Parameter expressions #139
Conversation
Codecov Report
@@ Coverage Diff @@
## master #139 +/- ##
=======================================
Coverage 99.29% 99.30%
=======================================
Files 7 7
Lines 284 287 +3
=======================================
+ Hits 282 285 +3
Misses 2 2
Continue to review full report at Codecov.
|
| @@ -227,37 +250,196 @@ def test_parameter_was_not_bound(self, recorder): | |||
| with recorder: | |||
| quantum_circuit(params={}) | |||
|
|
|||
| def test_invalid_parameter_expression(self, recorder): | |||
There was a problem hiding this comment.
If you're wondering, I also moved some tests around to try and organize the file better.
tests/test_converter.py
Outdated
| @@ -1063,3 +1075,38 @@ def circuit(params): | |||
| ] | |||
|
|
|||
| assert np.allclose(res, expected, **tol) | |||
|
|
|||
| def test_parameterexpression(self): | |||
There was a problem hiding this comment.
This is a new test.
| @@ -144,6 +144,29 @@ def test_quantum_circuit_with_gate_requiring_multiple_parameters(self, recorder) | |||
| assert recorder.queue[0].parameters == [0.5, 0.3, 0.1] | |||
| assert recorder.queue[0].wires == Wires([0]) | |||
|
|
|||
| def test_longer_parameter_expression(self): | |||
There was a problem hiding this comment.
this is a new test.
| @@ -1,5 +1,6 @@ | |||
| qiskit>=0.25 | |||
| pennylane>=0.15 | |||
| numpy | |||
| sympy | |||
There was a problem hiding this comment.
Usually, introducing a new dependency would be good to be considered, but in this case, Qiskit already depends on sympy. Worth noting that we can do this "without a cost" maybe in the PR description.
There was a problem hiding this comment.
do you mean in the changelog, or just in the PR?
| if p.parameters: # non-empty set = has unbound parameters | ||
| ordered_params = tuple(p.parameters) | ||
|
|
||
| f = lambdify(ordered_params, p._symbol_expr, modules=qml.numpy) |
There was a problem hiding this comment.
I wonder if this has any performance implications. 🤔 In Strawberry Fields, there was a case where using lambdify turned out to not be the best choice:
XanaduAI/strawberryfields#579
There was a problem hiding this comment.
Acoording to the sympy docs, lambdify is more efficient than evalf. Also, it allows us to use qml.numpy as the namespace for conversion. Using our wrapped numpy allows the expression to be differentiable.
I don't think this conversion is ever going to be very much of a bother compared to other bottlenecks. Qiskit does not seem very supportive of users creating elaborate expressions.
There was a problem hiding this comment.
Could we keep the condition as before and use the solution with lambdify only if len(p.parameters) > 1? That way we'll make use of it in fewer use cases.
There was a problem hiding this comment.
What is the expression is something like ParameterExpression(2*a). Then there's only one parameter, a, but we can't just use a's value in place of the expression. This is a case that would have fallen through the cracks before. If there's a viable alternative that's faster, I'm all for it, but I think this is the only way to support necessary functionality.
Co-authored-by: antalszava <[email protected]>
Closes Issue #129
This improvement allows the plugin to load circuits with more general parameter expressions.
For example,
Currently, the above code would raise an error.
I also add
sympyto the requirements, as it is explicitly used. Qiskit already has sympy as a dependency, so this does not add any cost to the requirements.