Bosonic backend nongaussian #541

elib20 · 2021-02-17T03:57:52Z

Context: The bosonic backend simulates states as linear combinations of Gaussians in phase space

Description of the Change:

Adds preparation backend and frontend methods for GKP, cat and Fock states to the bosonic backend
Adds the run_prog and init_circuit methods to bosonic/backend.py which are the engines for running sf Programs in the bosonic backend
Adds the bosonic compiler
Adds tests for these new backend methods

Benefits:

Simulation of bosonic qubits

Possible Drawbacks:

Related GitHub Issues:

Co-Authored-By: ilan-tz <57886357+ilan-tz@users.noreply.github.com> Co-Authored-By: Nicolas Quesada <991946+nquesada@users.noreply.github.com>

- Adds preparation methods for GKP, cat and Fock states - Adds the run_prog and init_circuit methods which are the engines for running sf Programs in the bosonic backend - Adds the frontend ops for calling these methods Co-Authored-By: ilan-tz <57886357+ilan-tz@users.noreply.github.com> Co-Authored-By: Nicolas Quesada <991946+nquesada@users.noreply.github.com> Co-Authored-By: GDauphinais <48553380+GDauphinais@users.noreply.github.com>

Tests for the bosonic backend

codecov · 2021-02-17T04:00:42Z

Codecov Report

Merging #541 (1356c90) into master (3bd5318) will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #541      +/-   ##
==========================================
+ Coverage   98.15%   98.19%   +0.04%     
==========================================
  Files          75       76       +1     
  Lines        8115     8327     +212     
==========================================
+ Hits         7965     8177     +212     
  Misses        150      150

Impacted Files	Coverage Δ
strawberryfields/backends/base.py	`88.98% <100.00%> (ø)`
...trawberryfields/backends/bosonicbackend/backend.py	`100.00% <100.00%> (ø)`
strawberryfields/compilers/__init__.py	`100.00% <100.00%> (ø)`
strawberryfields/compilers/bosonic.py	`100.00% <100.00%> (ø)`
strawberryfields/ops.py	`98.88% <100.00%> (+0.03%)`	⬆️

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 3bd5318...1356c90. Read the comment docs.

strawberryfields/backends/bosonicbackend/backend.py

strawberryfields/ops.py

- Ensures that new modes can be added in program contexts, and that state preparation for the new mode can still be run - Adds an error if more than one non-Gaussian state prep is attempted in the same mode - Adds unit tests for the changes

thisac

Mainly checked code quality, and had a few comments. Most fairly minor, or me just noticing small things that could be slightly improved. 🙂

strawberryfields/backends/bosonicbackend/backend.py

thisac · 2021-02-18T20:33:43Z

strawberryfields/backends/bosonicbackend/backend.py

+            GKP,
+            Ket,
+            MbSgate,
+            _New_modes,


This class should probably be updated to _NewModes to follow class naming convention.

I'm just using what it was already called in the ops.py file.

thisac · 2021-02-18T20:38:35Z

strawberryfields/backends/bosonicbackend/backend.py

+            _New_modes,
+        )
+
+        # Initialize the circuit. This applies all non-Gaussian state-prep


(very minor) I'm noticing that the comments sometimes are capitalized, sometimes not, and sometime use punctuation, sometimes not. It might be nice to use the same rules as we do with the docstrings: capitalize + punctuation when more than one sentence, otherwise all lowercase. Might just be me being a bit too picky here though. 😆

tests/backend/test_bosonic_backend.py

tests/backend/test_mbsqueeze_operation.py

strawberryfields/backends/bosonicbackend/backend.py

strawberryfields/ops.py

tests/backend/test_bosonic_backend.py

Co-authored-by: Theodor <theodor@xanadu.ai>

antalszava

Hi @elib20, left some comments. Looking awesome! 💯 😍

Nothing major, mostly suggestions related to code quality. Some cases where there still seems to be ambiguity is where there are multiple for/if nested blocks. In such cases may be worth trying to simplify them if easily feasible.

strawberryfields/compilers/bosonic.py

strawberryfields/backends/bosonicbackend/backend.py

antalszava · 2021-02-18T21:16:37Z

strawberryfields/backends/bosonicbackend/backend.py

+        from strawberryfields.ops import (
+            Bosonic,
+            Catstate,
+            DensityMatrix,
+            Fock,
+            GKP,
+            Ket,
+            MbSgate,
+            _New_modes,
+        )


A personal note here (which is also relevant for other parts of the code base): the absolute import import strawberryfields.ops as ops and prefixing each gate with ops. should help with organizing the import to the top level without causing cyclic import issues. No need to change though, just a side note.

