EmuKit/emukit

What it solves

Emukit addresses the challenge of decision making under uncertainty, specifically for complex systems where data is scarce or expensive to acquire. It provides tools to ensure that limited computational or experimental resources are used efficiently by propagating well-calibrated uncertainty estimates through design loops.

How it works

Emukit acts as a model-agnostic toolkit that integrates with the Python ecosystem. It allows users to build surrogate models (emulators) to approximate expensive functions. It supports various techniques including multi-fidelity emulation (combining data from sources with different costs/accuracy), Bayesian optimization for parameter tuning, and Bayesian quadrature for efficient integration.

Who it’s for

It is designed for researchers and engineers working with complex physical processes or machine learning algorithms where evaluating the system is costly and requires strategic experimental design.

Highlights

• Multi-fidelity emulation: Combines information from multiple data sources with varying fidelity and cost.
• Bayesian optimisation: Used for tuning ML parameters or optimizing physical experiments.
• Experimental design: Tools for active learning and designing the most informative experiments.
• Sensitivity analysis: Analyzes how inputs influence system outputs.
• Bayesian quadrature: Efficiently computes integrals of expensive functions.
• Framework agnostic: Compatible with various modeling tools in the Python ecosystem.