A project for hyperparameter selection with usage of Genetic Algorithm. All data in point cloud format is taken from A2D2 dataset. An aim of the solution is to automatize and optimize hyperparameter tuning of various classification algorithms, which is done manually nowadays. Genetic Algorithm was chosen for the project as a new and perspective way for optimization without ground truth data. EasyGA library is used for Genetic algorithm implementation, fitness function is based on existing clustering metrics
Chooses best hyperparameters
positional arguments:
lidar_path Directory where lidar files are stored
images_path Directory where images are stored
start_indx From which file algorithm should start
num_of_shots How many files are in the sequence
default Use default parameters or not
generation_goal Number of generations
population_size Number of chromosomes in generation
{1,2,3} Silhouette = 1, Calinski-Harabasz Index = 2, Davies-Bouldin Index = 3
{1,2,3,4} point clouds = 1, point clouds with between-clusters distances = 2, mapped images = 3, points selection mode = 4
verbose print side info or not