Model catalog

QuOptuna searches over quantum models (PennyLane) and classical models (scikit-learn). Searchable hyperparameters below come from MODEL_PARAM_KEYS.

Note

Kernel models (IQPKernelClassifier, ProjectedQuantumKernel, QuantumKitchenSinks, SeparableKernelClassifier) have no iterative training steps and are not prunable. Variational models are trained iteratively and are prunable. Variational quantum models are OvR-wrapped for multiclass. TreeTensorClassifier is a quantum model despite the name.

Note

The names below are the registry keys accepted by the CLI --models flag and the REST API. The /api/v1/models endpoint returns human-readable display names (for example Convolutional Neural Network) for use in the UI — those are not valid registry keys.

Quantum models (PennyLane)

Model	Type	Description	Searchable hyperparameters
CircuitCentricClassifier	variational	Circuit-centric variational classifier	max_vmap, batch_size, learning_rate, n_input_copies, n_layers
DataReuploadingClassifier	variational	Data re-uploading variational classifier	max_vmap, batch_size, learning_rate, n_layers, observable_type
DataReuploadingClassifierSeparable	variational	Separable data re-uploading classifier	max_vmap, batch_size, learning_rate, n_layers, observable_type
DressedQuantumCircuitClassifier	variational	Dressed quantum circuit (classical + quantum layers)	max_vmap, batch_size, learning_rate, n_layers
DressedQuantumCircuitClassifierSeparable	variational	Separable dressed quantum circuit	max_vmap, batch_size, learning_rate, n_layers
QuantumBoltzmannMachineSeparable	variational	Separable quantum Boltzmann machine	max_vmap, batch_size, learning_rate, visible_qubits, temperature
QuantumBoltzmannMachine	variational	Quantum Boltzmann machine	max_vmap, batch_size, learning_rate, visible_qubits, temperature
TreeTensorClassifier	variational	Tree-tensor-network circuit classifier	max_vmap, batch_size, learning_rate
IQPKernelClassifier	kernel	IQP-embedding quantum kernel classifier	max_vmap, repeats, C
ProjectedQuantumKernel	kernel	Projected quantum kernel classifier	max_vmap, gamma_factor, C, trotter_steps, t
QuantumKitchenSinks	kernel	Quantum kitchen sinks feature map	max_vmap, n_qfeatures, n_episodes
QuantumMetricLearner	variational	Quantum metric learning classifier	max_vmap, batch_size, learning_rate, n_layers
SeparableVariationalClassifier	variational	Separable variational classifier	batch_size, learning_rate, encoding_layers
SeparableKernelClassifier	kernel	Separable quantum kernel classifier	C, encoding_layers

Image-shaped models

These models are in the registry and selectable, but they reshape each row into a square 2D grid, so they only make sense for image-like datasets whose feature count is a perfect square. They are not suitable for general tabular data.

Model	Type	Description	Searchable hyperparameters
QuanvolutionalNeuralNetwork	variational	Quantum convolutional filters feeding a CNN	max_vmap, batch_size, learning_rate, n_qchannels, qkernel_shape, kernel_shape
WeiNet	variational	Quanvolutional model using fixed classical filters (edge_detect, smooth, sharpen)	max_vmap, batch_size, learning_rate, filter_name
ConvolutionalNeuralNetwork	classical	Classical CNN baseline	batch_size, learning_rate, kernel_shape

Classical models (scikit-learn)

Model	Description	Searchable hyperparameters
SVC	Support vector classifier (RBF)	gamma, C, class_weight
SVClinear	Linear SVC (LinearSVC)	C, class_weight
MLPClassifier	Multi-layer perceptron (hidden_layer_sizes required)	batch_size, hidden_layer_sizes, alpha, learning_rate
Perceptron	Linear perceptron	eta0, class_weight

See also

• Python API reference
• CLI reference
• Configuration reference