Define cirq.STATE_VECTOR_LIKE #2376

Strilanc · 2019-10-18T22:11:44Z

Add list-of-qid-values to the supported STATE_VECTOR_LIKE types
Add tensor-of-amplitudes to the supported STATE_VECTOR_LIKE types
Add non-numpy sequences to supported STATE_VECTOR_LIKE types
Infer values-vs-tensor-vs-vector using shape, fail when ambiguous (can only occur for non-standard shapes involving qudits of dimension 1)
Make num_qubits argument of to_valid_state_vector optional (if qid_shape specified)

- Add list-of-qid-values to the supported STATE_VECTOR_LIKE types - Add tensor-of-amplitudes to the supported STATE_VECTOR_LIKE types - Add non-numpy sequences to supported STATE_VECTOR_LIKE types - Infer values-vs-tensor-vs-vector using shape, fail when ambiguous (can only occur for non-standard shapes involving qudits of dimension 1) - Make `num_qubits` argument of `to_valid_state_vector` optional (if `qid_shape` specified)

# Conflicts: # cirq/sim/wave_function_simulator.py

kevinsung · 2019-10-18T22:19:20Z

cirq/sim/wave_function.py

+    # Full big-endian computational basis state index.
+    int,
+    # Per-qudit computational basis values.
+    Sequence[Union[int, bool]],


Since bools are ints, we can just say int instead of the union.

In [1]: isinstance(bool, int) Out[1]: False

>>> isinstance(True, int) True

>>> issubclass(bool, int) True

Hahaha, I made a dumb mistake and @cduck showed me the error in my ways. isinstance(False, int) returns true.

I do think that there's some user benefit to listing bool explicitly, but done.

yeah these type annotations are for the user so there's no harm in being explicit

cduck

I didn't review this thoroughly but I don't see any issues handling qudits.

cirq/sim/wave_function.py

cduck · 2019-10-18T22:29:21Z

cirq/sim/wave_function.py

+                                qid_shape: Tuple[int, ...],
+                                dtype: Type[np.number], atol) -> np.ndarray:
+
+    if isinstance(state_like, int):


In this case I really do want it to be an integer and not a generic number such as a float. I suppose numpy ints would also be acceptable, though.

cduck · 2019-10-18T22:36:13Z

cirq/sim/wave_function.py

+        if len(qid_shape) == prod:
+            if (not isinstance(state_like, np.ndarray) or
+                    state_like.dtype.kind != 'c'):
+                raise ValueError(
+                    'Because len(qid_shape) == product(qid_shape), it is '
+                    'ambiguous whether or not the given `state_like` is a '
+                    'state vector or a list of computational basis values for '
+                    'the qudits. In this situation you are required to pass '
+                    'in a state vector that is a numpy array with a complex '
+                    'dtype.')


This is worrying. As a user, I would it expect it to work for all input sizes if it works for most input sizes.

Because 2**n > n for all n >= 0, you never see this unless you have qudits with dimension 1. I think this is a strange enough corner case that we don't have to ensure it works perfectly.

We could reduce the number of relevant cases by checking if the array has non-integer entries in it (in which case it must be a state vector) or if it has integers summing to more than one (in which case it must be a qudit value list). The only truly ambiguous case is something like shape=(2, 2, 1, 1), value=[1, 0, 0, 0] where the value is one-hot and has an ambiguous length.

tfw your qudits have dimension 1

cduck · 2019-10-18T22:37:29Z

cirq/sim/wave_function.py

+    # Full big-endian computational basis state index.
+    int,
+    # Per-qudit computational basis values.
+    Sequence[Union[int, bool]],


>>> isinstance(True, int) True

>>> issubclass(bool, int) True

cduck · 2019-10-18T22:39:21Z

cirq/sim/wave_function.py

+                f'qid_shape={qid_shape!r}\n'
+                f'state={state!r}\n')
+
+    if state.dtype.kind[0] not in '?bBiu':


Is there a better way to check this?

I'm honestly not sure. Multiply it by 0, add 0.1, and see if it's still 0?

>>> issubclass(np.int8, np.integer) True >>> issubclass(np.float64, np.integer) False # But >>> issubclass(np.bool, np.integer) False >>> issubclass(np.bool, np.bool) True

https://stackoverflow.com/questions/934616/how-do-i-find-out-if-a-numpy-array-contains-integers

Checking against numbers.Integral should verify that the variable obeys integer arithmetic.

>>> issubclass(int, numbers.Integral) True >>> issubclass(bool, numbers.Integral) True >>> issubclass(np.int8, numbers.Integral) True >>> issubclass(np.int16, numbers.Integral) True >>> issubclass(np.bool, numbers.Integral) True >>> issubclass(float, numbers.Integral) False >>> issubclass(np.float64, numbers.Integral) False

There seems to be something tricky going on. Although np.bool is integral, the dtype type of an array initialized to have dtype np.bool is not:

>>> issubclass(np.bool, numbers.Integral) True

>>> issubclass(np.array([1, 2, 3], dtype=np.bool).dtype.type, numbers.Integral) False

>>> issubclass(np.array([1, 2, 3], dtype=np.int32).dtype.type, numbers.Integral) True

The trickery appears to be that when you ask for an array of np.bool you don't get an array of np.bool. Instead you get an array of np.bool_ which inherits from np.generic instead of np.int.

