Welcome to Odyssey docs!
Odyssey is a FastAPI service that exposes Bayesian optimisation (BO) as HTTP
endpoints for self-driving laboratories (SDLs) and materials-discovery
campaigns. It wraps BoTorch/GPyTorch
Gaussian-process models behind a multipart/form-data API, so any client
that can make an HTTP request can ask "what should I try next?" without
depending on the Python BO stack directly.
Odyssey provides:
- Single-objective suggestion (
/single/suggest) — a Gaussian-process optimiser with selectable acquisition functions (qLogNEI,qLogEI,PI,UCB). - Multi-objective suggestion (
/mobo/suggest,/mobo/qNEHVI-suggest) — qNEHVI/qNParEGO-based optimisers returning a next sample and the current Pareto front. - Space-filling design (
/design/{sobol,lhc}) — Sobol/Latin-hypercube sampling with optional discrete-step and sum-to-total constraints. - Mixture / simplex design and fitting (
/simplex/design,/simplex/fit,/simplex_model) — Scheffé simplex-lattice designs and polynomial response-surface fitting. - Visualisation endpoints (
/umap_overview,/parallel_coords,/pairwise_grid,/convergence) — Plotly views of the parameter space, GP posterior, and optimisation progress.