Bad Qubits — Learn Quantum by Doing

Multiple choice

In the kernel (feature-map) view of QML, why is the data encoding described as the most consequential choice?

The encoding fixes the kernel, and therefore the class of functions the model can ever represent, before a single parameter is trained
The encoding sets the learning rate of the classical optimizer used in the hybrid loop
The encoding determines how many shots each measurement needs, but does not affect which functions are representable
The encoding is irrelevant to the kernel; only the trainable ansatz layers determine the representable functions