Add cirq.TensoredConfusionMatrices for readout error mitigation.

[tanujkhattar]

Fixes #4800.

cc @mrwojtek @mpharrigan

[mpharrigan]

super cool. I think an abstraction to tensor-product a big cm from small ones is very useful to have.

I strongly worry about using the words and variable names "confusion matrix" to refer both to (ndarray, Sequence[Qid]) tuples and to ReadoutConfusionMatrix as a source of ... confusion

[mpharrigan]

This also uses raw numpy arrays for the "results" bitstrings and an explicit back-mapping to qubit identities; which I think is correct to make this functionality composable. But it puts a big onus on the experimenter like described in #4449

[MichaelBroughton]

This looks awesome! Few high level questions:

1. Should the measure_confusion_matrix function have a parallel vs isolated operation mode ? Thinking along the lines of how measure the data for the 2x2 confusion matrices slightly differently for things like p00_isolated vs p00_parallel. WDYT ?
2. Can we generalize this a little more to make the readout characterization that we also have in the experiments dir obsolete ?

[tanujkhattar]

1. Should the measure_confusion_matrix function have a parallel vs isolated operation mode ? Thinking along the lines of how measure the data for the 2x2 confusion matrices slightly differently for things like p00_isolated vs p00_parallel. WDYT ?

Yes, this is a good idea but I think we can do this in a follow-up PR.

2. Can we generalize this a little more to make the readout characterization that we also have in the experiments dir obsolete ?

I guess once 1. above is done, we should be able to make the single qubit readout calibration obsolete since it just represents single qubit confusion matrices for each qubit. We will have to go through a deprecation cycle though and I think it's a widely used method so there will be quite a few places to update.

I'd recommend to open separate issues for both 1 & 2 above and do them in separate PRs. LMK your thoughts.

I've addressed all other comments. cc @mpharrigan

[mrwojtek]

This is quite neat implementation!

[tanujkhattar]

@mrwojtek Addressed all comments. PTAL.

[mrwojtek]

Commented on the optimization failure above.

[tanujkhattar]

@mrwojtek Fixed.

[mrwojtek]

LGTM!

Stale review, changes have been addressed.

[CirqBot]

Automerge cancelled: A required status check is not present.

Missing statuses: ['Pytest MacOS (3.7)', 'Pytest MacOS (3.8)', 'Pytest MacOS (3.9)', 'Pytest Windows (3.7)', 'Pytest Windows (3.8)', 'Pytest Windows (3.9)', 'Typescript lint check', 'Typescript tests', 'Typescript tests coverage']