There's code in the notebook to add a label to the RB plot...this should be in the fitter code

The code in clifford_utils.py, lines 215-223 seems not to be the same as described in the appendix of https://arxiv.org/pdf/1210.7011.pdf (see also the jupyter notebook on the structure of the Clifford group in: https://github.ibm.com/IBMQuantum/qiskit-ignis-internal/pull/83).

In the appendix, the forth class is the "SWAP class", so I think that the code should be:

        h_gates(gatelist, 0, h0)
        h_gates(gatelist, 1, h1)

        v_gates(gatelist, 0, r0)
        v_gates(gatelist, 1, r1)

        cx_gates(gatelist, 0, 1)
        cx_gates(gatelist, 1, 0)
        cx_gates(gatelist, 0, 1)

However, the current code is (this was the original code - I didn't change it):

        h_gates(gatelist, 0, h0)
        h_gates(gatelist, 1, h1)

        cx_gates(gatelist, 0, 1)
        cx_gates(gatelist, 1, 0)
        cx_gates(gatelist, 0, 1)

        v_gates(gatelist, 0, r0)
        v_gates(gatelist, 1, r1)

I checked both options, and both of them produce 11520 elements in the Clifford group, so both options cover all the elements in the group (but using different gate representations).
I also checked test_rb and it runs OK, but test_clifford fails, since the tables (of lists of gates) are not the same.
So in order to change the code I also need to replace the file "test_tables_expected".

Currently the coherence notebooks and tests use phase_amplitude_damping_error and amplitude_damping_error, but they should use thermal_relaxation_error

The text in examples/relaxation_and_decoherence.ipynb talks about amplitude damping and phase damping errors, however the experiments use thermal relaxation errors. We can either change the text, or leave it as it is with a comment about the usage of another error in the code.

What is the expected enhancement?

Currently we have a good README.md file for the project and helpful notebooks. We can improve this further.

• Add a general description of the goals of state and process tomography.
• Add a clear description of the algorithms used (mathematics can be written in a dedicated notebook).
• Add precise documentation for the user-supplied parameters that can affect the tomography process.

A Jupyter notebook is already available in:
https://github.ibm.com/IBMQuantum/qiskit-ignis-internal/pull/83

What is the expected enhancement?

Should the MeasurementFilter be passed the qiskit terra result object and it return a qiskit results object.

We will move part of examples to https://github.com/Qiskit/qiskit-tutorials/tree/master/qiskit/ignis and the other part to https://github.com/Qiskit/qiskit-tutorials/tree/master/community.

Community

• Tomography overview
• Coherence Overview
• RB Overview
• Clifford Overview

Tutorials

• Tomography Examples
• Characterization Examples
• RB Examples
• Measurement Examples

The tests run with 1 or 2 qubits, so I actually wonder why they take so long.
We should profile, as @ShellyGarion did for randomized benchmarking. If compilation seems to be the problem, can we disable the transpiler?
Also related is the question of tests accuracy. The tests set the initial values for the fitter to be the correct fitting parameters. This does not mimic the reality, where the user does not know the correct parameters, so she sets as initial values her inaccurate estimates for the correct parameters.
If we stick with these accurate initial values, then we can decrease the gap (delta) that we allow between the real T1/T2/T2* and the computed one. This was indeed done in test_t2 and test_t2star, but not in test_t1. Alternatively, we can increase delta and decrease the number of shots and data points.

Fitter spits out a filter object that can be applied to correct the measurement data...

What is the expected behavior?

we need to set up the admin files

• Remove shots
• Ability to add all the circuits or add more shots from the save seed
• Ideally pull off the base fitter class

What is the expected behavior?

Let's merge the tomography overview that @gadial make into the examples notebooks and clean up the notation @chriseclectic

These will replace the current tomography notebooks in https://github.com/Qiskit/qiskit-tutorials/tree/master/qiskit/ignis so make sure they provide at least as much information and are accessible (they can have parts which are advanced, but lead off with simpler discussion)

Informations

• Qiskit Ignis version: 0.1.0
• Python version: 3.7.2
• Operating system: MacOS 10.14

What is the current behavior?

Running the first cell of code in the state_tomography tutorial notebook:

# Needed for functions
import numpy as np
import time

# Import Qiskit classes
import qiskit 
from qiskit import QuantumRegister, QuantumCircuit, ClassicalRegister, Aer
from qiskit.quantum_info import state_fidelity
from qiskit.providers.aer import noise

# Tomography functions
import sys, os
sys.path.append(os.path.abspath(os.path.join('..')))
import qiskit.ignis.verification.tomography as tomo
import qiskit.ignis.error_mitigation.measurement as mc

throws this error:
ModuleNotFoundError: No module named 'qiskit.ignis.error_mitigation'

Steps to reproduce the problem

Run the first code cell in the state_tomography notebook

What is the expected behavior?

