Quantify dataset - advanced examples#
See also
The complete source code of this tutorial can be found in
Here we will explore a few advanced usages of the Quantify dataset and how it can accommodate them.
Imports and auxiliary utilities
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import xarray as xr
from rich import pretty
from quantify_core.analysis.calibration import rotate_to_calibrated_axis
from quantify_core.analysis.fitting_models import exp_decay_func
from quantify_core.data import handling as dh
from quantify_core.utilities import dataset_examples
from quantify_core.utilities.dataset_examples import (
mk_nested_mc_dataset,
mk_shots_from_probabilities,
mk_surface7_cyles_dataset,
)
from quantify_core.utilities.examples_support import (
mk_iq_shots,
round_trip_dataset,
)
from quantify_core.utilities.inspect_utils import display_source_code
pretty.install()
dh.set_datadir(Path.home() / "quantify-data") # change me!
Dataset for an “unstructured” experiment#
Let’s take consider a Surface Code experiment, in particular the one portrayed in Fig. 4b from one of the papers from DiCarlo Lab [Marques et al., 2021].
For simplicity, we will not use exactly the same schedule, because what matters here are the measurements. It is difficult to deal with the results of these measurements because we have a few repeating cycles followed by a final measurement that leaves the overall dataset “unstructured”.
Source code for generating this schedule and visualizing it
If you want to create and visualize the schedule above using quantify-scheduler
,
you can use this code:
from quantify_scheduler import Schedule
from quantify_scheduler.operations.gate_library import CZ, Y90, Measure, Reset, X
from quantify_scheduler.visualization.circuit_diagram import circuit_diagram_matplotlib
def mk_surface7_sched(num_cycles: int = 3):
"""Generates a schedule with some of the feature of a Surface 7 experiment as
portrayed in Fig. 4b of :cite:`marques_logical_qubit_2021`.
Parameters
----------
num_cycles
The number of times to repeat the main cycle.
Returns
-------
:
A schedule similar to a Surface 7 dance.
"""
sched = Schedule("S7 dance")
q_d1, q_d2, q_d3, q_d4 = [f"D{i}" for i in range(1, 5)]
q_a1, q_a2, q_a3 = [f"A{i}" for i in range(1, 4)]
all_qubits = q_d1, q_d2, q_d3, q_d4, q_a1, q_a2, q_a3
sched.add(Reset(*all_qubits))
for cycle in range(num_cycles):
sched.add(Y90(q_d1))
for qubit in [q_d2, q_d3, q_d4]:
sched.add(Y90(qubit), ref_pt="start", rel_time=0)
sched.add(Y90(q_a2), ref_pt="start", rel_time=0)
for qubit in [q_d2, q_d1, q_d4, q_d3]:
sched.add(CZ(qC=qubit, qT=q_a2))
sched.add(Y90(q_d1))
for qubit in [q_d2, q_d3, q_d4]:
sched.add(Y90(qubit), ref_pt="start", rel_time=0)
sched.add(Y90(q_a2), ref_pt="start", rel_time=0)
sched.add(Y90(q_a1), ref_pt="end", rel_time=0)
sched.add(Y90(q_a3), ref_pt="start", rel_time=0)
sched.add(CZ(qC=q_d1, qT=q_a1))
sched.add(CZ(qC=q_d2, qT=q_a3))
sched.add(CZ(qC=q_d3, qT=q_a1))
sched.add(CZ(qC=q_d4, qT=q_a3))
sched.add(Y90(q_a1), ref_pt="end", rel_time=0)
sched.add(Y90(q_a3), ref_pt="start", rel_time=0)
sched.add(Measure(q_a2, acq_index=cycle))
for qubit in (q_a1, q_a3):
sched.add(Measure(qubit, acq_index=cycle), ref_pt="start", rel_time=0)
for qubit in [q_d1, q_d2, q_d3, q_d4]:
sched.add(X(qubit), ref_pt="start", rel_time=0)
# final measurements
sched.add(Measure(*all_qubits[:4], acq_index=0), ref_pt="end", rel_time=0)
return sched
sched = mk_surface7_sched(num_cycles=3)
f, ax = circuit_diagram_matplotlib(sched)
f.set_figwidth(30)
How do we store all the shots for this measurement? We might want this because, e.g., we know we have an issue with leakage to the second excited state of a transmon and we would like to be able to store and inspect raw data.
