`averaging_mode=AveragingMode.SINGLESHOT` default for probability routines

[aorgazf]

Many of the probabilities routines:

• rabi_amplitude
• rabi_length
• ramsey
• t1
• t2

use as SINGLESHOT as their default averaging_mode. This severely limits the amount of points that can be scanned, as the memory is dedicate to save the shots.

When the classification to be performed is merely a separation between states 0 and 1 by a straight line, this can be done efficiently within the FPGA of the control electronics. SEQUANTIAL should be the default configuration for all these routines while allowing to obtain SINGLESHOT information as an additional feature for singular classifications.

Most of the control electronics (QM, Qblox and ZI) allow more complex classifications using weights, to distinguish between higher states.

https://github.com/qiboteam/qibocal/blob/main/src/qibocal/protocols/characterization/rabi/amplitude.py#L100

[andrea-pasquale]

As discussed on Monday, all the previous protocols are available in AveragingMode.CYCLIC through the corresponding *_signal protocols. In the future we plan to unify the protocols and add a flag to choose between the two acquisition mode as already discussed in #832.

This severely limits the amount of points that can be scanned, as the memory is dedicate to save the shots.

Also on Monday, we discussed about the fact that usually for those type of protocols it is not necessary to run huge sweeps. Could you provide an example where such huge scans are necessary leading to an error in the code?

SEQUENTIAL should be the default configuration for all these routines while allowing to obtain SINGLESHOT information as an additional feature for singular classifications.

I believe that this point should be discussed with Qibolab devs, up to now we have preferred CYCLIC compared to SEQUENTIAL because it should be more robust to noise.