The examples in this folder all illustrate ways to leverage the framework in actual situations. Emphasis is on end-to-end solution notebooks including "all the glue stuff" required to make it work.
- Closed-loop optimization of known function
- Iterative optimization of a function, which can be sampled
- Maximize a multivariate function using either closed-loop or iterative approach
- Optimize a noisy, multivariate system
- Stop optimization early when convergence seems to be reached based on relative improvement in best response
- Functions with integer covariates
- Make the best brownie
In this example, we use the .auto-method to perform closed-loop maximization of a known univariate function.
File: Example 1 - Closed-loop optimization of known function.ipynb
Here, the iterative solution approach accessible via the .ask and .tell methods is used to find the maximum of a univariate function, which can be sampled. The example makes use of a well-defined function, but any samplable function could be used (also e.g. the readout from a physical system).
File: Example 2 - Iterative optimization of function, which can be sampled.ipynb
In this example, a variant of the multivariate Easom function is maximized using both the closed-loop approach of the .auto-method as well as the iterative approach of .ask and .tell methods. Similar to Example 2, the response function need not be explicitly defined for the iterative .ask-.tell methods to work, as long as the function can be sampled, the framework can be applied.
Towards the end, one of the two covariates (
File: Example 3 - Maximum of a multivariate function.ipynb
Here a noisy multivariate response function is optimized. This illustrates how to apply Bayesian optimization for noisy systems such as those found in many real-world applications. The noise could either be inherent in the system itself or be a measurement uncertainty.
File: Example 4 - Optimize a noisy multivariate system.ipynb
Example 5: Stop optimization early when convergence seems to be reached based on relative improvement in best response
In this example illustrate how to use rel_tol and rel_tol_steps-parameters in .auto-method to stop iterations before the specified number of iterations if the solution is deemed converged. This is assessed by looking at the relative improvement in best response between consecutive iterations. rel_tol defines the relative improvement threshold required for convergence; also setting rel_tol_steps requires that this threshold is reached for rel_tol_steps consecutive iterations.
The example discusses best practise for using these functionalities to obtain good convergence.
File: Example 5 - Optimization stopping criteria based on relative improvements.ipynb
In this example it is shown how to define integer covariates and illustrates how the framework handles cases of mixed covariate data types (some continuous, some integer) better than pure integer covariates.
File: Example 6 - Functions with integer covariates.ipynb
This example illustrates how to combine covariates of all the different data types (continuous, integer, categorical) to solve a complex and somewhat relevant real-life problem: finding the best brownie recipe!
File: Example 7 - Make the best brownie.ipynb