Quantum-Clustering

Quantum-clustering is a framework to train and test variational quantum kernel (VQK) functions in cooperation with classical clustering algorithms using pennylane for the quantum simulation and scikit-learn for the clustering algorithms. In order to test this concept, this framework supports many different different parameters, including cost functions, optimizers, quantum ansätze, datasets, clustering algorithms, etc.

Basic Usage

There are two main ways to use this framework, directly from the code or with configuration files.

Code

Creating the necessary python objects and starting the training by hand. Preferred if you only want to use parts of the existing framework for other purposes.

Configuration Files

Define a Json configuration file using the configuration options and execute the config file with the provided functions. Preferred for large scale test scenarios.

from qclustering.JsonHelper import run_json_file
run_json_file("config.json")

Configuration Options

The config.json file shows an example configuration file. The following list shows most of the possible configuration details.

• numpy_seed: int - Numpy seed to achieve reproducable results for the randomness
• shots: int - Number of shots used for the quantum calculations
• device: string = ["default.qubit"|"default.mixed"] - Quantum device used for the simulation.
• circuit: string = ["overlap"|"swap"] - Quantum circuit used for calculating the kernel function. Either "overlap" or "swap"
• num_processes: int - Number of processes used for calculating the full kernel matrix. Currently only supported for noiseless calculations.
• cost_func: string = ["KTA"|"triplet-loss"|"davies-bouldin"|"calinski-harabasz"] - Name of the cost function used to train the VQK
• cost_func_params: Object - All further parameters for the given cost function.
• ansatz: string = ["ansatz1"|"ansatz1noise"|"ansatz2"] - Name of the quantum ansatz to use.
• ansatz_params: Object - Further configuration for the ansatz.
  • wires: int - Number of qubits to use.
  • layers: int - Number of layers to use.
• dataset: string = ["iris"|"moons"|"donuts"] - Name of the dataset to use.
• dataset_params: Object - Further parameters for the dataset.
  • train_size: int - Number of samples to use for training.
  • val_size: int - Number of samples to use for validation.
  • test_size: int - Number of samples to use for testing.
  • apply_pca: bool - Apply PCA to the dataset.
  • num_features: int - Define the desired number of features for PCA.
• epochs: int - Number of epochs to train the VQK.
• batch_size: int - Number of samples in each batch.
• learning_rate: float - Step size for the optimizer.
• learning_rate_decay: float - Change of the learning rate in each epoch.
• optimizer: string = ["GradientDescent"|"Adam"] - Name of the optimizer to use.
• optimizer_params: Object - Further paramters for the optimizer to use.
• test_clustering_interval: int - Interval at which the VQK is used for clustering the testing data during the training process.
• testing_algorithms: Array = ["kmeans"|"spectral"|"dbscan"|"agglomerative"] - List of the names of clustering algorithms to use for testing.
• testing_algorithm_params: Array - List of objects. Each object is used as further parameters for each algorithm defined in testing_algorithms.