Allowing arbitrary initialization of the input nodes

[mgarnier59]

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.

[shinich1]

@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?

[mgarnier59]

@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.

[shinich1]

@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.

[shinich1]

~~@mgarnier59 @thierry-martinez have you checked the examples run without errors?~~

[thierry-martinez]

@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]

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

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 86.31579% with 26 lines in your changes missing coverage. Please review.

Project coverage is 72.26%. Comparing base (30a3f6b) to head (19dfd8a).

Files	Patch %	Lines
graphix/sim/tensornet.py	45.83%	13 Missing :warning:
graphix/sim/statevec.py	93.22%	4 Missing :warning:
graphix/linalg_validations.py	40.00%	3 Missing :warning:
graphix/sim/density_matrix.py	93.02%	3 Missing :warning:
graphix/states.py	90.62%	3 Missing :warning:

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #135      +/-   ##
==========================================
+ Coverage   71.84%   72.26%   +0.41%     
==========================================
  Files          30       32       +2     
  Lines        5359     5458      +99     
==========================================
+ Hits         3850     3944      +94     
- Misses       1509     1514       +5     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[shinich1]

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

[thierry-martinez]

Thank you very much for your review, @EarlMilktea. I believe I addressed or commented all your comments. Do you mind to check if everything is ok for you now? Thank you again!

[EarlMilktea]

Thank you for your suggestions/fixes, @thierry-martinez ! Could you resolve my new comment on conftest.py ?

[thierry-martinez]

@EarlMilktea , I fixedconftest.py in 55331b5 according to your comments. Thanks!

[shinich1]

@EarlMilktea , I fixedconftest.py in 55331b5 according to your comments. Thanks!

Hi @thierry-martinez , it looks like the recent changes you mention on this page are on separate branch, e.g. 55331b5, and is not showing up on the diff of this PR. could you take a look and push to this branch? After that, this PR should be ready for merge.

[thierry-martinez]

@EarlMilktea , I fixedconftest.py in 55331b5 according to your comments. Thanks!

Hi @thierry-martinez , it looks like the recent changes you mention on this page are on separate branch, e.g. 55331b5, and is not showing up on the diff of this PR. could you take a look and push to this branch? After that, this PR should be ready for merge.

Sorry, I didn't push on the right repo! Fixed.

[shinich1]

@EarlMilktea please check that your comments are resolved, and approve if so. From my side, this is nearly ready except the check related to TN backend above.

[EarlMilktea]

@thierry-martinez

I still can't see some of your changes correctly (such as SV_Data -> SVData).

Could you verify again that you're pushing to the right repo.?

[thierry-martinez]

@thierry-martinez

I still can't see some of your changes correctly (such as SV_Data -> SVData).

Could you verify again that you're pushing to the right repo.?

All references to SV_Data have been renamed into Data in cabd019, except one occurrence in the doc-comment of the Statevec class: fixed in 20d0237.

[thierry-martinez]

@EarlMilktea please check that your comments are resolved, and approve if so. From my side, this is nearly ready except the check related to TN backend above.

Fixed in 05aa7c9.

[EarlMilktea]

@thierry-martinez

Even though I still see some small problems regarding type hints, it would not affect the functionality itself.

LGTM!

@shinich1

I will submit a new PR to thoroughly fix type-related problems.

[thierry-martinez]

Thank you very much! Merged.