions (qfuncs).
tape_transform (function or single_tape_transform) – the single tape transform to convert into the qfunc transform.
A qfunc transform, that acts on any qfunc, and returns a new qfunc as per the tape transform. Note that if
tape_transformtakes additional parameters beyond a single tape, then the created qfunc transform will take the same parameters, prior to being applied to the qfunc.
- Return type
Given a single tape transform
my_transform(tape, x, y), you can use this function to convert it into a qfunc transform:
>>> my_qfunc_transform = qfunc_transform(my_transform)
It can then be used to transform an existing qfunc:
>>> new_qfunc = my_qfunc_transform(0.6, 0.7)(old_qfunc) >>> new_qfunc(params)
It can also be used as a decorator:
@qml.qfunc_transform def my_transform(tape, x, y): for op in tape.operations + tape.measurements: if op.name == "CRX": wires = op.wires param = op.parameters qml.RX(x * param, wires=wires) qml.RY(y * qml.math.sqrt(param), wires=wires) qml.CZ(wires=[wires, wires]) else: op.queue() @my_transform(0.6, 0.1) def qfunc(x): qml.Hadamard(wires=0) qml.CRX(x, wires=[0, 1])
>>> dev = qml.device("default.qubit", wires=2) >>> qnode = qml.QNode(qfunc, dev) >>> print(qml.draw(qnode)(2.5)) 0: ──H───────────────────╭Z──┤ 1: ──RX(1.5)──RY(0.158)──╰C──┤
The transform weights provided to a qfunc transform are fully differentiable, allowing the transform itself to be differentiated and trained. For more details, see the Differentiability section under Usage Details.
qfunc transforms, when used inline (that is, not as a decorator), take the following form:
if they do not permit any parameters. We can break this down into distinct steps, to show what is happening with each new function call:
Create a transform defined by the transform weights:
>>> specific_transform = my_transform(transform_weights)
Note that this step is skipped if the transform does not provide any weights/parameters that can be modified!
Apply the transform to the qfunc. A qfunc transform always acts on a qfunc, returning a new qfunc:
>>> new_qfunc = specific_transform(ansatz)
Finally, we evaluate the new, transformed, qfunc:
So the syntax
simply ‘chains’ these three steps together, into a single call.
When applying a qfunc transform, not only is the newly transformed qfunc fully differentiable, but the qfunc transform parameters themselves are differentiable. This allows us to train both the quantum function, as well as the transform that created it.
Consider the following example, where a pre-defined ansatz is transformed within a QNode:
dev = qml.device("default.qubit", wires=2) def ansatz(x): qml.Hadamard(wires=0) qml.CRX(x, wires=[0, 1]) @qml.qnode(dev) def circuit(param, transform_weights): qml.RX(0.1, wires=0) # apply the transform to the ansatz my_transform(*transform_weights)(ansatz)(param) return qml.expval(qml.PauliZ(1))
We can print this QNode to show that the qfunc transform is taking place:
>>> x = np.array(0.5, requires_grad=True) >>> y = np.array([0.1, 0.2], requires_grad=True) >>> print(qml.draw(circuit)(x, y)) 0: ──RX(0.1)───H──────────╭Z──┤ 1: ──RX(0.05)──RY(0.141)──╰C──┤ ⟨Z⟩
Evaluating the QNode, as well as the derivative, with respect to the gate parameter and the transform weights:
>>> circuit(x, y) 0.9887793925354269 >>> qml.grad(circuit)(x, y) (array(-0.02485651), array([-0.02474011, -0.09954244]))
Internally, the qfunc transform works as follows:
def transform(old_qfunc, params): def new_qfunc(*args, **kwargs): # 1. extract the tape from the old qfunc, being # careful *not* to have it queued. tape = make_tape(old_qfunc)(*args, **kwargs) # 2. transform the tape new_tape = tape_transform(tape, params) # 3. queue the *new* tape to the active queuing context new_tape.queue() return new_qfunc
Note: this is pseudocode; the actual implementation is significantly more complicated!
Steps (1) and (3) are identical for all qfunc transforms; it is only step (2),
tape_transformand the corresponding tape transform parameters, that define the qfunc transformation.
That is, given a tape transform that defines the qfunc transformation, the decorator elevates the tape transform to one that works on quantum functions rather than tapes. This decorator therefore automates the process of adding in the queueing logic required under steps (1) and (3), so that it does not need to be repeated and tested for every new qfunc transform.