Key Features

What you can do

Define-by-run API

Allows dynamic and Pythonic construction of hyperparameter search spaces using conditionals and loops during code execution.

Efficient Optimization Algorithms

Includes algorithms such as TPE, CMA-ES, GP-based Bayesian optimization, NSGA-II for multi-objective optimization, and pruning of unpromising trials.

Parallelization

Supports parallel execution of optimization trials across tens or hundreds of workers with minimal code changes.

Integrations

Compatible with machine learning libraries including PyTorch, TensorFlow, XGBoost, LightGBM, Keras, Catboost, MLflow, and Weights & Biases.

Optuna Dashboard

Provides real-time visualization of optimization history and hyperparameter importance through graphs and tables.