Simplify single shot protocols by andrea-pasquale · Pull Request #1278 · qiboteam/qibocal

andrea-pasquale · 2025-10-30T08:00:16Z

Closes #1241
In this PR I'm planning to simplify the code related to the single experiments.
Some modifications include:

Use LDA to compute angle threshold
Drop outdated parameters related to ML classifiers
Simplify plot using directly plotly
Add confusion matrix
Propagate changes to qutrit classification
Plot qutrit fit
Update readout optimization protocols

codecov · 2025-10-31T11:10:12Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.75%. Comparing base (ed74278) to head (77c7329).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1278      +/-   ##
==========================================
- Coverage   96.86%   96.75%   -0.11%     
==========================================
  Files         133      131       -2     
  Lines       10457    10139     -318     
==========================================
- Hits        10129     9810     -319     
- Misses        328      329       +1

Flag	Coverage Δ
unittests	`96.75% <100.00%> (-0.11%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/qibocal/cli/report.py	`100.00% <100.00%> (ø)`
...qibocal/protocols/classification/classification.py	`100.00% <100.00%> (+5.45%)`	⬆️
.../protocols/classification/qutrit_classification.py	`100.00% <100.00%> (ø)`
src/qibocal/protocols/classification/utils.py	`100.00% <100.00%> (ø)`
...cols/qubit_spectroscopies/qubit_spectroscopy_ef.py	`100.00% <100.00%> (ø)`
src/qibocal/protocols/rabi/ef.py	`100.00% <ø> (ø)`
...ocal/protocols/readout/readout_characterization.py	`100.00% <100.00%> (ø)`
...qibocal/protocols/readout_optimization/__init__.py	`100.00% <100.00%> (ø)`
...ibocal/protocols/readout_optimization/amplitude.py	`100.00% <100.00%> (ø)`
...ibocal/protocols/readout_optimization/frequency.py	`100.00% <100.00%> (ø)`
... and 2 more

... and 5 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

andrea-pasquale · 2025-11-04T11:45:49Z

I believe that this should be ready to review now.
Here is how the readout optimization protocols are looking now: http://login.qrccluster.com:9000/dvp18iWdQSOweRyNVWh63g==/
Here is the "new" single shot and readout characterization: http://login.qrccluster.com:9000/AsPZh9W8QeWQEAxUbaTSCA==/

In the last report you can see how by lowering the readout amplitude we get better QND reducing the excitations to state 2.

for more information, see https://pre-commit.ci

jevillegasdatTII · 2026-01-13T13:01:53Z

Is it possible that some of the data is mislabeled in the readout characterization?

Here the distribution of the last measurement of state |0> is almost identical to the first one of state |1> and very differnt form the first measurement. I get the same with other qubits in the same QPU

alecandido · 2026-01-13T14:20:31Z

Is it possible that some of the data is mislabeled in the readout characterization?

Everything is possible :)

Here the distribution of the last measurement of state |0> is almost identical to the first one of state |1> and very differnt form the first measurement.

I agree, it would make sense that if you say you are measuring the same thing, they should be in the same position.

@lballerio will certainly take a look into this (as soon as he's done with some of the other open PRs 🙄)

Add average states markers in ssro plots and change the scaling so that the clusters are more visible.

for more information, see https://pre-commit.ci

jevillegasdatTII · 2026-01-14T11:24:51Z

The plots in the SSRO stretch ugly with the report window; in #1334 I added an anchor to the x axis that helps the plot be "prettier" and added the average state markers.

