[TAPE] Incorporate the tape refactor into core

[josh146]

Context:

• This is the final tape refactor PR. In this PR, the many tape components, including interfaces, measurements, operations, QNodes, and the tapes, are moved out of the pennylane.beta, and better organized into a new pennylane.tape subpackage. We also provide helper functions here to make it easier to use the new tape core if wanted.

Description of the Change:

• All tape components were moved into the new pennylane.tape directory. The tests were similarly restructured.

• Documentation has been written, see it here: https://pennylane--823.org.readthedocs.build/en/823/code/qml_tape.html

• Three new mocks were added to the QuantumTape() context:

  # create mock measurement functions
  mocks += [mock.patch.object(qml, "expval", qml.tape.measure.expval)]
  mocks += [mock.patch.object(qml, "var", qml.tape.measure.var)]
  mocks += [mock.patch.object(qml, "sample", qml.tape.measure.sample)]
  mocks += [mock.patch.object(qml, "probs", qml.tape.measure.probs)]

  That is, within a tape context, calling qml.expval et al. will instead result in a call to the functions within the new qml.tape.measure module. After the context closes, these locations will resume pointing to the original core components.

• It is now possible to create a tape QNode by simply changing the qnode decorator to qml.tape.qnode:

  import pennylane as qml
  from pennylane import numpy as np
  dev = qml.device("default.qubit", wires=1)
  
  @qml.tape.qnode(dev)
  def circuit(x):
      qml.RX(x, wires=0)
      return qml.expval(qml.PauliZ(0))
  
  >>> circuit(0.5)
  0.8775825618903726

  This allows a single Python file to contain a mixture of old and new QNodes.

• Finally, two new convenience functions have been added; qml.enable_tape() and qml.disable_tape():

  dev = qml.device("default.qubit", wires=1)
  
  qml.enable_tape()
  
  @qml.qnode(dev)
  def circuit(x):
      qml.RX(x, wires=0)
      return qml.expval(qml.PauliZ(0))
  
  >>> print(circuit(0.5))
  0.8775825618903726
  >>> print(type(circuit))
  <class 'pennylane.tape.qnode.QNode'>
  >>> print(circuit.qtape)
  <AutogradQuantumTape: wires=[0], params=1>
  
  qml.disable_tape()
  
  @qml.qnode(dev)
  def circuit(x):
      qml.RX(x, wires=0)
      return qml.expval(qml.PauliZ(0))
  
  >>> print(circuit(0.5))
  0.8775825618903726
  >>> print(type(circuit))
  <class 'pennylane.interfaces.autograd.to_autograd.<locals>.AutogradQNode'>

  This allows entire scripts/files to be globally switched to using the new tape core, by simply adding qml.enable_tape() near the top. This will be useful with testing compatibility, and benchmarking.

• In tape mode, we allow qml.state() to be used. Outside of tape mode, an error will be raised.

Benefits: n/a

Possible Drawbacks:

In tape-mode, the QNode does not yet have feature-parity with the standard PennyLane QNode. Features currently not available in tape mode include:

• Circuit drawing and visualization

• Metric tensor computation

• The ability to automatically extract the layer structure of variational circuits

• Hashing and serialization.

• The qml.qnn module

These can all be done in subsequent PRs.

Related GitHub Issues: n/a

[codecov]

Codecov Report

Merging #823 into master will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #823      +/-   ##
==========================================
+ Coverage   91.05%   91.09%   +0.03%     
==========================================
  Files         129      130       +1     
  Lines        8678     8701      +23     
==========================================
+ Hits         7902     7926      +24     
+ Misses        776      775       -1     

Impacted Files	Coverage Δ
pennylane/tape/queuing.py	97.05% <ø> (ø)
pennylane/tape/tapes/reversible.py	100.00% <ø> (ø)
pennylane/__init__.py	63.52% <100.00%> (+0.87%)	⬆️
pennylane/tape/__init__.py	100.00% <100.00%> (ø)
pennylane/tape/circuit_graph.py	100.00% <100.00%> (ø)
pennylane/tape/interfaces/__init__.py	100.00% <100.00%> (ø)
pennylane/tape/interfaces/autograd.py	100.00% <100.00%> (ø)
pennylane/tape/interfaces/tf.py	96.22% <100.00%> (ø)
pennylane/tape/interfaces/torch.py	100.00% <100.00%> (ø)
pennylane/tape/measure.py	96.87% <100.00%> (ø)
... and 7 more

---

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 9ac47c8...8b40221. Read the comment docs.

[antalszava]

Hi @josh146 this looks good overall! Left some questions/suggestions, would be keen to hear your thoughts, but overall would be good from my side. Also, I think there might be a couple of places where a search and replace might be needed to cover the qml.variable, qml.various occurrences in the tests.

A thing I was wondering about is if we'd like to move tape out of beta though it would stay an experimental feature. 🤔

[josh146]

Also, I think there might be a couple of places where a search and replace might be needed to cover the qml.variable, qml.various occurrences in the tests.

Thanks for catching this Antal, I did use a lot of find and replace here 😬

A thing I was wondering about is if we'd like to move tape out of beta though it would stay an experimental feature. 🤔

It is definitely still an experimental feature. Perhaps we could raise a warning when the tape is enabled? Or would that be too annoying?

[antalszava]

Perhaps we could raise a warning when the tape is enabled? Or would that be too annoying?

Hmm, it could indeed be a bit repetitive each time. I'm sort of on the fence. If users get to know about the features brought with tape, they might already know that it's experimental, so it should be fine. Perhaps other reviewers would have further suggestions.

Thanks for the answers! Overall this looks good to me 😊

[trbromley]

Thanks @josh146, looks awesome!

I noticed a few spots in test docstrings where qml. had been added (my favourite was qml.various). I left (a few too many) suggestions correcting them - sorry for the spam! (Just noticed @antalszava pointed it out too)

[trbromley]

Minor thing: could a novice reader get confused at seeing another "tape"-like object and be mixed up between our tape and the TF tape? A remedy to this could be to use the autograd interface for this example.

[josh146]

Good point, I agree!

[josh146]

Thanks @trbromley and @antalszava! I've incorporated all your changes.

Apologies for the qml.variable and qml.various --- my text editors refactoring mode was... more zealous than I expected. I've gone through and undone any find-and-replaces inside docstrings and comments.

[mariaschuld]

Maybe check the sentence above again, is the . supposed to be a comma?

[mariaschuld]

Hope you had a good search & replace function there :)

[mariaschuld]

Really a great idea to have tape mode! We can now extend the tape with new feature, staying in the experimental mode.

I checked out the branch and played around a bit, works like a charm.

[co9olguy]

🍾