To support such use-cases we will have a dimension in the dataset for the repeating cycles and one extra dimension for the final measurement.
# mock data parameters
num_shots = 128 # NB usually >~1000 in real experiments
ground = -0.2 + 0.65j
excited = 0.7 + 4j
centroids = ground, excited
sigmas = [0.1] * 2
display_source_code(mk_iq_shots)
display_source_code(mk_shots_from_probabilities)
display_source_code(mk_surface7_cyles_dataset)
def mk_iq_shots(
num_shots: int = 128,
sigmas: Union[Tuple[float], NDArray[np.float64]] = (0.1, 0.1),
centers: Union[Tuple[complex], NDArray[np.complex128]] = (-0.2 + 0.65j, 0.7 + 4j),
probabilities: Union[Tuple[float], NDArray[np.float64]] = (0.4, 0.6),
seed: Union[int, None] = 112233,
) -> NDArray:
"""
Generate clusters of (I + 1j*Q) points with a Gaussian distribution.
Utility to mock the data coming from qubit readout experiments.
Clusters are centered around ``centers`` and data points are distributed between
them according to ``probabilities``.
.. seealso:: :ref:`howto-utilities-examples-ssro`
Parameters
----------
num_shots
The number of shot to generate.
sigma
The sigma of the Gaussian distribution used for both real and imaginary parts.
centers
The center of each cluster on the imaginary plane.
probabilities
The probabilities of each cluster being randomly selected for each shot.
seed
Random number generator seed passed to ``numpy.random.default_rng``.
"""
if not len(sigmas) == len(centers) == len(probabilities):
raise ValueError(
f"Incorrect input. sigmas={sigmas}, centers={centers} and "
f"probabilities={probabilities} must have the same length."
)
rng = np.random.default_rng(seed=seed)
cluster_indices = tuple(range(len(centers)))
choices = rng.choice(a=cluster_indices, size=num_shots, p=probabilities)
shots = []
for idx in cluster_indices:
num_shots_this_cluster = np.sum(choices == idx)
i_data = rng.normal(
loc=centers[idx].real,
scale=sigmas[idx],
size=num_shots_this_cluster,
)
q_data = rng.normal(
loc=centers[idx].imag,
scale=sigmas[idx],
size=num_shots_this_cluster,
)
shots.append(i_data + 1j * q_data)
return np.concatenate(shots)
def mk_shots_from_probabilities(probabilities: Union[np.ndarray, list], **kwargs):
"""Generates multiple shots for a list of probabilities assuming two states.
Parameters
----------
probabilities
The list/array of the probabilities of one of the states.
**kwargs
Keyword arguments passed to
:func:`~quantify_core.utilities.examples_support.mk_iq_shots`.
Returns
-------
:
Array containing the shots. Shape: (num_shots, len(probabilities)).
"""
shots = np.array(
tuple(
mk_iq_shots(probabilities=[prob, 1 - prob], **kwargs)
for prob in probabilities
)
).T
return shots
def mk_surface7_cyles_dataset(num_cycles: int = 3, **kwargs) -> xr.Dataset:
"""
See also :func:`mk_surface7_sched` inlined in the documentation as an example in:
:ref:`sec-dataset-advanced-examples`
Parameters
----------
num_cycles
The number of repeating cycles before the final measurement.
**kwargs
Keyword arguments passed to :func:`~.mk_shots_from_probabilities`.
