QuantumCircuit to PennyLane template conversion

[antalszava]

Context:
Each quantum software framework has its architecture when it comes to how they handle creating circuits and applying gates. Therefore a converter between Qiskit's QuantumCircuit and a custom template in PennyLane could be provided. This converter could then be utilized by users who are new to PennyLane but have prior experience with other quantum computing software frameworks.

Description of the Change:

Adds the converter function load in converter.py that creates a PennyLane template from an arbitrary

• QuantumCircuit object;
• QASM file;
• QASM string.

Adds the entrypoint pennylane.io such that PennyLane can use the load function.

Benefits:

Allows the conversion from Qiskit circuit object to Pennylane templates. Once the conversion has been done, these templates can then be placed within a quantum circuit in PennyLane.

Circuits containing parameters converted this way can be parametrized by passing a dictionary as an argument to load.

Each converted circuit can be added to a quantum circuit in PennyLane an arbitrary many times. The user can also specify different parameters each time if desired.

Further allows shifting the converted circuit. Passing the wire_shift argument allows to shift the indices of wires already defined in the original QuantumCircuit object e.g. if the QuantumCircuit acts on wires [0, 1] and if wire_shift == 1, then the returned Pennylane template will act on wires [1, 2].

Possible Drawbacks:
N/A

Related GitHub Issues:
N/A

[codecov]

Codecov Report

Merging #55 into master will decrease coverage by 0.81%.
The diff coverage is 94.23%.

@@            Coverage Diff             @@
##           master      #55      +/-   ##
==========================================
- Coverage     100%   99.18%   -0.82%     
==========================================
  Files           8        9       +1     
  Lines         316      368      +52     
==========================================
+ Hits          316      365      +49     
- Misses          0        3       +3

Impacted Files	Coverage Δ
pennylane_qiskit/qiskit_device.py	100% <ø> (ø)	⬆️
pennylane_qiskit/__init__.py	100% <100%> (ø)	⬆️
pennylane_qiskit/ops.py	100% <100%> (ø)	⬆️
pennylane_qiskit/converter.py	94% <94%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4aee6c9...ca97645. Read the comment docs.

[josh146]

Looks great Antal, mostly minor comments. Great job with the really thorough tests!

[josh146]

By the way, it looks like the tests are failing on Python 3.5 - could be a dictionary ordering issue

[codecov-io]

Codecov Report

Merging #55 into master will not change coverage.
The diff coverage is 100%.

@@          Coverage Diff          @@
##           master    #55   +/-   ##
=====================================
  Coverage     100%   100%           
=====================================
  Files           8      9    +1     
  Lines         316    378   +62     
=====================================
+ Hits          316    378   +62

Impacted Files	Coverage Δ
pennylane_qiskit/qiskit_device.py	100% <ø> (ø)	⬆️
pennylane_qiskit/__init__.py	100% <100%> (ø)	⬆️
pennylane_qiskit/ops.py	100% <100%> (ø)	⬆️
pennylane_qiskit/converter.py	100% <100%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4aee6c9...4fbd76e. Read the comment docs.

[josh146]

By the way @antalszava, it looks like Codecov is failing because the lines 130-131 aren't being covered by the tests:

else:
    float_params = [float(param) for param in parameters]
    operation(*float_params, wires=operation_wires)

[antalszava]

Thank you for all the feedback, it was really valuable!

1. As far as the changes in PennyLane are concerned, I would envision something like the following in __init__.py (similar to how the device entrypoint is treated):

def load(name: str, quantum_circuit_object):
    """load(name, quantum_circuit_object)
    Load a plugin :func:`~.load` which can then be used to convert objects of
    quantum circuits from other frameworks.

    Args:
        name (str): the name of the plugin to convert from
        quantum_circuit_object: the quantum circuit that will be converted
            to a PennyLane template

    Returns:
        _function: the PennyLane template created from the quantum circuit
            object
    """

    if name in plugin_converters:

        # loads the plugin load function
        plugin_converter = plugin_converters[name].load()

        # calls the load function of the converter on the quantum circuit object
        return plugin_converter(quantum_circuit_object)

    raise ValueError(
        "Converter does not exist. Make sure the required plugin is installed "
        "and supports conversion."
    )

Once this is defined, the following from_qasm can also be added

def from_qasm(quantum_circuit: str):
    """load(name, quantum_circuit)
    Loads quantum circuits defined in QASM by using the converter in the
    PennyLane-Qiskit plugin.

    Args:
        quantum_circuit (str): the name for the QASM file or string

    Returns:
        _function: the PennyLane template created from the quantum circuit
            object
    """
    return load('qasm', quantum_circuit)

In the latest version, the entrypoints were extended with:
'qasm = pennylane_qiskit:load_qasm',

Therefore, each plugin will be available through entrypoints and by passing their string as the name argument to load.

In order to make this seamless, I have created one single function for QASM and added a simple check.

Furthermore, the load function has been refactored, it has been split into smaller functions such that it is more readable.

[josh146]

Really good @antalszava! My only comments were with respect to the load function docstring, and the load_qasm file. I think it's always best to have an explicit function to load from files, as it can avoid unexpected behaviour (cf. json.load for files, json.loads for strings).

Happy to approve conditional on those small remaining comments.

[co9olguy]

Thanks @antalszava, looking good. I left only a couple very minor comments/questions/suggestions.