This unfortunately didn't work :(

strawberryfields/backends/bosonicbackend/backend.py

antalszava · 2021-02-18T23:34:54Z

strawberryfields/backends/bosonicbackend/backend.py

+                pars = cmd.op.p
+                for reg in labels:
+                    # All the possible preparations should go in this loop
+                    if isinstance(cmd.op, Bosonic):


Would consider moving the branching if statement structure into a separate method if feasible (maybe not due to the new modes branch).

That would also help with:

removing the too many branches linter check,

shortening this method,

including the comments as the docstring of the new method.

Yes I think the _New_modes messes up that approach.

antalszava · 2021-02-18T23:37:38Z

strawberryfields/backends/bosonicbackend/backend.py

+                if val is not None:
+                    for i, r in enumerate(cmd.reg):
+                        if r.ind not in self.ancillae_samples_dict.keys():
+                            self.ancillae_samples_dict[r.ind] = [val[:, i]]


Suggest checking if using a defaultdict could help to decrease the nested nature of this part of the code (that would allow having each member to be initialized as an empty list by default, though suggest checking the quirks of it).

Let's keep this on the list of things to investigate in the next round of improvements.

strawberryfields/backends/bosonicbackend/backend.py

antalszava · 2021-02-18T23:44:16Z

strawberryfields/backends/bosonicbackend/backend.py

@@ -54,6 +69,205 @@ def __init__(self):
        self._supported["mixed_states"] = True
        self._init_modes = None
        self.circuit = None
+        self.ancillae_samples_dict = {}
+
+    def run_prog(self, prog, **kwargs):


How come this method is defined? Mostly curious about how it is going to be used with the engine as other backends do not seem to define such a method.

@antalszava : the idea here is that we want to have a custom run_prog that checks the type of initialization that happens for the different modes. If you know in advance how the different modes are initialized then you can directly allocate the required arrays. Otherwise, everytime a new mode is parsed one would have to generate a new array with the extra dimensions required for the particular preparation. For example if you have a circuit with two cats, this method will recognize that it needs 4x4 = 16 coefficients, means and covariances. If you instead used the usual run_prog you would first allocate 4 coefficients, means and covariances and then later when you read the preparation of the second cat you would have to generate the new arrays having a first index running from 0 to 16-1.

It was decided early on to build a custom engine for the bosonic backend, which is essentially what this method is, mainly since we had some complications with the initializations for each mode. You will see the changes made to the engine file in the next PR. I'm tagging @ilan-tz since he can give a more thorough answer.

Yep. The original backends (Fock, Gaussian, and TF) all are run 'step-by-step'. The LocalEngine will initialize the backend with the number of modes and some other program independent properties, and will then separately loop over the program operations and call the corresponding backend methods.

This does not work well for the bosonic backend; information about the program is needed at initialization, but the current LocalEngine does not provide! In fact, the LocalEngine tries to do too much by itself.

A better approach is how the RemoteEngine works - it passes the program directly to the backend, waits for the backend to interpret and execute it, and receives the results. This is the approach we have decided to go with for the Bosonic backend.

Sidenote: we did look into attempting to modify the local engine to provide this information to the backend, but you end up going down a rabbit hole of needing to patch the local engine, patch the existing backends, and still have the whole thing look like a bit of a hotfix.

strawberryfields/ops.py

strawberryfields/backends/bosonicbackend/backend.py

strawberryfields/compilers/bosonic.py

Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com>

strawberryfields/compilers/bosonic.py

antalszava

@elib20 Looks good to me, amazing! 🥇 😍 So good to see this addition. 🙂

strawberryfields/backends/bosonicbackend/backend.py

tests/backend/test_bosonic_backend.py

strawberryfields/ops.py

Co-authored-by: antalszava <antalszava@gmail.com>

elib20 and others added 5 commits February 16, 2021 22:39

Adds bosonic compiler

c531327

Co-Authored-By: ilan-tz <57886357+ilan-tz@users.noreply.github.com> Co-Authored-By: Nicolas Quesada <991946+nquesada@users.noreply.github.com>