"""
cycles = range(num_cycles)
mock_data = mk_shots_from_probabilities(
probabilities=[np.random.random() for _ in cycles], **kwargs
)
mock_data_final = mk_shots_from_probabilities(
probabilities=[np.random.random()], **kwargs
)
# %%
data_vars = {}
# NB same random data is used for all qubits only for the simplicity of the mock!
for qubit in (f"A{i}" for i in range(3)):
data_vars[f"{qubit}_shots"] = (
("repetitions", "dim_cycle"),
mock_data,
mk_main_var_attrs(
unit="V", long_name=f"IQ amplitude {qubit}", has_repetitions=True
),
)
for qubit in (f"D{i}" for i in range(4)):
data_vars[f"{qubit}_shots"] = (
("repetitions", "dim_final"),
mock_data_final,
mk_main_var_attrs(
unit="V", long_name=f"IQ amplitude {qubit}", has_repetitions=True
),
)
cycle_attrs = mk_main_coord_attrs(long_name="Surface code cycle number")
final_msmt_attrs = mk_main_coord_attrs(long_name="Final measurement")
coords = dict(
cycle=("dim_cycle", cycles, cycle_attrs),
final_msmt=("dim_final", [0], final_msmt_attrs),
)
dataset = xr.Dataset(
data_vars=data_vars,
coords=coords,
attrs=mk_dataset_attrs(),
)
return dataset
dataset = mk_surface7_cyles_dataset(
num_shots=num_shots, sigmas=sigmas, centers=centroids
)
assert dataset == round_trip_dataset(dataset) # confirm read/write
dataset
<xarray.Dataset> Size: 27kB Dimensions: (repetitions: 128, dim_cycle: 3, dim_final: 1) Coordinates: cycle (dim_cycle) int64 24B 0 1 2 final_msmt (dim_final) int64 8B 0 Dimensions without coordinates: repetitions, dim_cycle, dim_final Data variables: A0_shots (repetitions, dim_cycle) complex128 6kB (-0.23630343679164473... A1_shots (repetitions, dim_cycle) complex128 6kB (-0.23630343679164473... A2_shots (repetitions, dim_cycle) complex128 6kB (-0.23630343679164473... D0_shots (repetitions, dim_final) complex128 2kB (-0.23630343679164473... D1_shots (repetitions, dim_final) complex128 2kB (-0.23630343679164473... D2_shots (repetitions, dim_final) complex128 2kB (-0.23630343679164473... D3_shots (repetitions, dim_final) complex128 2kB (-0.23630343679164473... Attributes: tuid: 20241012-040831-584-eca63f dataset_name: dataset_state: None timestamp_start: None timestamp_end: None quantify_dataset_version: 2.0.0 software_versions: {} relationships: [] json_serialize_exclude: []
dataset.A1_shots.shape, dataset.D1_shots.shape
((128, 3), (128, 1))
dataset_gridded = dh.to_gridded_dataset(
dataset, dimension="dim_cycle", coords_names=["cycle"]
)
dataset_gridded = dh.to_gridded_dataset(
dataset_gridded, dimension="dim_final", coords_names=["final_msmt"]
)
dataset_gridded
<xarray.Dataset> Size: 27kB Dimensions: (repetitions: 128, cycle: 3, final_msmt: 1) Coordinates: * cycle (cycle) int64 24B 0 1 2 * final_msmt (final_msmt) int64 8B 0 Dimensions without coordinates: repetitions Data variables: A0_shots (repetitions, cycle) complex128 6kB (-0.23630343679164473+0.7... A1_shots (repetitions, cycle) complex128 6kB (-0.23630343679164473+0.7... A2_shots (repetitions, cycle) complex128 6kB (-0.23630343679164473+0.7... D0_shots (repetitions, final_msmt) complex128 2kB (-0.2363034367916447... D1_shots (repetitions, final_msmt) complex128 2kB (-0.2363034367916447... D2_shots (repetitions, final_msmt) complex128 2kB (-0.2363034367916447... D3_shots (repetitions, final_msmt) complex128 2kB (-0.2363034367916447... Attributes: tuid: 20241012-040831-584-eca63f dataset_name: dataset_state: None timestamp_start: None timestamp_end: None quantify_dataset_version: 2.0.0 software_versions: {} relationships: [] json_serialize_exclude: []
dataset_gridded.A0_shots.real.mean("repetitions").plot(marker="o", label="I-quadrature")
dataset_gridded.A0_shots.imag.mean("repetitions").plot(marker="^", label="Q-quadrature")
_ = plt.gca().legend()
Dataset for a “nested MeasurementControl” experiment#
Now consider a dataset that has been constructed by an experiment involving the
operation of two
MeasurementControl
objects. The second of
them performs a “meta” outer loop in which we sweep a flux bias and then perform
several experiments to characterize a transmon qubit, e.g. determining the frequency of
a read-out resonator, the frequency of the transmon, and its T1 lifetime.
