PyData Amsterdam 2017
You are given access to an espresso machine with many buttons and knobs to tweak. Your task is to brew the best cup of espresso possible. How do you find the best settings before dying of a caffeine overdose? The answer is bayesian optimisation: the art of minimising an extremely expensive to evaluate function in as few calls as possible.
Bayesian optimisation can be applied to finding the optimal hyper-parameters for a deep neural network, optimizing the click-through-rate in online advertising or simulator settings of an optimal physics experiment. This talk will teach you about the basics of bayesian optimisation and how to use Scikit-Optimize to apply it to your own real world problems.
In addition to learning how to use this toolkit, you will also learn the answers to questions like:
When is bayesian optimisation useful? What is bayesian about bayesian optimisation? When can a statistical model be used for bayesian optimisation? What is an acquisition function and how does it help to exploit / explore the search space?
We will end the talk with an example to optimize hyperparameters of a neural-network using bayesian optimisation.