Keywords: Automated machine learning, interpretable machine learning, partial dependence plots
Abstract: Automated hyperparameter optimization (HPO) can support practitioners to obtain peak performance in machine learning models. However, there is often a lack of valuable insights into the effects of different hyperparameters on the final model performance. This lack of comprehensibility and transparency makes it difficult to trust and understand the automated HPO process and its results. We suggest using interpretable machine learning (IML) to gain insights from the experimental data obtained during HPO and especially discuss the popular case of Bayesian optimization (BO). BO tends to focus on promising regions with potential high-performance configurations and thus induces a sampling bias. Hence, many IML techniques, like Partial Dependence Plots (PDP), carry the risk of generating biased interpretations. By leveraging the posterior uncertainty of the BO surrogate model, we introduce a variant of the PDP with estimated confidence bands. In addition, we propose to partition the hyperparameter space to obtain more confident and reliable PDPs in relevant sub-regions. In an experimental study, we provide quantitative evidence for the increased quality of the PDPs within sub-regions.
Ethics Statement: We consider our work as core research in the field of AutoML. In particular, we do not address any specific application which may expose any ethical concerns. We emphasize that our work aims to increase transparency in AutoML and foster a human-centered approach, which in our opinion can, in general, promote the realization of ethical principles.
Crc Pdf: pdf
Poster Pdf: pdf
Original Version: pdf