Allow measuring subset of qubits when QuantumKernel.evaluate() is run on MPS simulator

[qiskit-community]

Allow measuring subset of qubits when QuantumKernel.evaluate() is run on MPS simulator

[woodsp-ibm]

From the link this looks like a performance enhancement when using MPS simulator when all qubits are being measured. You can pass in an Aer simulator set for MPS mode and configure it accordingly. Having only just skimmed over the PR you linked what is it that you are proposing is changed here?

As far as the title is concerned Allow measuring subset of qubits - I got the impression from reading the linked PR that the change did not cater to that, but it might be possible to do later. Can you explain further...

[omarshehab]

@woodsp-ibm , The QuantumKernel.evaluate() API currently measures all qubits so we cannot take advantage of measuring subset of qubits when we are using the MPS simulator.

[woodsp-ibm]

I guess I am confused - the PR you linked says this as a summary

Performance improvement for the case where all qubits are measured.

and later

It is possible to implement for cases where a subset of the qubits is measured, but this will require additional effort to decide when propagation is necessary and when not.

which I took to mean that it was not implemented at present anyway.

[omarshehab]

@woodsp-ibm , it has been implemented and merged as shown below.

Unfortunately, QuantumKernel.evaluate(), while running on MPS simulator, cannot take advantage of this at this moment.

[adekusar-drl]

@omarshehab Could you please write a bit detailed description? I don't fully understand what is suggested to do here.

[omarshehab]

@adekusar-drl , measurement of only specific qubits is required for qiskit.aqua.algorithms.VQC(). When this API is run on MPS simulator, we want to make sure that MPS simulator is taking advantage of this partial measurement.

[adekusar-drl]

As I see in qiskit_machine_learning.kernels.QuantumKernel implementation, only a required number of qubits is measured, basically it is the number of qubits in the feature map passed to QuantumKernel. If you don't pass a feature map, then ZZFeatureMap(2) is used. So, that's why I don't fully understand what is wrong in the implementation.

By the way, aqua and qiskit.aqua.algorithms.VQC are deprecated, I don't think we will update them with new features.

[adekusar-drl]

I'm closing this issue as stale and unclear. Feel free to re-open with more details provided.