RB Experiment is missing tests about qiskit-experiments HOT 2 CLOSED

In ignis the tests were split into 3 parts:

• tests for the circuit generation
• tests for the fitters, based on pre-generated JSON data
• functions to generate the JNON files (not running automatically, but we need to run them when the code is updated to update the JSON files)

However, in qiskit.experiments we can also have end-to-end tests, as in the T1 PR #5
This approach is useful for T1 and similar experiments, but can take long time when generating circuits for the advanced methods, such as RB and all its variants (interleaved, cnot-dihedral, purity etc.), QV etc, in particular if we want to check composite experiments.

So, which approach should we take here?

from qiskit-experiments.

At minimum we should have independent tests for circuit generation and analysis.

For RB circuit generation you can check that each circuit implements an identity by running Clifford(circuit), and check that the metadata reports the correct number of cliffords (this might be a bit trickier since might have to count barriers for this since we decompose when constructing the circuit), and that you have the right number of sequences. You could also check setting a seed produces deterministic output for repeat experiments.
For fitter we can use pre-generated or synthetic data to check fitting works.

Having an end to end test would be nice, but for RB we might have to wait until we can speed up the RB code since it will take too long.

from qiskit-experiments.