Update format of the generator chain

[alex-simm]

What

Changes the form of the signal chain that the generator accepts into a dictionary. For each device, the dict specifies which devices are used as inputs, for example:

"d1": {
    "LO": [],
    "AWG": [],
    "DigitalToAnalog": ["AWG"],
    "Response": ["DigitalToAnalog"],
    "Mixer": ["LO", "Response"],
    "VoltsToHertz": ["Mixer"]
}

instead of
"d1": ["LO", "AWG", "DigitalToAnalog", "Response", "Mixer", "VoltsToHertz"]

Also, I added an optional callback to the generator which can be used for debugging during the signal generation.

Why

Closes #29

Previously, the order of the devices in the chain depended on the way the generator internally stored the signals on a stack. The dictionary can be created more intuitively and does not depend on the actual implementation of the generator.

How

The generator brings the directed graph into topological order. Then it iterates through the devices, only storing the signals as long as they are needed. This implementation should be able to handle any signal chain, even if the output of one device is used as input for several others. In case of a cycle, it will raise an error.

Remarks

I didn't test it with complicated signal chains, but some are covered by the unit tests.

Checklist

• Tests - Added unit tests for new code, regression tests for bugs and updated the integration tests if required
• Formatting & Linting - black and flake8 have been used to ensure styling guidelines are met
• Type Annotations - All new code has been type annotated in the function signatures using type hints
• Docstrings - Docstrings have been provided for functions in the numpydoc style
• Documentation - The tutorial style documentation has been updated to explain changes & new features
• Notebooks - Example notebooks have been updated to incorporate changes and new features
• Changelog - A short note on this PR has been added to the Upcoming Release section

[alex-simm]

So far, the code is ready and should work. Some of the tests are failing because I used a package from python 3.9. Do we need that backwards compatibility?

[lazyoracle]

So far, the code is ready and should work. Some of the tests are failing because I used a package from python 3.9. Do we need that backwards compatibility?

Yep, we only very recently moved to provide support for Python 3.9 and there is no plan to drop support for 3.6, 3.7 and 3.8 in the upcoming releases. So ideally we should not use any features that are Python 3.9 only

[lazyoracle]

A possible alternative to consider (which makes for a good choice despite being an external dependency) is networkx. It is widely used, stable, been around for a long time and is also used by other quantum programming toolchains (eg qiskit which also has their own re-implementation in rust retworkx which also has support for Python 3.6). I would suggest we use networkx and drop support for Python 3.6 instead of dropping all previous versions in favour of the still pretty new Python 3.9 which many libraries don't fully support yet.

[shaimach]

If all we need is a topological sort of the signal generation DAG, then we can implement it ourselves in < 1 day. If we need more significantly more extensive graph algorithms, then using an existing package makes sense.

…

On Sun, 1 Aug 2021 at 13:16, Anurag Saha Roy ***@***.***> wrote: A possible alternative to consider (which makes for a good choice despite being an external dependency) is networkx <https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.dag.topological_sort.html>. It is widely used, stable, been around for a long time and is also used by other quantum programming toolchains (eg qiskit which also has their own re-implementation in rust retworkx <https://github.com/Qiskit/retworkx>. I would suggest we use networkx and drop support for Python 3.6 instead of dropping all previous versions in favour of the still pretty new Python 3.9 which many libraries don't fully support yet. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#129 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAH3Q4LURLTT3QHICO73BSDT2UNGHANCNFSM5AUF3K5A> .

[CLAassistant]

All committers have signed the CLA.

[alex-simm]

I added a simple implementation of topological sorting. This should work without any dependence on python 3.9 now.

[lazyoracle]

Putting this on hold until #134 is fixed

[lazyoracle]

Updates to CHANGELOG should follow the style in 4d6c8e2. Sorry if this wasn't adequately clear in the CONTRIBUTING document. The changelog should be updating by adding new details about what was done without removing content from the top.

[codecov]

Codecov Report

Merging #129 (7757235) into dev (b0ce40c) will increase coverage by 0.53%.
The diff coverage is 89.58%.

@@            Coverage Diff             @@
##              dev     #129      +/-   ##
==========================================
+ Coverage   71.28%   71.81%   +0.53%     
==========================================
  Files          36       36              
  Lines        5237     5475     +238     
==========================================
+ Hits         3733     3932     +199     
- Misses       1504     1543      +39     

Impacted Files	Coverage Δ
c3/generator/generator.py	91.00% <89.58%> (-3.03%)	⬇️
c3/experiment.py	78.05% <0.00%> (-4.36%)	⬇️
c3/model.py	84.79% <0.00%> (-1.06%)	⬇️
c3/main.py	57.30% <0.00%> (+0.48%)	⬆️
c3/utils/tf_utils.py	48.73% <0.00%> (+0.63%)	⬆️
c3/libraries/fidelities.py	34.10% <0.00%> (+12.44%)	⬆️
c3/libraries/propagation.py	57.09% <0.00%> (+26.76%)	⬆️

---

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 b0ce40c...7757235. Read the comment docs.

[nwittler]

Good to go.