This:

Instead of:

jevillegasdatTII · 2026-01-14T14:21:59Z

In the amplitude optimization the swept value of amplitude appears to be relative to the amplitude in the platform and not absolute. For example running between 0.05 and 1 I get thius

But SSRO with pulse amplitude = 0.95 (best ro amplitude) returns:

Assignment Fidelity | 0.767

While SSRO with pulse amplitde = 0.15 returns

Assignment Fidelity | 0.966

Clearly very different to those computed in the amplitude optimization plot. I am only guessing the swept value is re;lative and not absolute because if I set the amplitude to 1 and run the sweep, I get

I checked the qibocal implementation and it does look like is sweeping the amplitude parameter correctly (in fact I don't know if relative sweepers for amplitude are still supported)

lballerio · 2026-01-15T07:20:14Z

Hi @jevillegasdatTII, did you open an issue?

modified: src/qibocal/protocols/classification/classification.py

jevillegasd

im ok with the changes as they are and it can be worked around the smaller fixes commented (sweeps being realitve) afterwards.

for more information, see https://pre-commit.ci

alecandido · 2026-02-23T06:50:31Z

@lballerio is about to double-check this with QPU 118, as soon as he gets suitable frequencies for the qubits.

Whenever he confirms as well, we can merge

lballerio

Here is my review about this PR...some comments are docs/ minor fixes, but I think some comments, especially about 1->2 transition experiments are more relevant.
However I briefly tested the experiments and they seem to work, but I would not merge now the PR.

lballerio · 2026-02-26T12:46:42Z

    html_list = []
    for figure in figures:
-        figure.write_html(buffer, include_plotlyjs=False, full_html=False)
+        figure.write_html(buffer, include_plotlyjs="cdn", full_html=False)


I want to link write_html help function and focus on the discussion about include_plotlyjs parameter:

write_html_api_reference

Long story short, this configuration of include_plotlyjs requires an active internet connection (except if cached).
Is it what we want?

Yes, that's reasonable. Otherwise, each report would be just 3 MB larger, only because of repeated content. Most of the time, you also need to be connected to the internet to retrieve these reports.

The only other interesting option would be directory, and self-host the JavaScript package.
This is fine, but requires more effort on our side. Also because of potential upgrades and compatibility concerns.

Until this becomes explicitly a problem, I would keep cdn.

lballerio · 2026-02-26T13:29:16Z

 ROC_LENGHT = 800
 ROC_WIDTH = 800
 DEFAULT_CLASSIFIER = "qubit_fit"


ROC_LENGHT, ROC_WIDTH and DEFAULT_CLASSIFIER are not used in this script nor in other files.

Suggested change

ROC_LENGHT = 800

ROC_WIDTH = 800

DEFAULT_CLASSIFIER = "qubit_fit"

They were, before this PR.

Since now they are not any longer used, it makes perfect sense to remove them. Good catch :)

lballerio · 2026-02-27T08:13:37Z

+                platform.channels[platform.qubits[q].drive].lo: {
+                    "frequency": platform.config(platform.qubits[q].drive).frequency
+                    + params.lo_offset * GHZ_TO_HZ
+                },


same as previous comment, here we are moving LO frequencies, don't we have to recalibrate mixers then?

Yes, we should: much better to drop this shift, and keep it manual (for the time being).

If strongly needed, we could automate this process later on, when we will have stabilized the mixer calibration API.

lballerio · 2026-02-27T08:24:19Z

I think here the file name and the protocol name is misleading.
I think ef.py is quite meaningless, and also I realized that this is a signal experiment, not a probability, but we export it as rabi_amplitude_ef, which is inconsistent with all other protocols where we can measure both in probability and signel+phase.

Additionally, I was wondering if it is worth to put rabi and rabi_ef all together in a single protocol and use a keyword to trigger either 0->1 or 1->2 transitions since the experiments are the same and rabi_ef code often recall rabi classes and fuctions.
Also might be worth to implement all Rabi experiments even for 1->2 transition (?)
The bottleneck might be that is quite boring and refactoring the code is time consuming.

I think ef.py is quite meaningless, and also I realized that this is a signal experiment, not a probability, but we export it as rabi_amplitude_ef, which is inconsistent with all other protocols where we can measure both in probability and signel+phase.

Agreed. We could call the file amplitude_signal_ef.py and the protocol rabi_amplitude_signal_ef.

