Automatically determine interface from constructed quantum script

[AlbertMitjans]

The QNode class now accepts an auto interface, which automatically detects the interface of the given input.

dev = qml.device("default.qubit", wires=2)

@qml.qnode(dev, interface="auto")
def circuit(weight):
    qml.RX(weight[0], wires=0)
    qml.RY(weight[1], wires=1)
    return qml.expval(qml.PauliZ(0))

interface_tensors = [
    np.array([0, 1]),
    torch.Tensor([0, 1]),
    tf.Variable([0, 1], dtype=float),
    jnp.array([0, 1]),
    [0, 1],
]

for tensor in interface_tensors:
    res = circuit(weight=tensor)
    print(res, type(res))

Prints

1.0 <class 'pennylane.numpy.tensor.tensor'>
tensor(1.0000, dtype=torch.float64) <class 'torch.Tensor'>
tf.Tensor(0.9999999896704281, shape=(), dtype=float64) <class 'tensorflow.python.framework.ops.EagerTensor'>
1.0 <class 'jaxlib.xla_extension.DeviceArray'>
1.0 <class 'pennylane.numpy.tensor.tensor'>

[codecov]

Codecov Report

Merging #3132 (964da3d) into master (afb2070) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #3132   +/-   ##
=======================================
  Coverage   99.69%   99.69%           
=======================================
  Files         276      276           
  Lines       24273    24292   +19     
=======================================
+ Hits        24200    24219   +19     
  Misses         73       73           

Impacted Files	Coverage Δ
pennylane/interfaces/execution.py	100.00% <100.00%> (ø)
pennylane/qnode.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[rmoyard]

Hi @AlbertMitjans, great changes! One thing that should be done is adding those changes to _execute_new in the execution.py file. Also do you think we could change the default interface to auto?

[AlbertMitjans]

One thing that should be done is adding those changes to _execute_new in the execution.py file.

We are calling _execute_new with the updated interface, thus I believe we don't need to add any logic there.

Also do you think we could change the default interface to auto?

@albi3ro mentioned that we should first try this out as an optional feature to make sure it works correctly. If it does we can make it the default.

[rmoyard]

@AlbertMitjans Thanks for the second point! Our methodology for the return type project is to duplicate the function (one public and one private). Right at the beginning, the public one calls the private one. The private one should be able to replace the public one without having to double check if something is missing (when we will deprecate it). So in this case I would suggest to move the code bloc after the qml.active_return and add the bloc in the function _execute_new.

[antalszava]

As per Romain's find, will just double-check that the new return-type pipeline works as expected.

[AlbertMitjans]

@AlbertMitjans Thanks for the second point! Our methodology for the return type project is to duplicate the function (one public and one private). Right at the beginning, the public one calls the private one. The private one should be able to replace the public one without having to double check if something is missing (when we will deprecate it). So in this case I would suggest to move the code bloc after the qml.active_return and add the bloc in the function _execute_new.

Oh OK! I did what you suggested. Thanks for this catch! 🎉

[rmoyard]

Thanks @AlbertMitjans ! 💯 You can add ["autograd", "auto"] in the test file test_qnode_autograd_new.py and test_autograd_new.py, it should solve the coverage issues.

[AlbertMitjans]

You can add ["autograd", "auto"] in the test file test_qnode_autograd_new.py and test_autograd_new.py, it should solve the coverage issues.

Added! @rmoyard

[albi3ro]

The docstring for QNode and execute should be updated to include the option "auto".

[AlbertMitjans]

The docstring for QNode and execute should be updated to include the option "auto".

Done!

[albi3ro]

Glad to have this! 👍 Looks good to me.

[antalszava]

Looks good! 🎉 Awesome feature, great work 🙂 😍