Incorrect ValueError while validating inputs in `BaseKernel`

[SaashaJoshi]

Environment

• Qiskit Machine Learning version: 0.8.2/0.8.3
• Python version: 3.10
• Operating system: MacOS

What is happening?

While creating a custom feature map (FM), I was mistakenly using all the parameters as trainable parameters.

In the TrainableFidelityQuantumKernel class this would mean overriding the value of num_features.

qiskit-machine-learning/qiskit_machine_learning/kernels/trainable_fidelity_quantum_kernel.py

Lines 94 to 95 in ef44f5a

	# override the num of features defined in the base class
	self._num_features = self.feature_map.num_parameters - self._num_training_parameters

So, with my custom FM,

num_features = 0,  #since
self.feature_map.num_parameters = self._num_training_parameters

Next step in the program is to call the evaluate function in the FidelityQuantumKernel class:

qiskit-machine-learning/qiskit_machine_learning/kernels/fidelity_quantum_kernel.py

Line 100 in ef44f5a

def evaluate(self, x_vec: np.ndarray, y_vec: np.ndarray | None = None) -> np.ndarray:

And, from here, the code logic goes into the _validate_inputs method in the BaseKernel class, which does this,

qiskit-machine-learning/qiskit_machine_learning/kernels/base_kernel.py

Lines 110 to 120 in ef44f5a

	if x_vec.shape[1] != self._num_features:
	# before raising an error we try to adjust the feature map
	# to the required number of qubit.
	try:
	self._feature_map.num_qubits = x_vec.shape[1]
	except AttributeError as a_e:
	raise ValueError(
	f"x_vec and class feature map have incompatible dimensions.\n"
	f"x_vec has {x_vec.shape[1]} dimensions, "
	f"but feature map has {self._feature_map.num_parameters}."
	) from a_e

Now, the logic in this code checks if x_vec.shape[1] != self._num_features but returns a different ValueError.

With my custom FM, x_vec,shape[1]=8, self._num_features=0, and feature_map.num_parameters=8. However, the ValueError incorrectly returns this,
...but feature map has {self._feature_map.num_parameters}.

This translates to the error that I have been receiving: x_vec and class feature map have incompatible dimensions.
x_vec has 8 dimensions, but feature map has 8.

Instead the ValueError should be,
...but feature map has {self._num_features}.

There is an error with my custom FM and an incorrect ValueError made it difficult to debug. I think the ValueError should be corrected.

How can we reproduce the issue?

Create a custom feature map with all the parameters as trainable parameters.

You can also refer to this slack discussion thread between me and @woodsp-ibm about this issue: https://qiskit.slack.com/archives/CB6C24TPB/p1745436047257769

What should happen?

ValueError must be corrected to reflect the logic it checks, i.e., if x_vec.shape[1] != self._num_features

raise ValueError(
      f"x_vec and class feature map have incompatible dimensions.\n"
      f"x_vec has {x_vec.shape[1]} dimensions, "
      f"but feature map has {self._num_features}."
  ) from a_e

Any suggestions?

No response