Bosonic backend tests

16b9d7f

Tests for the bosonic backend

ran black

2309738

Tidying docstrings

5012f85

elib20 requested a review from nquesada February 17, 2021 03:57

elib20 added 10 commits February 16, 2021 23:31

Codefactor changes

bec7a44

Ran black and tidied test

181d03b

codefactor changes

e1844ed

minor bug fixes

e4329e7

codefactor and docstrings

a723e55

Change importing

f5e03f3

ran black + pylint disable

2be5bfe

docstrings and pytest marking

156d24f

ran black

6a84a24

Fix hbar in test

c9fc8aa

elib20 commented Feb 17, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

elib20 commented Feb 17, 2021

View reviewed changes

strawberryfields/ops.py Outdated Show resolved Hide resolved

nquesada requested review from antalszava and removed request for antalszava February 17, 2021 13:31

elib20 added 2 commits February 17, 2021 11:48

Fix add new mode and noninitial state prep

f0f6f45

- Ensures that new modes can be added in program contexts, and that state preparation for the new mode can still be run - Adds an error if more than one non-Gaussian state prep is attempted in the same mode - Adds unit tests for the changes

Disable too many branches for init_circuit

6f56639

elib20 requested review from antalszava and thisac February 17, 2021 17:14

thisac reviewed Feb 18, 2021

View reviewed changes

elib20 commented Feb 18, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

elib20 commented Feb 18, 2021

View reviewed changes

strawberryfields/ops.py Outdated Show resolved Hide resolved

elib20 commented Feb 18, 2021

View reviewed changes

tests/backend/test_bosonic_backend.py Outdated Show resolved Hide resolved

Apply suggestions from code review

59d78a9

Co-authored-by: Theodor <theodor@xanadu.ai>

Ran black

7ba79ee

antalszava reviewed Feb 18, 2021

View reviewed changes

josh146 reviewed Feb 19, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

josh146 reviewed Feb 19, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

josh146 reviewed Feb 19, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

elib20 commented Feb 19, 2021

View reviewed changes

strawberryfields/backends/bosonicbackend/backend.py Outdated Show resolved Hide resolved

elib20 commented Feb 19, 2021

View reviewed changes

strawberryfields/compilers/bosonic.py Outdated Show resolved Hide resolved

elib20 and others added 8 commits February 19, 2021 17:17

Apply suggestions from code review

6796383

Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com>

ran black

711d82f

Minor changes based on code review

a777590

Fixes circular import

102cfc2

Minor updates based on code review

fe72df1

Docstring and comment tidying

039cef1

Minor edit for array construction

777f832

Modifying code readability

b6fc10b

elib20 requested review from thisac and antalszava February 22, 2021 21:54

elib20 commented Feb 23, 2021

View reviewed changes

strawberryfields/compilers/bosonic.py Outdated Show resolved Hide resolved

elib20 added 3 commits February 22, 2021 19:05

rename MSgate in compiler

4687361

Fix heterodyne return shape

b5acfe1

Add MSgate to list of channels

fa3e14d

antalszava approved these changes Feb 23, 2021

View reviewed changes

Apply suggestions from code review

1356c90

Co-authored-by: antalszava <antalszava@gmail.com>

elib20 merged commit 9ee646f into master Feb 23, 2021

elib20 deleted the bosonic_backend_nongaussian branch February 23, 2021 18:17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Bosonic backend nongaussian #541

Bosonic backend nongaussian #541

elib20 commented Feb 17, 2021

codecov bot commented Feb 17, 2021 •

edited

Loading

thisac left a comment

thisac Feb 18, 2021

elib20 Feb 22, 2021

thisac Feb 18, 2021

antalszava left a comment

antalszava Feb 18, 2021

elib20 Feb 22, 2021

antalszava Feb 18, 2021

elib20 Feb 22, 2021

antalszava Feb 18, 2021

elib20 Feb 22, 2021

antalszava Feb 18, 2021

nquesada Feb 19, 2021

elib20 Feb 19, 2021

josh146 Feb 19, 2021

josh146 Feb 19, 2021

antalszava left a comment

Bosonic backend nongaussian #541

Bosonic backend nongaussian #541

Conversation

elib20 commented Feb 17, 2021

codecov bot commented Feb 17, 2021 • edited Loading

Codecov Report

thisac 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

antalszava 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

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

antalszava left a comment

Choose a reason for hiding this comment

codecov bot commented Feb 17, 2021 •

edited

Loading