Packages should be imported without any errors

Suggested solutions

I replaced the line: import qiskit.ignis.error_mitigation.measurement as mc with
import qiskit.ignis.mitigation.measurement as mc as I saw the folder 'qiskit/ignis/mitigation' was similar to what was being originally called. After I did this I ran through the notebook and the imports did not throw any errors. The other cells seemed to work as well.

https://github.com/Qiskit/qiskit/blob/master/docs/ignis/characterization.rst

Should be a usability guide and a terse overview of features.

Informations

• Qiskit Ignis version: 0.1.0
• Python version: 3.7.2
• Operating system: MacOS version 10.14

What is the current behavior?

AttributeError: 'list' object has no attribute 'split' being thrown in randomized_benchmarking tutorial

Steps to reproduce the problem

Execute all of the cells up to this cell:

backend = qiskit.Aer.get_backend('unitary_simulator')
basis_gates = ['u1', 'u2', 'u3', 'cx']
job = qiskit.execute(qc, backend=backend, basis_gates=basis_gates)
print(np.around(job.result().get_unitary(),3))

Executing this cell will throw the error.

What is the expected behavior?

This output:

[[-0.-1.j  0.+0.j  0.+0.j  0.+0.j -0.+0.j  0.+0.j  0.+0.j  0.+0.j]
 [ 0.+0.j -0.-1.j  0.+0.j  0.+0.j  0.+0.j  0.-0.j  0.+0.j  0.+0.j]
 [ 0.-0.j  0.+0.j -0.-1.j  0.+0.j  0.+0.j  0.+0.j -0.+0.j  0.+0.j]
 [ 0.+0.j  0.-0.j  0.+0.j -0.-1.j  0.+0.j  0.+0.j -0.-0.j  0.-0.j]
 [-0.+0.j  0.+0.j  0.+0.j  0.+0.j -0.-1.j  0.+0.j  0.+0.j  0.+0.j]
 [-0.+0.j  0.+0.j  0.+0.j  0.+0.j  0.+0.j -0.-1.j  0.+0.j  0.+0.j]
 [ 0.+0.j  0.+0.j -0.+0.j -0.+0.j  0.-0.j  0.+0.j -0.-1.j  0.-0.j]
 [ 0.+0.j  0.+0.j -0.+0.j  0.+0.j  0.+0.j  0.-0.j -0.+0.j -0.-1.j]]

Suggested solutions

Converting basis_gates from a list to a string, as the issue seems to be because basis_gates is a list instead of a string.

Right now the simultaneous sequences are squashed to the measurement. Add the option to add barriers across sequences.

What is the expected behavior?

For T1, I see that the BaseCoherenceFitter constructor is initialized with expected_state=1. Then BaseCoherenceFitter._calc_data contains the lines:

expected_state_list = [self._expected_state] * self._num_of_qubits
expected_state_str = ''.join(expected_state_list)

which generate, for two qubits, expected_state_str='11', instead of 01.

Perhaps I'm wrong, as test_t1 is supposed to fail if there really is a bug.

initial_c in test_t1 should be 0.

Right now because some of the tests use aer they are both slow and fail since they are non-deterministic. Can we copy the model of rb and pkl some results to test the fitter? We should still run the circuits through the idea simulator to make sure it doesn't fail.

Informations

• Qiskit Ignis version:
• Python version:
• Operating system:

What is the current behavior?

Steps to reproduce the problem

What is the expected behavior?

Suggested solutions

Informations

• Qiskit Ignis version: 0.1.0
• Python version: 3.6
• Operating system: Linux

What is the current behavior?

@gadial showed me that test_zz failed for him with:
RuntimeError: Optimal parameters not found: The maximum number of function evaluations is exceeded.

It happened when Gadi submitted a pull request, where the change did not affect test_zz.

The tests in test_characterization.py verify that the following flow runs: circuits generation - backend execution - fitting. The goal is to identify cases where the interface of one of these three components changes, requiring updates in the other components. The error of exceeding the maximum number of evaluations has to be ignored.

Steps to reproduce the problem

What is the expected behavior?