>>> test_array = np.array([1, 2, 3], dtype=np.bool) >>> test_array.dtype.type == np.bool False >>> test_array.dtype.type <class 'numpy.bool_'> >>> test_array.dtype.type.__bases__ (<class 'numpy.generic'>,) >>> np.bool.__bases__ (<class 'int'>,) >>> test_array.dtype.type == np.bool_ True

np.bool_s do not behave like numbers with respect to arithmetic since they do not inherit from np.number. I don't know of a way for a numpy array of boolean values to have a dtype other than np.bool_. Whenever you index such an array, you will get a np.bool_ object. In order to make a numpy array give you python bools it seems that you need to use the item function.

You will probably need to add a check for np.bool_ in addition to the numbers.Integral check.
if not (isinstance(state.dtype, np.bool_) or issubclass(state.dtype, numbers.Integral)):

If the array has at least one element, an equivalent check would be given by the below code since the item function would return a python bool which is a numbers.Integral.
if not isinstance(state.item(0), numbers.Integral):

As written, if someone passed in an empty array for state and an empty tuple for qid_shape, the output of this function will be a Scalar of type dtype. Is this desired?

_qudit_values_to_state_tensor(state=np.empty(shape=(0),dtype=np.int64), qid_shape=tuple(), dtype=np.int64) array(1) _qudit_values_to_state_tensor(state=np.empty(shape=(0),dtype=np.int64), qid_shape=tuple(), dtype=np.bool) array(True)

mpharrigan

looks comprehensive

mpharrigan · 2019-11-06T00:15:52Z

cirq/circuits/circuit.py

@@ -1410,7 +1410,7 @@ def unitary(self,

    def final_wavefunction(
            self,
-            initial_state: Union[int, np.ndarray] = 0,
+            initial_state: 'cirq.STATE_VECTOR_LIKE' = 0,


I mean ... this really exacerbates the "wavefunction vs. state vector" terminology question

You have a point.

mpharrigan · 2019-11-06T00:17:34Z

cirq/sim/wave_function.py

+        if len(qid_shape) == prod:
+            if (not isinstance(state_like, np.ndarray) or
+                    state_like.dtype.kind != 'c'):
+                raise ValueError(
+                    'Because len(qid_shape) == product(qid_shape), it is '
+                    'ambiguous whether or not the given `state_like` is a '
+                    'state vector or a list of computational basis values for '
+                    'the qudits. In this situation you are required to pass '
+                    'in a state vector that is a numpy array with a complex '
+                    'dtype.')


tfw your qudits have dimension 1

Strilanc requested review from dabacon and cduck October 18, 2019 22:11

googlebot added the cla: yes Makes googlebot stop complaining. label Oct 18, 2019

Merge branch 'master' of github.com:quantumlib/Cirq into initial_state

8fd3f0f

# Conflicts: # cirq/sim/wave_function_simulator.py

kevinsung reviewed Oct 18, 2019

View reviewed changes

cduck reviewed Oct 18, 2019

View reviewed changes

Strilanc added 2 commits October 18, 2019 15:58

tests, lint, types, coverage

5999865

tweaks

1d1839d

Strilanc added the Ready for Re-Review For when reviewers take their time. label Oct 29, 2019

Strilanc requested a review from mpharrigan October 29, 2019 21:51

mpharrigan approved these changes Nov 6, 2019

View reviewed changes

Strilanc mentioned this pull request Nov 6, 2019

Add cirq.final_density_matrix utility method #2487

Merged

Strilanc added automerge Tells CirqBot to sync and merge this PR. (If it's running.) and removed Ready for Re-Review For when reviewers take their time. labels Nov 6, 2019

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Nov 6, 2019

Merge branch 'master' into initial_state

7765b01

CirqBot merged commit 58a794f into master Nov 6, 2019

CirqBot removed the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Nov 6, 2019

CirqBot deleted the initial_state branch November 6, 2019 01:10

CirqBot removed the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Nov 6, 2019

viathor mentioned this pull request Oct 15, 2021

Consolidate quantum state representations in cirq #4582

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Define cirq.STATE_VECTOR_LIKE #2376

Define cirq.STATE_VECTOR_LIKE #2376

Strilanc commented Oct 18, 2019

kevinsung Oct 18, 2019

Strilanc Oct 18, 2019

cduck Oct 18, 2019 •

edited

Loading

Strilanc Oct 18, 2019

Strilanc Oct 18, 2019

mpharrigan Nov 6, 2019

cduck left a comment

cduck Oct 18, 2019

Strilanc Oct 18, 2019

cduck Oct 18, 2019

Strilanc Oct 18, 2019

mpharrigan Nov 6, 2019

cduck Oct 18, 2019 •

edited

Loading

cduck Oct 18, 2019

Strilanc Oct 18, 2019

cduck Oct 18, 2019 •

edited

Loading

c-poole Oct 21, 2019

Strilanc Oct 21, 2019

c-poole Oct 21, 2019 •

edited

Loading

c-poole Oct 21, 2019

mpharrigan left a comment

mpharrigan Nov 6, 2019

Strilanc Nov 6, 2019

mpharrigan Nov 6, 2019

Define cirq.STATE_VECTOR_LIKE #2376

Define cirq.STATE_VECTOR_LIKE #2376

Conversation

Strilanc commented Oct 18, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cduck Oct 18, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cduck left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cduck Oct 18, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cduck Oct 18, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

c-poole Oct 21, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mpharrigan left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cduck Oct 18, 2019 •

edited

Loading

cduck Oct 18, 2019 •

edited

Loading

cduck Oct 18, 2019 •

edited

Loading

c-poole Oct 21, 2019 •

edited

Loading