Update tutorials on calibration #4258
Conversation
commit d006a19 Author: Ryan LaRose <[email protected]> Date: Sun Jun 20 21:38:58 2021 -0400 Format. commit 4c28929 Author: Ryan LaRose <[email protected]> Date: Sun Jun 20 21:23:58 2021 -0400 Temp remove from book.yaml. commit 519d0be Author: Ryan LaRose <[email protected]> Date: Sun Jun 20 21:15:04 2021 -0400 Update advanced calibration tutorial.
rmlarose
requested review from
cduck,
vtomole,
wcourtney and
a team
as code owners
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| @@ -182,14 +182,20 @@ upper_tabs: | |||
| - heading: "Google hardware" | |||
| - title: "Getting started with QCS" | |||
| path: /cirq/tutorials/google/start | |||
| - title: "Floquet calibration" | |||
| path: /cirq/tutorials/google/floquet | |||
There was a problem hiding this comment.
I just tweeted a link to this page to go along with today's Nature publication. In general, we should be careful about URL backwards compatibility. Would it be possible to give the new "Floquet calibration: Example and benchmark" the /floquet URL? Does docsite have the ability to alias pages from multiple endpoints? @lamberta
There was a problem hiding this comment.
Yes, maintaining URL stability is important, particularly when SEO kicks in and you want to move pages.
You can't alias a page to two separate URLs, but you can add a redirect to point to the new page. Add an entry in the _redirects.yaml file located in the root of your docsite server directory.
There was a problem hiding this comment.
Indeed thanks! I'll add a redirect, and, more importantly, I'll like the tweet :)
There was a problem hiding this comment.
@rmlarose Did you add the redirect already? I can't see any changes to docs/_redirects.yaml (assuming that's where it's supposed to go).
| "id": "Y6Wh1qtGMghL" | ||
| }, | ||
| "outputs": [], | ||
| "source": [ |
There was a problem hiding this comment.
this inspired me to finally open #4260
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
"are tools to characterize sqrt(iswap) gates" --> "a family of gates including sqrt(iswap)" The family is represented by cirq.PhasedFSimGate blah blah. sqrt(iswap) corresponds to (-pi/4, 0, 0, 0, 0)
"insertion of Z phases" -> "insertion of Z gates" (or "ZPowGates" or "RzGates")
"decreasing the effects of two sources of coherent errors:"
Reply via ReviewNB
There was a problem hiding this comment.
Done. Reads a lot better and is more precise, thanks!
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
kindof confusing that theta and phi are set to "False" below if we say we can get estimates for all angles except chi.
It could be mildly helpful to put the latin-alphabet-spelling (chi) after the symbol so people can relate it to the code. It looks like an X
Reply via ReviewNB
There was a problem hiding this comment.
Set all to False execpt for chi, + added the latin spelling of chi in the text.
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
classical simulation -> classical processing
"Unlike floquet calibration, XEB requires significant classical processing to characterize angles"
Reply via ReviewNB
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
people are going to complain that this is slow and slows down notebook testing. The other notebooks loosen the tolerance parameters for fitting so the optimization finishes faster. But of course we don't want to set bad defaults that people will blindly copy.
Did @dabacon's scheme to regex replace parameters for notebook testing get merged? If so; we need to have replacement rules for this notebook
Reply via ReviewNB
There was a problem hiding this comment.
It did get merged, see the .tst files next to the notebooks.
There was a problem hiding this comment.
Yes, calibration_api.tst reduces the cycle depths to speed up testing here.
| @@ -0,0 +1,609 @@ | |||
| { | |||
There was a problem hiding this comment.
There was a problem hiding this comment.
The function name is make_zeta_chi_gamma_compensation_for_moments, which kind of implies that chi can be compensated. So it's a little confusing either way, but I agree your suggestion to say chi cannot be compensated is less confusing.
There was a problem hiding this comment.
Fixed. Also fixed some left headings in the table not being bold.
| @@ -39,7 +39,7 @@ | |||
| "id": "AF-CFuvzPx4x" | |||
There was a problem hiding this comment.
I know this isn't changed in the current PR, but:
this circuit diagram wraps really badly on my machine because of the use of print because you want to set the qubit order. If you just have the circuit as the last thing in the cell, jupyter will print its text representation in a cell with a horizontal scroll bar. I don't know if there's any fix here
Reply via ReviewNB
There was a problem hiding this comment.
Thanks for pointing out, I also don't know of a fix. At least it appears nicely on the website with a horizontal scroll bar.
Worth noting: Not specifying a qubit order makes the circuit much harder to understand here, even if it displays nicely.
| @@ -0,0 +1,674 @@ | |||
| { | |||
There was a problem hiding this comment.
this cell doesn't work as written and requires an actual processor id. We should 1) open a bug to provide mock engine calibrations 2) in the mean time, wrap this in a conditional to skip if processor_id isn't supplied
Reply via ReviewNB
There was a problem hiding this comment.
Agree on (1). For (2), I think this tutorial will look strange saying "look at calibration and pick good qubits" then not showing any calibration data. Do you think its sufficient to just save outputs for now, then follow-up after being unblocked by (1)?
There was a problem hiding this comment.
I agree that we can open a bug and fix this in a follow-up PR. For now, let's use the saved outputs.
| @@ -0,0 +1,674 @@ | |||
| { | |||
There was a problem hiding this comment.
this doesn't work as written without a processor_id. Why not set up sampler to optionally be a coherent-error sampler like in the floquet notebook?
Reply via ReviewNB
There was a problem hiding this comment.
Because the first step is looking at calibration to pick qubits - which can't be done with a coherent-error sampler AFAIK - this notebook really requires a processor_id. So the strategy was to just save all outputs.
| @@ -0,0 +1,674 @@ | |||
| { | |||
There was a problem hiding this comment.
|
Very cool stuff! I think the url stability thing is the most important plus some nits |
| @@ -0,0 +1,674 @@ | |||
| { | |||
There was a problem hiding this comment.
Line #11. calibration.heatmap(metric).plot(ax=axes[i] if len(metrics) > 1 else axes);
If it's useful here, I think it might be worth showing off the heatmap feature which allows you to specify a different scaling (eg: LogNorm, PowerNorm etc.) for the mapping of colors. For Eg:
heatmap.plot(ax, collection_options = {'norm' : mpl.colors.LogNorm()}) could be useful to have better distinction among qubits with low error rates. This is also something which people use internally (eg: Wojtek had reached out to me asking for this feature).
Reply via ReviewNB
|
I'm happy to see the changes in this PR. All the content seem to be cleaner and more accessible now! |
For posterity: |
balopat
added
the
automerge
CirqBot
added
the
front_of_queue_automerge
CirqBot
removed
automerge
Thanks!! |
Closes quantumlib#3933 and updates calibration tutorials. - quantumlib@825d29a repurposes the old "advanced calibration tutorial" to an example only using XEB calibration. - quantumlib@31df17a updates the Floquet calibration tutorial. - quantumlib@1e0c941 adds an overview tutorial on calibration, showing basic usage for both Floquet and XEB. Note: This is the same as quantumlib#4226 but with an easier-to-review commit history as requested in quantumlib#4226 (comment).
Closes #3933 and updates calibration tutorials.
Note: This is the same as #4226 but with an easier-to-review commit history as requested in #4226 (comment).