Suggested solutions

Three things, each can be done independently from the others:

1. Fix the random seed in characterization tests. This will hopefully ensure that such unexpected errors will not occur, in spite of the difference in the simulator's behavior in different platforms.
2. Catch exceptions: encapsulate calls to fitters' constructors with a try-catch, which will be able to identify errors of mismatch in interfaces. This will probably require also encapsulating the call to curve_fit in the fitters' code.
3. Consider changing Travis, such that it will run only that tests that can be affected by the change introduced in the pull request.

Here is the documentation of t1_circuits:

    Generates circuit for T1 measurement.
    Each circuit consists of an X gate, followed by a sequence of identity
    gates.

    Args:
       num_of_gates (list of integers): the number of identity gates in each
                                        circuit. Must be in an increasing
                                        order.
       gate_time (float): time in micro-seconds of running a single gate.
       qubits (list of integers): indices of the qubits whose T1 are
       to be measured.

    Returns:
       A list of QuantumCircuit
       xdata: a list of delay times in seconds

While gate_time is in micro-seconds, the unit of delay time is xdata are said to be seconds. I assume that this is incorrect, for xdata? The implementation itself does not convert between units.

There's an error if one of the standard deviations is zero
eg

[{'mean': array([0.98 , 0.905, 0.815, 0.58 , 0.43 , 0.42 , 0.38 ]),
  'std': array([0.   , 0.015, 0.045, 0.04 , 0.05 , 0.03 , 0.05 ])}]

ValueError: Residuals are not finite in the initial point.

See method https://arxiv.org/abs/1203.4550

What is the expected behavior?

https://github.com/Qiskit/qiskit-ignis/blob/8f1bdb61f98c003199c808b39df62057ae0fe2bb/qiskit/ignis/verification/tomography/fitters/cvx_fit.py#L281

What is the expected enhancement?

Can we remove the readme from tomography and add content to documentation.

To cope with 0 std, we have in BaseCoherenceFitter._calc_data:

if self._ydata['std'][-1]==0:
      self._ydata['std'][-1]=1e-4

How about when self._ydata['std'][-2] is zero?

What is the expected behavior?

Currently the Tomography module lacks any sort of tests, and although the code quality is good, there are still some linter errors.

• Ensure the code is lint-error free.
• Add unit tests to the major parts of the code (fitters, circuit generation)
• Add tests for the full run of the workflow (similar to what is performed in the notebooks).

It seems to me that currently state tomography consists of one clear entrance point - a circuit and a list of qubits - and one clear exit point, namely the density operator. Currently, it is up to the user to move data around - generate the tomography circuits, run the simulator on them, format the data for the fitter and run the fitter itself.

It will be helpful to have a wrapper function such as
def perform_state_tomography(circuit, measured_qubits, **kwargs)
that outputs the density operator after the tomography is complete. The control over the tomography process (e.g. the number of shots in the simulator runs, the simulator to use, the fitter to use etc.) will be done via keywords with sensible default values.

We can also add support for a hash, qobj-like data structure which will explicitly contain the circuits, the intended measured qubits and the additional keywords. I already have a basic implementation along these lines.

What is the expected behavior?

Should we remove shots from the coherence fitters. We don't need this in the fitters.

The following code works well:

q3 = QuantumRegister(3)
circuit = QuantumCircuit(q3)
circuit.h(q3[2])
rho = perform_state_tomography(circuit, q3)

Where perform_state_tomography is a wrapper function I wrote for the state tomography workflow. The result is the correct 8x8 density matrix.

I am also able to compute the 4x4 density matrix for the first two qubits by performing

q2 = [q3[0], q3[1]]
rho = perform_state_tomography(circuit, q2)

However, this fails for other subsets. For instance,

q2 = [q3[0], q3[2]]
rho = perform_state_tomography(circuit, q2)

Fails with the following error:

