Discriminator Experiment Class#30
Conversation
|
It appears that you've implemented something for multiple experiments that are parallel independent experiments. If they are indeed independent, then the correct thing to do would be to implement circuit and analysis functions that see only one qubit, and to handle multiple qubits by the parallel experiment class. |
|
If the analysis function assumes that the circuits were executed with |
|
I see that |
|
Perhaps you can leverage the IQ test backend in the tutorial: https://github.com/Qiskit/qiskit-experiments/blob/960f6395db248711ec5d460e7371d275b041c2f6/qiskit_experiments/test/mock_iq_backend.py#L46 |
eggerdj
left a comment
eggerdj
left a comment
There was a problem hiding this comment.
This looks like it is going in the right direction. I think we can make the integration with the data processing nodes a bit leaner. For instance, it looks like one could combine the LDADiscriminator and QDADiscriminator nodes.
| class DiscriminatorExperiment(BaseExperiment): | ||
| """Discriminator Experiment class""" |
There was a problem hiding this comment.
How would this work with say 0, 1, and 2? I.e. if we want to use a discriminator for a qutrit experiment. Will there be a new experiment or should we make the current one more flexible?
There was a problem hiding this comment.
The ESP discriminator is being implemented here: coruscating#4
Because it requires pulse, we felt it's significantly different from the standard discriminator and so it's a completely different class.
eggerdj
left a comment
eggerdj
left a comment
There was a problem hiding this comment.
Thanks for updating the code. Given how the code is written it looks like the Discriminate node is not trainable so it should be Discriminate(DataAction). Do we want to allow untrained data processors to be used as nodes, i.e. Discriminate(TrainableDataAction)? If so I would expect the processor to train itself on a subset of the data e.g. if an experiment contains circuits of the form:
# Calibration circuits
|0> - meas
|0> - X - meas
# Experiment circuits
Some cool circuits
then a discriminator could train itself.
There was a problem hiding this comment.
Can this be an untrained discriminator as the base class suggests it could be?
| """Base class for discriminator processor. Takes IQ data and calibrated discriminator as input, | ||
| outputs counts.""" |
There was a problem hiding this comment.
Could this take an untrained discriminator as input as the base class would suggest?
| """Base class for discriminator processor. Takes IQ data and calibrated discriminator as input, | ||
| outputs counts.""" | ||
|
|
||
| def __init__(self, handle: ExperimentData, validate: bool = True): |
There was a problem hiding this comment.
What is the rational for giving the init a handle to the data? This deviates from the other nodes which are called on the data.
| from .discriminator_analysis import DiscriminatorAnalysis | ||
|
|
||
|
|
||
| class Discriminator(BaseExperiment): |
There was a problem hiding this comment.
Could we either change the name to indicate that this discriminates 0 and 1 or provide the functionality to extend this to higher excited states? I think changing the name would be sufficient at first.
| "meas_level": self.run_options.meas_level, | ||
| "meas_return": self.run_options.meas_return, |
There was a problem hiding this comment.
I believe we can remove these now. Is there a reason to keep them?
|
@coruscating Do you intend to complete this PR ? |
|
If not in the near future then we don't have to close it, we can keep it open and move it to the IceBox pipeline |
|
@yaelbh Yes, I'm working on the updates and almost done. |
|
Will open new PR for this. |
4 participants
Summary
Basic discriminator class that implements LDA and QDA for multiple qubits. One calibration is done for |00..0> and |11..1> state, then each qubit is fitted separately. Discriminator data processing node, once trained, can be used to discriminate future data.