Below we showcase what the data from the dataset containing the T1 experiment results could look like
fig, ax = plt.subplots()
rng = np.random.default_rng(seed=112244) # random number generator
num_t1_datasets = 7
t1_times = np.linspace(0, 120e-6, 30)
for tau in rng.uniform(10e-6, 50e-6, num_t1_datasets):
probabilities = exp_decay_func(
t=t1_times, tau=tau, offset=0, n_factor=1, amplitude=1
)
dataset = dataset_examples.mk_t1_av_with_cal_dataset(t1_times, probabilities)
round_trip_dataset(dataset) # confirm read/write
dataset_g = dh.to_gridded_dataset(
dataset, dimension="main_dim", coords_names=["t1_time"]
)
# rotate the iq data
rotated_and_normalized = rotate_to_calibrated_axis(
dataset_g.q0_iq_av.values, *dataset_g.q0_iq_av_cal.values
)
rotated_and_normalized_da = xr.DataArray(dataset_g.q0_iq_av)
rotated_and_normalized_da.values = rotated_and_normalized
rotated_and_normalized_da.attrs["long_name"] = "|1> Population"
rotated_and_normalized_da.attrs["units"] = ""
rotated_and_normalized_da.real.plot(ax=ax, label=dataset.tuid, marker=".")
ax.set_title("Results from repeated T1 experiments\n(different datasets)")
_ = ax.legend()
Since the raw data is now split among several datasets, we would like to keep a reference to all these datasets in our “combined” datasets. Below we showcase how this can be achieved, along with some useful xarray features and known limitations.
We start by generating a mock dataset that combines all the information that would have been obtained from analyzing a series of other datasets.
Source code for mk_nested_mc_dataset function
display_source_code(mk_nested_mc_dataset)
def mk_nested_mc_dataset(
num_points: int = 12,
flux_bias_min_max: tuple = (-0.04, 0.04),
resonator_freqs_min_max: tuple = (7e9, 7.3e9),
qubit_freqs_min_max: tuple = (4.5e9, 5.0e9),
t1_values_min_max: tuple = (20e-6, 50e-6),
seed: Optional[int] = 112233,
) -> xr.Dataset:
"""
Generates a dataset with dataset references and several coordinates that serve to
index the same variables.
Note that the each value for ``resonator_freqs``, ``qubit_freqs`` and ``t1_values``
would have been extracted from other dataset corresponding to individual experiments
with their own dataset.
Parameters
----------
num_points
Number of datapoints to generate (used for all variables/coordinates).
flux_bias_min_max
Range for mock values.
resonator_freqs_min_max
Range for mock values.
qubit_freqs_min_max
Range for mock values.
t1_values_min_max
Range for mock random values.
seed
Random number generator seed passed to ``numpy.random.default_rng``.
