Refactor linear inversion state tomography and notebook #4

mpharrigan · 2019-01-08T21:48:13Z

fixes #7

Use pyquil operator expectation code
style improvements
pretty plots in example notebook
tests use a fixture to avoid re-sampling tomography data

TODO:

refactor acquire_data / results class
unshim MLE estimation
make MLE estimation work with expectation values rather than counts
swap in pyqvm in tests + more shots + tighter tolerance

Clean up by using `vec` and experiment settings in place of `n_qubit_pauli_basis`

Why am I getting transposed?

joshcombes · 2019-01-10T21:32:53Z

forest_qcvv/tomography.py

+    dim = 2 ** len(qubits)
+    rho = rho.reshape((dim, dim))
+    # TODO: how did this get transposed?
+    return rho.T


I think at one point (bitbucket) rho was vectorized, the transpose was possibly used then. The state rho is Hermitan so the dagger transpose should not change things but transpose by itself will.

can test with a simple 1 qubit state rho = (1/2) (x X + yY + z Z)

Solved this in e87e949 and a65c671 by using unvec!

Why didn't I think of this before??

mpharrigan · 2019-01-11T03:22:37Z

@joshcombes @kylegulshen

Y = 1j * PROJ_PLUS - 1j * PROJ_MINUS

is this correct?

np.round(Y, 2)
Out[4]: 
array([[0.-0.j, 0.+1.j],
       [0.+1.j, 0.-0.j]])

mpharrigan · 2019-01-11T03:46:17Z

In the interest of keeping this PR from becoming ginormous and forever-taking, I've shimmed over iterative_mle and estimate_variance to flip between the previous and current data structures. I'll open a follow-up ticket for refactoring that to use ExperimentResult objects directly, which should be done simultaneously with modifying it to use expectation values rather than projector-success-counts.

This PR depends on rigetti/pyquil#552

kylegulshen · 2019-01-11T19:43:25Z

@mpharrigan

Y = 1j * PROJ_PLUS - 1j * PROJ_MINUS

is this correct?

np.round(Y, 2)
Out[4]: 
array([[0.-0.j, 0.+1.j],
       [0.+1.j, 0.-0.j]])

If PROJ_PLUS is the plus eigenvector of X, then no this isn't correct. Notice Y isn't Hermitian so something must have gone wrong. I'm not sure what the context is to say more.

kylegulshen · 2019-01-11T20:08:56Z

examples/Hinton-Plots.ipynb

+    "Z_EFFECTS = [PROJ_ZERO, PROJ_ONE]\n",
+    "X_EFFECTS = [PROJ_PLUS, PROJ_MINUS]\n",
+    "X = PROJ_PLUS - PROJ_MINUS\n",
+    "Y = 1j * PROJ_PLUS - 1j * PROJ_MINUS\n",


Found the context. Notice that this would simply be i*X since X = ProjPlus - ProjMinus. The top right element of Y should be negative and the bottom left positive.

kylegulshen · 2019-01-11T20:12:45Z

examples/Hinton-Plots.ipynb

+    "PLUS = np.array([[1], [1]]) / np.sqrt(2)\n",
+    "PROJ_PLUS = PLUS @ PLUS.T.conj()\n",
+    "PROJ_MINUS = ID - PROJ_PLUS\n",
+    "Z_EFFECTS = [PROJ_ZERO, PROJ_ONE]\n",


Should proj_one be negative here? That would make it more analogous to X_EFFECTS because in the x basis the X_EFFECTS are PROJ_ZERO, -PROJ_ONE

Sorry, I don't understand here. xeffects are proj_plus and proj_minus, not proj_zero and -proj_one.

by analogy, there's no minus sign in either z_effects or x_effects as currently written

kylegulshen · 2019-01-11T20:13:53Z

examples/Hinton-Plots.ipynb

+    "P10 = np.kron(PROJ_ONE, PROJ_ZERO)\n",
+    "P11 = np.kron(PROJ_ONE, PROJ_ONE)\n",
+    "ID_2Q = P00 + P01 + P10 + P11\n",
+    "ZZ_EFFECTS = [P00, P01, P10, P11]"


