Allowing arbitrary initialization of the input nodes #135

mgarnier59 · 2024-05-03T13:05:47Z

Allowing arbitrary intialization of the input nodes for statevector and density matrix backends + tests.

Several format of initialization are supported (Statevec or DensityMatrix objects, list of newly-defined State objects, numerical data).
Main change is that basic states (|0>, |+>, ...) are now in states.BasicStates and no longer in ops.py.

Indeed, the attribute Statevec.Nqubit leads to problems that have to be fixed

some test fixed, some other not (related to DM and TN backend)

Add test to check simulation at the circuit level is the same as at pattern level in the statevec backend; DM to come

graphix/states.py

.pre-commit-config.yaml

.vscode/settings.json

graphix/ops.py

graphix/pattern.py

graphix/simulator.py

tests/test_statevec.py

graphix/sim/density_matrix.py

graphix/sim/statevec.py

tests/test_statevec_backend.py

shinich1 · 2024-05-08T14:21:40Z

tests/test_statevec_backend.py

@@ -29,8 +32,10 @@ def test_measurement_into_each_XYZ_basis(self):
        n = 3
        k = 0
        # for measurement into |-> returns [[0, 0], ..., [0, 0]] (whose norm is zero)
-        for state in [States.plus, States.zero, States.one, States.iplus, States.iminus]:
-            m_op = np.outer(state, state.T.conjugate())
+        # NOTE weird choice (MINUS is orthogonal to PLUS so zero)


tests/test_pattern.py

graphix/states.py

thierry-martinez · 2024-05-13T15:12:26Z

@mgarnier59 @thierry-martinez how do the performance of simulation compare with original implementation? I presume no significant difference since it's only modifying the initialization part?

We didn't investigate it. What's more it provides more feature compared to the previous versions. I tend to think that global performance issues will be adressed later.

I believe we should at least test it's not surprisingly slow by 2x or more, or have completely different scaling as a function of problem size, which indicates something is wrong with the code. if it's up to a few tens of % (which I believe is most likely the case) indeed it can be safely addressed later.

I obtained these benchmarks (input-init is the version I just pushed, which includes the last changes in master; input-init^1 is the version before the merge). Performances are quite close indeed.

+-------------------------------------------------------------------------------------+
|                       statement                      |master|input-init|input-init^1|
+------------------------------------------------------+------+----------+------------+
|                    pytest.main([])                   | 16.28|   16.41  |    14.67   |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=15, depth=1)| 3.12 |   3.30   |    3.27    |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=16, depth=1)| 9.69 |   8.24   |    8.45    |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=17, depth=1)| 17.61|   18.91  |    17.72   |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=18, depth=1)| 39.14|   41.15  |    42.17   |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=19, depth=1)| 72.78|   76.26  |    77.80   |
+------------------------------------------------------+------+----------+------------+
|random_circuits(seed=42, times=20, nqubit=20, depth=1)|155.43|  162.73  |   163.46   |
+-------------------------------------------------------------------------------------+

shinich1 · 2024-05-13T15:19:24Z

@thierry-martinez Thanks a lot, that looks great!

codecov · 2024-05-13T15:27:00Z

Codecov Report

Attention: Patch coverage is 85.88235% with 24 lines in your changes are missing coverage. Please review.

Project coverage is 72.17%. Comparing base (dd15e5d) to head (8836a61).

❗ Current head 8836a61 differs from pull request most recent head 6ebfcbe

Please upload reports for the commit 6ebfcbe to get more accurate results.

Files	Patch %	Lines
graphix/sim/tensornet.py	38.09%	13 Missing ⚠️
graphix/sim/statevec.py	92.98%	4 Missing ⚠️
graphix/states.py	87.09%	4 Missing ⚠️
graphix/sim/density_matrix.py	92.68%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #135      +/-   ##
==========================================
+ Coverage   71.75%   72.17%   +0.42%     
==========================================
  Files          30       32       +2     
  Lines        5353     5441      +88     
==========================================
+ Hits         3841     3927      +86     
- Misses       1512     1514       +2

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Reported by Maxime Garnier. TeamGraphix@f377001#r142237724 The test for Pauli preprocessing in density_matrix.py should be ``` pattern._pauli_preprocessed and input_state != graphix.states.BasicStates.PLUS ``` without `not`. This bug wasn't caught by `pytest` because the density matrix backend was never tested with Pauli-preprocessed patterns. I think a natural place to add such a test is in `test_pattern.py`, by "parametrizing" `test_pauli_measurment` with `backend`. This was not enough because `test_pauli_measurment` was defined twice in the same file and the second definition was hiding the first one. Therefore, this commit: - rename the first `test_pauli_measurment` into `test_pauli_measurement_random_circuit` (fixing the typo by the way); - "parametrize" `test_pauli_measurement_random_circuit` with `backend`, testing `"statevector"` and `"densitymatrix"`; - a TODO is added for tensor network backend, since the default `graph_prep` mode (`auto`) selects the mode `parallel` despite this mode is incompatible with non-standardize patterns. I propose to leave that for another PR.

shinich1 · 2024-05-23T17:04:17Z

@EarlMilktea could you take a quick look at the changes?