"""
rng = np.random.default_rng(seed=seed) # random number generator
flux_bias_vals = np.linspace(*flux_bias_min_max, num_points)
resonator_freqs = np.linspace(*resonator_freqs_min_max, num_points)
qubit_freqs = np.linspace(*qubit_freqs_min_max, num_points)
t1_values = rng.uniform(*t1_values_min_max, num_points)
resonator_freq_tuids = [dh.gen_tuid() for _ in range(num_points)]
qubit_freq_tuids = [dh.gen_tuid() for _ in range(num_points)]
t1_tuids = [dh.gen_tuid() for _ in range(num_points)]
coords = dict(
flux_bias=(
"main_dim",
flux_bias_vals,
mk_main_coord_attrs(long_name="Flux bias", unit="A"),
),
resonator_freq_tuids=(
"main_dim",
resonator_freq_tuids,
mk_main_coord_attrs(
long_name="Dataset TUID resonator frequency", is_dataset_ref=True
),
),
qubit_freq_tuids=(
"main_dim",
qubit_freq_tuids,
mk_main_coord_attrs(
long_name="Dataset TUID qubit frequency", is_dataset_ref=True
),
),
t1_tuids=(
"main_dim",
t1_tuids,
mk_main_coord_attrs(long_name="Dataset TUID T1", is_dataset_ref=True),
),
)
data_vars = dict(
resonator_freq=(
"main_dim",
resonator_freqs,
mk_main_var_attrs(long_name="Resonator frequency", unit="Hz"),
),
qubit_freq=(
"main_dim",
qubit_freqs,
mk_main_var_attrs(long_name="Qubit frequency", unit="Hz"),
),
t1=(
"main_dim",
t1_values,
mk_main_var_attrs(long_name="T1", unit="s"),
),
)
dataset_attrs = mk_dataset_attrs()
dataset = xr.Dataset(data_vars=data_vars, coords=coords, attrs=dataset_attrs)
return dataset
dataset = mk_nested_mc_dataset(num_points=num_t1_datasets)
assert dataset == round_trip_dataset(dataset) # confirm read/write
dataset
<xarray.Dataset> Size: 2kB Dimensions: (main_dim: 7) Coordinates: flux_bias (main_dim) float64 56B -0.04 -0.02667 ... 0.02667 0.04 resonator_freq_tuids (main_dim) <U26 728B '20241012-040832-623-b6fcf0' .... qubit_freq_tuids (main_dim) <U26 728B '20241012-040832-624-a3b258' .... t1_tuids (main_dim) <U26 728B '20241012-040832-624-d124b7' .... Dimensions without coordinates: main_dim Data variables: resonator_freq (main_dim) float64 56B 7e+09 7.05e+09 ... 7.3e+09 qubit_freq (main_dim) float64 56B 4.5e+09 4.583e+09 ... 5e+09 t1 (main_dim) float64 56B 4.238e-05 ... 4.154e-05 Attributes: tuid: 20241012-040832-625-10bb66 dataset_name: dataset_state: None timestamp_start: None timestamp_end: None quantify_dataset_version: 2.0.0 software_versions: {} relationships: [] json_serialize_exclude: []
In this case the four main coordinates are not orthogonal coordinates, but instead just different labels for the same data points, also known as a “multi-index”.
fig, axs = plt.subplots(3, 1, figsize=(10, 10), sharex=True)
_ = dataset.t1.plot(x="flux_bias", marker="o", ax=axs[0].twiny(), color="C0")
x = "t1_tuids"
_ = dataset.t1.plot(x=x, marker="o", ax=axs[0], color="C0")
_ = dataset.resonator_freq.plot(x=x, marker="o", ax=axs[1], color="C1")
_ = dataset.qubit_freq.plot(x=x, marker="o", ax=axs[2], color="C2")
for tick in axs[2].get_xticklabels():
tick.set_rotation(15) # avoid tuid labels overlapping
It is possible to work with an explicit MultiIndex within a (python) xarray object:
dataset_multi_indexed = dataset.set_index({"main_dim": tuple(dataset.t1.coords.keys())})
dataset_multi_indexed
<xarray.Dataset> Size: 2kB Dimensions: (main_dim: 7) Coordinates: * main_dim (main_dim) object 56B MultiIndex * flux_bias (main_dim) float64 56B -0.04 -0.02667 ... 0.02667 0.04 * resonator_freq_tuids (main_dim) <U26 728B '20241012-040832-623-b6fcf0' .... * qubit_freq_tuids (main_dim) <U26 728B '20241012-040832-624-a3b258' .... * t1_tuids (main_dim) <U26 728B '20241012-040832-624-d124b7' .... Data variables: resonator_freq (main_dim) float64 56B 7e+09 7.05e+09 ... 7.3e+09 qubit_freq (main_dim) float64 56B 4.5e+09 4.583e+09 ... 5e+09 t1 (main_dim) float64 56B 4.238e-05 ... 4.154e-05 Attributes: tuid: 20241012-040832-625-10bb66 dataset_name: dataset_state: None timestamp_start: None timestamp_end: None quantify_dataset_version: 2.0.0 software_versions: {} relationships: [] json_serialize_exclude: []