Same thing here, I would think P01 and P10 should be negative.

kylegulshen · 2019-01-11T20:19:20Z

examples/state_tomography.ipynb

+    "\n",
+    "qc = get_qc('2q-pyqvm')\n",
+    "from forest_qcvv.compilation import basic_compile\n",
+    "qc.compiler.quil_to_native_quil = basic_compile\n",


It would be good to explain what this achieves with a comment.

kylegulshen · 2019-01-11T20:32:40Z

examples/Hinton-Plots.ipynb

+    "\n",
+    "# Two qubit defs\n",
+    "P00 = np.kron(PROJ_ZERO, PROJ_ZERO)\n",
+    "P01 = np.kron(PROJ_ZERO, PROJ_ONE)\n",


We need to be careful about standardizing qubit ordering in tensor products. I had done everything this way for process tomography before noticing that state tomography was reversed. I think process tomography currently uses the convention opposite of the one used here. I prefer the convention here.

yeah this is a problem. With this refactor, I've given linear_inv_state_estimate an explicit argument: qubits and the resulting rho will be tensored according to this list

specifically for these, I'm going to leave them as-is. As you've probably discovered, this is just copied from test_state_tomography.py. We can clean those up in a follow-on pr

kylegulshen · 2019-01-11T20:46:07Z

forest_qcvv/tomography.py

+from pyquil.operator_estimation import ExperimentSetting, \
+    TomographyExperiment as PyQuilTomographyExperiment, ExperimentResult, measure_observables
+from pyquil.paulis import sI, sX, sY, sZ, PauliSum, PauliTerm
+from pyquil.unitary_tools import lifted_pauli


Is this in pyquil master?

no, it's in rigetti/pyquil#552

kylegulshen · 2019-01-11T20:49:58Z

forest_qcvv/tomography.py

@@ -100,6 +113,19 @@ class TomographyData:
    """number of shots used to calculate the `expectation`"""


+def shim_pyquil_results_to_TomographyData(program, qubits, results: List[ExperimentResult]):
+    return TomographyData(
+        in_ops=[r.setting.in_operator for r in results[1:]],


Not sure if significant, but in_ops is currently Optional[List[str]]

yeah yeah. The shimming is a little rough-and-ready, under the basis that it will become unnecessary after we refactor all the functions. In my tests, this wasn't an actual problem since in_ops is just a record of what was done and isn't used in the analysis routines. If it becomes a problem we can make the shim more robust by casting to strings

kylegulshen · 2019-01-11T20:56:33Z

forest_qcvv/tomography.py

-    :param operators: A list of ndarray-like objects.
-    :return: An ndarray that when applied to a vector of expectation values returns a vectorized
-    estimate of the state.
+    :param results: A tomographically complete list of results.


missing qubits in docstring

kylegulshen · 2019-01-11T21:01:20Z

tests/test_state_tomography.py

+Z_EFFECTS = [PROJ_ZERO, PROJ_ONE]
+X_EFFECTS = [PROJ_PLUS, PROJ_MINUS]
+X = PROJ_PLUS - PROJ_MINUS
+Y = 1j * PROJ_PLUS - 1j * PROJ_MINUS


Needs to be corrected as above, though not used.

kylegulshen · 2019-01-11T21:03:56Z

tests/test_state_tomography.py

+PLUS = np.array([[1], [1]]) / np.sqrt(2)
+PROJ_PLUS = PLUS @ PLUS.T.conj()
+PROJ_MINUS = ID - PROJ_PLUS
+Z_EFFECTS = [PROJ_ZERO, PROJ_ONE]


This seems to match what was there before but I'm not convinced it's right

yeah all this stuff I took verbatim from what was there before. Will have to do some investigating on who wrote it to make sure it's correct

mpharrigan · 2019-01-16T22:32:03Z

addressed non-(defining projectors correctly) review items

mpharrigan · 2019-01-17T22:49:19Z

gonna merge and open a follow-on ticket

Matthew Harrigan added 12 commits January 8, 2019 13:43

Refactor state tomography

f818f06

Hinton in its own notebook

8fdcad2

Port linear_inv_state_estimate to accept results

bbc7015

Clean up by using `vec` and experiment settings in place of `n_qubit_pauli_basis`

Update docstring

931c566

Light formatting

fcbd048

Linear inversion tests

b10b712

Why am I getting transposed?

