Update T1 experiment

[eliottrosenberg]

Makes two changes to the $T_1$ experiment in Cirq:

1. Changes the fit function to $f(t) = A e^{-t/T_1} + B$ instead of just $f(t) = e^{-t/T_1}$ to capture the effects of SPAM errors.
2. Makes the wait times logarithmically spaced instead of linearly spaced, which is more suitable since we expect an exponential decay.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.76%. Comparing base (bc76660) to head (3c107ec).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6487      +/-   ##
==========================================
- Coverage   97.76%   97.76%   -0.01%     
==========================================
  Files        1105     1105              
  Lines       94997    94999       +2     
==========================================
+ Hits        92871    92872       +1     
- Misses       2126     2127       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[eliottrosenberg]

@NoureldinYosri do you know why the Windows pytest is failing?

[NoureldinYosri]

@eliottrosenberg it's not clear to me what's happening with the windows test. I suggest adding some information to the assert statement to try to understand what's going on.

[NoureldinYosri]

can you change this line to

desired = ...
assert results == desired, f'{results.data=} {desired.data=}'

so that we get a more informative error message.

[eliottrosenberg]

I tried this, and it just outputs

>       assert results == desired, f'{results.data=} {desired.data=}'
E       AssertionError: results.data=   delay_ns  false_count  true_count
E         0       100           10           0
E         1       215           10           0
E         2       464           10           0
E         3      1000           10           0 desired.data=   delay_ns  false_count  true_count
E         0       100           10           0
E         1       215           10           0
E         2       464           10           0
E         3      1000           10           0

[eliottrosenberg]

I suspect it may have something to do with data types since I changed line 83 in t1_decay_experiment.py to np.logspace(np.log10(min_delay_nanos), np.log10(max_delay_nanos), num_points, dtype=int).

[NoureldinYosri]

looks like it's indeed a types issue specifically for the delay_ns column

>>> results.data.dtypes
delay_ns       float64
false_count      int64
true_count       int64
dtype: object
>>> desired.data.dtypes
delay_ns       int64
false_count    int64
true_count     int64
dtype: object

[eliottrosenberg]

That's pretty strange. When I run it on linux, delay_ns is int64, and in line 83 of t1_decay_experiment.py, I specifically ask for dtype=int. What do you recommend? Should I manually convert delay_ns to integers in the test? It seems like there is some deeper issue when running on windows.

[NoureldinYosri]

the result is computed by sampler.sample which is probably changing the dtype to float internally. I suggest that you ensure that the desired result is also float.

Why do you want to force delay_ns to be int?

[eliottrosenberg]

I changed it back to float

[NoureldinYosri]

same as above

[NoureldinYosri]

LGTM%nits

[NoureldinYosri]

use the original parameter to make it clear that this happens when no min_delay is supplied

if min_delay is None

[eliottrosenberg]

I want to keep it as is in case the user specifies 0 min delay.

[NoureldinYosri]

nit: you can assign directly

self.popt, _ = optimize.curve_fit(exp_decay, xs, probs, p0=[t1_guess, 1.0, 0.0])