Is there any way to get the intermediate steps (similar to the callback functions in Tensorflow or Pytorch Adams optimizer) when using Qiskit's internal ADAM optimizer?
I am implementing a Variational Quantum Circuit(VQC) to train the quantum machine learning model using Qiskit's Adam optimizer.
I looked through the document of ADAM optimizer in Qiskit, but I could not find the appropriate methods to get them.
I could get the intermediate training information with the Pytorch Adam optimizer, but I wanted to know whether I could do it with Qiskit.
I am using qiskit v0.45.2.
As you observed the ADAM optimizer in Qiskit Algorithms does not have a callback like the other optimizers. It has been noted in this issue https://github.com/qiskit-community/qiskit-algorithms/issues/60 which was more around unification of the signatures of the optimizers, in qiskit algorithms, so it's consistent over the set of them. Now ADAM does have a snapshot_dir parameters which can be used to have its internal parameters saved if that is of use at all for you now.
The source to ADAM is here https://github.com/qiskit-community/qiskit-algorithms/blob/main/qiskit_algorithms/optimizers/adam_amsgrad.py so you can see what it does. It was moved out of Qiskit and qiskit.algorithms as part of the move to have a separate repository/package for qiskit algorithms. With the source you could always make a copy and edit it to add a callback for yourself, if that's important for you to have.