EarlMilktea

memo: duplicated comments are added only once (ex. use from __future__ import annotations)

EarlMilktea · 2024-05-24T01:23:51Z

graphix/random_objects.py

@@ -1,5 +1,6 @@
+from typing import Optional, Union


Please use from __future__ import annotations instead.

EarlMilktea · 2024-05-24T01:27:22Z

graphix/random_objects.py

@@ -68,7 +69,7 @@ def rand_dm(dim: int, rank: int = None, dm_dtype=True) -> DensityMatrix:
        return dm


-def rand_gauss_cpx_mat(dim: int, sig: float = 1 / np.sqrt(2)) -> npt.NDArray:
+def rand_gauss_cpx_mat(dim: int, sig: float = 1 / np.sqrt(2)) -> np.ndarray:


I prefer to use npt.NDArray unless absolutely necessary (ex. runtime checking of return type) because we cannot specify element type.

EarlMilktea · 2024-05-24T01:28:22Z

graphix/sim/density_matrix.py

-from graphix.sim.statevec import CNOT_TENSOR, CZ_TENSOR, SWAP_TENSOR, meas_op
+from graphix.sim.statevec import CNOT_TENSOR, CZ_TENSOR, SWAP_TENSOR, Statevec
+
+Data = typing.Union[


typing.Union is deprecated.

EarlMilktea · 2024-05-24T01:28:38Z

graphix/sim/density_matrix.py

+    graphix.states.State,
+    "DensityMatrix",
+    Statevec,
+    typing.Iterable[graphix.states.State],


Please import from collections.abc instead.

EarlMilktea · 2024-05-24T01:29:21Z

graphix/sim/density_matrix.py

+            return
+        if isinstance(data, typing.Iterable):
+            input_list = list(data)
+            if len(input_list) != 0:


No need to compute len.

EarlMilktea · 2024-05-24T02:04:44Z

graphix/states.py

+    """
+
+    @abc.abstractmethod
+    def get_statevector(self) -> np.ndarray:


npt.NDArray if possible.

EarlMilktea · 2024-05-24T02:04:59Z

graphix/states.py

+    plane: graphix.pauli.Plane
+    angle: float
+
+    def __repr__(self):


Please specify return type explicitly.

EarlMilktea · 2024-05-24T02:06:28Z

tests/test_density_matrix.py

+        nqb = fx_rng.integers(2, 5)
+        print(f"nqb is {nqb}")
+        rand_angles = fx_rng.random(nqb) * 2 * np.pi
+        rand_planes = fx_rng.choice(np.array([i for i in graphix.pauli.Plane]), nqb)


Is it necessary to cast to np.ndarray before choice?

EarlMilktea · 2024-05-24T02:07:59Z

tests/test_statevec.py

+from graphix.states import BasicStates, PlanarState
+
+
+class TestStatevec(unittest.TestCase):


Please do not use unittest.

EarlMilktea · 2024-05-24T02:11:38Z

tests/test_pattern.py

+# for testing with arbitrary inputs
+# SV and DM backend
+class TestPatternSim:
+    def test_sv_sim(self, fx_rng: Generator, nqb, rand_circ):


Please specify -> None explicitly.

mgarnier59 and others added 19 commits May 3, 2024 14:55

update

89288f8

up

a6da6d3

statevector constructor for input initialization

187e650

fixed TN files for BasicStates modification

8d7f18b

add tests for statevec + some black

cbedf08

modifying Backends and simulator, add tests

818844d

updates

46bdfa5

cancelled modification in tensor.

86ccb49

Indeed, the attribute Statevec.Nqubit leads to problems that have to be fixed

Update statevec.py

c2487a4

commented all statevec.Nqubit attributes

882cac7

some test fixed, some other not (related to DM and TN backend)

update statevec.__init__. DM ongoing

3415369

Constructors for statevec and DM

ab6781d

Fix tests for input states

2c0c2ee

implem arbitrary inputs for circuit statevector

506f9f6

update test_pattern.py

10169b4

Add test to check simulation at the circuit level is the same as at pattern level in the statevec backend; DM to come

Circuit measure handling

db6d54e

Some simplifications

25aa338

Update CHANGELOG.md, isort, black

7d488c3

Use typing.Union instead of | to please Python 3.8/3.9

209e77c

mgarnier59 requested a review from shinich1 May 3, 2024 14:28

shinich1 reviewed May 4, 2024

View reviewed changes

graphix/states.py Outdated Show resolved Hide resolved

shinich1 linked an issue May 6, 2024 that may be closed by this pull request

arbitrary initial state #53

Open

mgarnier59 added 2 commits May 6, 2024 11:40

add docs + cleanup

714dc12

Update statevec.py

4aafe65

shinich1 reviewed May 7, 2024

View reviewed changes

graphix/states.py Outdated Show resolved Hide resolved

shinich1 reviewed May 7, 2024

View reviewed changes

.pre-commit-config.yaml Outdated Show resolved Hide resolved

shinich1 reviewed May 7, 2024

View reviewed changes

.vscode/settings.json Outdated Show resolved Hide resolved

shinich1 reviewed May 7, 2024

View reviewed changes

graphix/ops.py Outdated Show resolved Hide resolved

shinich1 reviewed May 7, 2024

View reviewed changes

graphix/pattern.py Outdated Show resolved Hide resolved

shinich1 reviewed May 7, 2024

View reviewed changes

graphix/pattern.py Outdated Show resolved Hide resolved