Additionally, I was wondering if it is worth to put rabi and rabi_ef all together in a single protocol and use a keyword to trigger either 0->1 or 1->2 transitions since the experiments are the same and rabi_ef code often recall rabi classes and fuctions.
Also might be worth to implement all Rabi experiments even for 1->2 transition (?)
The bottleneck might be that is quite boring and refactoring the code is time consuming.

It could be worth, to deduplicate code. But let's not do it in this PR. You could open an issue about it.

lballerio · 2026-02-27T08:39:55Z

+                signal=result.T[0],
+                phase=result.T[1],


are we sure results are signal phase and magnitude and not I-Q components?

Actually, since they are coming directly from Qibolab, they are exactly IQ pairs.

This is a plain bug

lballerio · 2026-02-27T09:41:53Z

            qubit: float(platform.calibration.single_qubits[qubit].qubit.frequency_01)
            for qubit in targets


shouldn't this be the drive frequency instead?

this is what I was thinking about:

Suggested change

qubit: float(platform.calibration.single_qubits[qubit].qubit.frequency_01)

for qubit in targets

qubit: float(platform.config(platform.qubits[qubit].drive).frequency)

for qubit in targets

lballerio · 2026-02-27T09:56:29Z

+        # # mask values where fidelity is below 80%
+        # averaged_qnd[grids["fidelity"] < 0.8] = np.nan
+        # # exclude values where QND is larger than 1
+        # averaged_qnd[averaged_qnd > 1] = np.nan


indeed, in my last calibration with qpus 118 and 156 I had problems with the old implementation of this protocol since all fidelities where <80%, hence all the data were np.nan and the experiment failed. I think this mask is useless, or at least we can decrease the threshold to 50%.

Suggested change

# # mask values where fidelity is below 80%

# averaged_qnd[grids["fidelity"] < 0.8] = np.nan

# # exclude values where QND is larger than 1

# averaged_qnd[averaged_qnd > 1] = np.nan

No, I would not put it to 50%. I agree it is just useless.

If we want to consider taking a different point, we can engineer a suitable cost function (not necessarily linear).
Same thing for the plots: if we want to avoid losing details in the high-fidelity patch, we could even change the colorscale with a non-linear rescaling. Though, it is possibly not even needed.

alecandido

I mostly read @lballerio's comments, and the related code. However, I also had a quick look to some of the other changes.

All in all, conditional to the result to be functioning, we do not need a detailed review of all the code changes. Which are massive, mostly because of a huge refactoring (but you could assume the old code to be messy and unreliable - no need to spend ages navigating the diff).

However, as @lballerio already pointed out, there is at least one outright bug (if I'm not missing anything myself). And I would implement the small improvements he suggested, postponing the larger changes (possibly opening issues, if relevant).

alecandido · 2026-03-12T13:25:42Z

Btw, once the review is done, the PR needs to be rebased. Though it should be straightforward, since GitHub is not reporting any conflict

andrea-pasquale added this to the Protocols milestone Oct 30, 2025

andrea-pasquale self-assigned this Oct 30, 2025

alecandido changed the title ~~Simply single shot protocols~~ Simplify single shot protocols Oct 30, 2025

andrea-pasquale force-pushed the refactor-ssr branch from eae0fd8 to 9148d2d Compare October 30, 2025 14:37

andrea-pasquale mentioned this pull request Oct 31, 2025

sweepers for ro optimization #1277

Closed

alecandido linked an issue Oct 31, 2025 that may be closed by this pull request

sweepers for ro optimization #1277

Closed

andrea-pasquale marked this pull request as ready for review November 4, 2025 11:40