AerBackend: simulation failed
Traceback (most recent call last):
  File "C:/Users/gadia/Documents/Quantum/qiskit-ignis-internal/test.py", line 116, in <module>
    rho = perform_state_tomography(circuit, q2)
  File "C:\Users\gadia\Documents\Quantum\qiskit-ignis-internal\verification\tomography\interface.py", line 11, in perform_state_tomography
    tomography_data_results = tomography_data(job.result(), tomography_circuits)
  File "C:\Program Files\Anaconda3\lib\site-packages\qiskit_aer\backends\aerjob.py", line 37, in _wrapper
    return func(self, *args, **kwargs)
  File "C:\Program Files\Anaconda3\lib\site-packages\qiskit_aer\backends\aerjob.py", line 92, in result
    return self._future.result(timeout=timeout)
  File "C:\Program Files\Anaconda3\lib\concurrent\futures\_base.py", line 398, in result
    return self.__get_result()
  File "C:\Program Files\Anaconda3\lib\concurrent\futures\_base.py", line 357, in __get_result
    raise self._exception
  File "C:\Program Files\Anaconda3\lib\concurrent\futures\thread.py", line 55, in run
    result = self.fn(*self.args, **self.kwargs)
  File "C:\Program Files\Anaconda3\lib\site-packages\qiskit_aer\backends\aerbackend.py", line 161, in _run_job
    self._validate_controller_output(output)
  File "C:\Program Files\Anaconda3\lib\site-packages\qiskit_aer\backends\aerbackend.py", line 190, in _validate_controller_output
    raise AerSimulatorError(res.get("status", None))
qiskit_aer.backends.aersimulatorerror.AerSimulatorError: 'ERROR: invalid stoull argument'

I'm not sure I understand what causes this error. I can debug it myself, but maybe you have a clear idea on why this fails.

A lighter version of the RB & Clifford module in Ignis only for 1&2 qubits, independent of the general Clifford module that should be added in Terra.

The output Cliffords will be in the form: (H.P.V) where H is either H or I, P is the Pauli and V is the axis swap group (I, V, W).

The code will include the following files:

• Clifford.py // Clifford object
• clifford_utils.py // generating the Clifford tables, random Clifford, inverse Clifford
• randomizedbenchmarking.py // RB code
• rb_fitters.py // RB fitters

As well as Jupyter tutorial notebooks:

• RBn.ipynb // RB for 2 qubuts
• Clifford.ipynb // Cliffords for 1&2 qubits

What is the expected behavior?

I'm taking the issues Jay opened in the internal repo:

• state tomography (linear with wizard)
  See:

• Maximum Likelihood, Minimum Effort

• Fast state tomography with optimal error bounds

• state tomography (cvx)

• process tomography (linear inversion)

• process tomography (cvx)

We need to create a readme

https://github.com/Qiskit/qiskit/tree/master/docs/ignis

Should be a usability guide and a terse discussion of the features. Not a tutorial on the Cliffords.

Currently the measurement mitigation constructs a correction matrix by preparing all 2^N basis states. If measurement noise is independent (ie. no crosstalk) we can construct the correction matrix from 2 (or N) calibrations (each qubit in the excited state together or separately). Still need to construct a 2^N matrix although we might be able to use sparse matrix tools.

What is the expected behavior?

It might make sense to define a base TomographyFitter class so that additional fitters can be added that share a common interface.

At minimum it would needs to be able to read load the output from the tomography_data function and return a fitted density matrix.

With respect to the current code, this would mean the fitter_data function which converts the tomography data into the format for the least squares fitting used by the mle and cvx fit would just become an internal function called by those fitters, and the mle_fit and cvx_fit functions would be inside the classes.

As we work on Ignis, we need to understand how we cope with other repositories on whom Ignis depends, e.g. Terra and Aer. Essentially there are two options:

1. Always try that Ignis master branch works with the master branches of the other repositories.
2. Make Ignis master branch work with the stable branches of the other repositories.

I prefer Option (2), but I think Aer and Aqua tend to work with Option (1), so it's OK too. The reason that I prefer Option (2) is that, with Option (1), we may find ourselves too busy in adjusting to frequent changes in Terra interfaces. Also we will depend on the master branch of Aer to be updated with the master branch of Terra. I think Option (2) is cleaner.

Decision here will affect:

• Instructions in .github/CONTRIBUTING.rst.
• Content of requirements.txt and requirements-dev.txt.
• A question that came up in the context of randomized benchmarking, about the representation of basis gates (as a string, e.g. 'u1,u2,u3,cx', or as a list of strings, e.g. ['u1', 'u2', 'u3', 'cx']).

See the plot in the jupyter notebook example.

It needs state labels.

This is MeasurementFitter.plot_calibration

What is the expected enhancement?

• Add more description and background (see https://github.com/Qiskit/qiskit-tutorials/tree/master/qiskit/ignis)

https://github.com/Qiskit/qiskit-ignis/blob/640c2401b7d3821d77179a713b5ca14348a09b40/qiskit/ignis/randomized_benchmarking/circuits.py#L183 assumes that the qubits in the pattern go from [0,nq-1], but in general this is not true

Put very rough logging in

In the T2 Echo there is one pi pulse, generalize to a CPMG sequence (essentially the same sequence except the number of echo's is a parameter).