Skip to main content

Hyperparameter Optimisation

Hyperparameter optimization fine-tunes machine learning models by systematically searching for the best combination of hyperparameters to enhance performance.

We provide built-in support for hyperparameter optimization using predefined default settings, derived through extensive experimentation. Additionally, users can manually fine-tune model hyperparameters.

We offer three hyperparameter optimization algorithms, detailed below.

Select Hyperparameter Optimisation in evoML​

  1. Create a New Trial

  2. Under Task Options choose Optimizer. The default hyperparameter optimisation strategy applied is Tree-Structured Parzen Estimator.

    optimizer

Hyperparameter Optimisation algorithms​

1. Tree-Structured Parzen Estimator (TPE)​

TPE is a Bayesian optimisation algorithm for global optimisation problems that models the search space as a tree-structured probability density function and iteratively estimates the optimal region in order to improve solution accuracy.

2. Non-Dominated Sorting Genetic Algorithm II (NSGA-II)​

NSGA-II is an evolutionary algorithm that simultaneously optimises multiple objectives by selecting, mutating and recombining candidates based on their performance and diversity, driving the search toward the best solution.

3. Random​

Random search is an optimisation algorithm that randomly samples the search space making it simple and effective in high-dimensional spaces. However, it may be less efficient than other optimisation methods since it does not utilise any information about the quality of previous solutions to guide its search.