andrea-pasquale added the calibration label Nov 4, 2025

This was linked to issues Nov 14, 2025

Check why QND could be larger than 1 #1249

Open

Use average QND as figure of merit for selecting readout amplitude #1250

Open

Compute QND using the implementation in readout_characterization #1206

Open

andrea-pasquale and others added 17 commits November 26, 2025 08:30

refactor: Start simplyfing single shot

b844d20

refactor: Add fit to plot

ebf304c

refactor: Adding fidelities

1e15c6b

refactor: Adapt qutrit classification

348c5e5

refactor: Start refactoring ro optimization

12bb2aa

[pre-commit.ci] auto fixes from pre-commit.com hooks

a359925

for more information, see https://pre-commit.ci

refactor: Finish qutrit plot

ee2414c

fix: Minor fix in plot

9dead77

refactor: Modify readout optimization

69b3482

refactor: Modify readout optimization protocols in 1D

12e980f

fix: Tests passing

d818d0e

refactor: Update 2d readout optimization

07afb0e

refactor: Fix QND evaluation

ba11b36

doc: Update plots in documentation

4011bcc

style: Refactor plot

9f620e1

fix: Update qutrits as well

5743a65

refactor: Reduce report size

c107378

andrea-pasquale force-pushed the refactor-ssr branch from 027ae14 to c107378 Compare November 26, 2025 07:30

alecandido assigned lballerio and unassigned andrea-pasquale Jan 13, 2026

jevillegasd and others added 2 commits January 14, 2026 11:21

modified: src/qibocal/protocols/classification/classification.py

7c286d7

Add average states markers in ssro plots and change the scaling so that the clusters are more visible.

[pre-commit.ci] auto fixes from pre-commit.com hooks

2862569

for more information, see https://pre-commit.ci

lballerio mentioned this pull request Jan 15, 2026

SSRO: add average states markers in plots and change the scaling so that the clusters are more visible. #1334

Merged

jevillegasdatTII mentioned this pull request Jan 15, 2026

Amplitude Sweeps are relative #1339

Open

Merge pull request #1334 from qiboteam/refactor-ssr_update_ssr

22c2a40

modified: src/qibocal/protocols/classification/classification.py

jevillegasdatTII mentioned this pull request Jan 21, 2026

Plot 1D experimental data as scatter points #1293

Open

jevillegasdatTII requested a review from jevillegasd February 4, 2026 09:06

jevillegasd approved these changes Feb 12, 2026

View reviewed changes

jevillegasd and others added 2 commits February 18, 2026 12:09

Merge branch 'main' into refactor-ssr

46838a4

[pre-commit.ci] auto fixes from pre-commit.com hooks

77c7329

for more information, see https://pre-commit.ci

lballerio requested changes Feb 27, 2026

View reviewed changes

lballerio requested review from alecandido and jevillegasd February 27, 2026 09:59

alecandido requested changes Mar 12, 2026

View reviewed changes

alecandido linked an issue Apr 1, 2026 that may be closed by this pull request

Remove Qblox classifier #1030

Open

	ROC_LENGHT = 800
	ROC_WIDTH = 800
	DEFAULT_CLASSIFIER = "qubit_fit"

		qubit: float(platform.calibration.single_qubits[qubit].qubit.frequency_01)
		for qubit in targets

Uh oh!

Conversation

andrea-pasquale commented Oct 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

andrea-pasquale commented Nov 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jevillegasdatTII commented Jan 13, 2026

Uh oh!

alecandido commented Jan 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jevillegasdatTII commented Jan 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jevillegasdatTII commented Jan 14, 2026

Uh oh!

lballerio commented Jan 15, 2026

Uh oh!

jevillegasd left a comment

Choose a reason for hiding this comment

Uh oh!

alecandido commented Feb 23, 2026

Uh oh!

lballerio left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lballerio Feb 27, 2026 • edited by alecandido Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alecandido left a comment

Choose a reason for hiding this comment

Uh oh!

alecandido commented Mar 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

andrea-pasquale commented Oct 30, 2025 •

edited

Loading

codecov Bot commented Oct 31, 2025 •

edited

Loading

andrea-pasquale commented Nov 4, 2025 •

edited

Loading

alecandido commented Jan 13, 2026 •

edited

Loading

jevillegasdatTII commented Jan 14, 2026 •

edited

Loading

lballerio Feb 27, 2026 •

edited by alecandido

Loading