ADAM bug when calculating the gradient in batches

[proeseler]

Environment

• Qiskit Algorithms version: 1.0.1
• Python version: 3.11
• Operating system: Linux

What is happening?

There is only one small error that prevents ADAM from calculating the gradient_num_diff for batches with max_evals_grouped. ADAM only gives "fun, self._eps" as argument when calling gradient_num_diff. This leads to max_evals_grouped=None, which leads to max_evals_grouped=1. Therefore, regardless of the call to set max_evals_grouped, max_evals_grouped=1 will always apply for ADAM.
Optimizer class method minimize:

ADAM class method gradient_num_diff:

Here are the corresponding files:

• https://qiskit-community.github.io/qiskit-algorithms/_modules/qiskit_algorithms/optimizers/optimizer.html#Optimizer
• https://qiskit-community.github.io/qiskit-algorithms/_modules/qiskit_algorithms/optimizers/adam_amsgrad.html#ADAM

How can we reproduce the issue?

Create the ADAM optimizer. Call set_max_evals_grouped with any limit with the optimizer. Now call minimize with the optimizer and you will not notice any change in the runtime/CPU usage.

What should happen?

I could quickly fix the error by adding the missing argument.

Any suggestions?

No response

[woodsp-ibm]

So that would simply be to add self._max_evals_grouped to the call. If you would like to contribute the change that would be great. Would you though please add a unit test that makes sure things work if max evals grouped is set when using Adam in that the outcome is as expected still - and a release note (reno) to note the fix. As to how a set of points affects CPU depends on how things are computed by the function, locally they could be parallel processed, if remote it was done to reduce queuing overhead by allowing bundling circuits for each point in the gradient into a single call/run.

[proeseler]

I will do so👍