Generate state tomography test

f8786b0

Generate 2q state tomography test

2e562ac

Use unvec

e87e949

Merge remote-tracking branch 'origin/master' into state-tomo-refactor

b4ee278

Extend unvec() to accept 1d arrays too

a65c671

VAR global

11f5c71

joshcombes reviewed Jan 10, 2019

View reviewed changes

Matthew Harrigan added 5 commits January 10, 2019 17:12

Port iterative mle estimation by shimming

2eda16d

Update notebook to not reference explicit shimming

e015828

tol to something that actually works

a156ccc

Hedged single-q

1ebc2d5

Use a test fixture

5adc4a8

Why didn't I think of this before??

Matthew Harrigan added 3 commits January 10, 2019 19:24

Shim (and unshim) estimate_variance

900bd91

Shim different return values in estimate_variance

7eff4b3

Fix tests / unitarity

ea05d14

mpharrigan changed the title ~~[WIP] Refactor state tomography~~ Refactor state tomography Jan 11, 2019

mpharrigan requested review from kylegulshen and joshcombes January 11, 2019 03:46

mpharrigan mentioned this pull request Jan 11, 2019

Refactor iterative mle state tomography to use pyquil operator estimation #13

Closed

mpharrigan changed the title ~~Refactor state tomography~~ Refactor linear inversion state tomography and notebook Jan 11, 2019

kylegulshen suggested changes Jan 11, 2019

View reviewed changes

Matthew Harrigan added 4 commits January 16, 2019 15:07

Merge branch 'master' into state-tomo-refactor

1c65e21

motivate basic_compile

cd55d77

functools

e8121b5

Document qubits argument

b8ec409

mpharrigan merged commit 2156653 into master Jan 17, 2019

mpharrigan deleted the state-tomo-refactor branch January 17, 2019 22:49

This was referenced Jan 17, 2019

Definitions of gates and operators wrong in test_state_tomography #16

Closed

Process tomography using pyQuil's measure_observables #17

Merged

mpharrigan mentioned this pull request Feb 6, 2019

Refactor DFE to use pyquil measure_observables #33

Merged

karalekas added this to the v0.3 milestone Mar 20, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refactor linear inversion state tomography and notebook #4

Refactor linear inversion state tomography and notebook #4

mpharrigan commented Jan 8, 2019 •

edited

Loading

joshcombes Jan 10, 2019

mpharrigan Jan 10, 2019

mpharrigan commented Jan 11, 2019 •

edited

Loading

mpharrigan commented Jan 11, 2019

kylegulshen commented Jan 11, 2019

kylegulshen Jan 11, 2019

kylegulshen Jan 11, 2019

mpharrigan Jan 16, 2019

kylegulshen Jan 11, 2019

kylegulshen Jan 11, 2019

kylegulshen Jan 11, 2019

mpharrigan Jan 16, 2019

mpharrigan Jan 16, 2019

kylegulshen Jan 11, 2019

mpharrigan Jan 16, 2019

kylegulshen Jan 11, 2019

mpharrigan Jan 16, 2019

kylegulshen Jan 11, 2019

kylegulshen Jan 11, 2019

kylegulshen Jan 11, 2019

mpharrigan Jan 16, 2019

mpharrigan commented Jan 16, 2019

mpharrigan commented Jan 17, 2019

Refactor linear inversion state tomography and notebook #4

Refactor linear inversion state tomography and notebook #4

Conversation

mpharrigan commented Jan 8, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mpharrigan commented Jan 11, 2019 • edited Loading

mpharrigan commented Jan 11, 2019

kylegulshen commented Jan 11, 2019

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

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

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

mpharrigan commented Jan 16, 2019

mpharrigan commented Jan 17, 2019

mpharrigan commented Jan 8, 2019 •

edited

Loading

mpharrigan commented Jan 11, 2019 •

edited

Loading