The MultiIndex is very handy for selecting data in different ways, e.g.:
index = 2
dataset_multi_indexed.qubit_freq.sel(
qubit_freq_tuids=dataset_multi_indexed.qubit_freq_tuids.values[index]
)
<xarray.DataArray 'qubit_freq' (main_dim: 1)> Size: 8B array([4.66666667e+09]) Coordinates: * main_dim (main_dim) object 8B MultiIndex * flux_bias (main_dim) float64 8B -0.01333 * resonator_freq_tuids (main_dim) <U26 104B '20241012-040832-624-141422' * t1_tuids (main_dim) <U26 104B '20241012-040832-624-90d82f' qubit_freq_tuids <U26 104B '20241012-040832-624-16cdae' Attributes: unit: Hz long_name: Qubit frequency is_main_var: True uniformly_spaced: True grid: True is_dataset_ref: False has_repetitions: False json_serialize_exclude: []
dataset_multi_indexed.qubit_freq.sel(t1_tuids=dataset.t1_tuids.values[index])
<xarray.DataArray 'qubit_freq' (main_dim: 1)> Size: 8B array([4.66666667e+09]) Coordinates: * main_dim (main_dim) object 8B MultiIndex * flux_bias (main_dim) float64 8B -0.01333 * resonator_freq_tuids (main_dim) <U26 104B '20241012-040832-624-141422' * qubit_freq_tuids (main_dim) <U26 104B '20241012-040832-624-16cdae' t1_tuids <U26 104B '20241012-040832-624-90d82f' Attributes: unit: Hz long_name: Qubit frequency is_main_var: True uniformly_spaced: True grid: True is_dataset_ref: False has_repetitions: False json_serialize_exclude: []
Known limitations#
Unfortunately, at the moment the MultiIndex has the problem of not being compatible with the NetCDF format used to write to disk:
try:
assert dataset_multi_indexed == round_trip_dataset(dataset_multi_indexed)
except NotImplementedError as exp:
print(exp)
variable 'main_dim' is a MultiIndex, which cannot yet be serialized. Instead, either use reset_index() to convert MultiIndex levels into coordinate variables instead or use https://cf-xarray.readthedocs.io/en/latest/coding.html.
We could make our load/write utilities take care of setting and resetting the index under the hood. Though there are some nuances there as well. If we would do that then some extra metadata needs to be stored in order to store/restore the multi-index. At the moment, the MultiIndex is not supported when writing a Quantify dataset to disk. Below we show a few complications related to this.
Fortunately, the MultiIndex can be reset back:
dataset_multi_indexed.reset_index(dims_or_levels="main_dim")
<xarray.Dataset> Size: 2kB Dimensions: (main_dim: 7) Coordinates: flux_bias (main_dim) float64 56B -0.04 -0.02667 ... 0.02667 0.04 resonator_freq_tuids (main_dim) <U26 728B '20241012-040832-623-b6fcf0' .... qubit_freq_tuids (main_dim) <U26 728B '20241012-040832-624-a3b258' .... t1_tuids (main_dim) <U26 728B '20241012-040832-624-d124b7' .... Dimensions without coordinates: main_dim Data variables: resonator_freq (main_dim) float64 56B 7e+09 7.05e+09 ... 7.3e+09 qubit_freq (main_dim) float64 56B 4.5e+09 4.583e+09 ... 5e+09 t1 (main_dim) float64 56B 4.238e-05 ... 4.154e-05 Attributes: tuid: 20241012-040832-625-10bb66 dataset_name: dataset_state: None timestamp_start: None timestamp_end: None quantify_dataset_version: 2.0.0 software_versions: {} relationships: [] json_serialize_exclude: []
all(dataset_multi_indexed.reset_index("main_dim").t1_tuids == dataset.t1_tuids)
True
But, for example, the dtype
has been changed to object
(from fixed-length string):
dataset.t1_tuids.dtype, dataset_multi_indexed.reset_index("main_dim").t1_tuids.dtype
(dtype('<U26'), dtype('<U26'))
dataset.t1_tuids.dtype == dataset_multi_indexed.reset_index("main_dim").t1_tuids.dtype
True