AutoRL Hyperparameter Landscapes

Aditya Mohan, Carolin Benjamins, Konrad Wienecke, Alexander Dockhorn, Marius Lindauer
Proceedings of the Second International Conference on Automated Machine Learning, PMLR 224:13/1-27, 2023.

Abstract

Although Reinforcement Learning (RL) has shown to be capable of producing impressive results, its use is limited by the impact of its hyperparameters on performance. This often makes it difficult to achieve good results in practice. Automated RL (AutoRL) addresses this difficulty, yet little is known about the dynamics of the hyperparameter landscapes that hyperparameter optimization (HPO) methods traverse in search of optimal configurations. In view of existing AutoRL approaches dynamically adjusting hyperparameter configurations, we propose an approach to build and analyze these hyperparameter landscapes not just for one point in time but at multiple points in time throughout training. Addressing an important open question on the legitimacy of such dynamic AutoRL approaches, we provide thorough empirical evidence that the hyperparameter landscapes strongly vary over time across representative algorithms from RL literature (DQN, PPO and SAC) in different kinds of environments (Cartpole, Bipedal Walker and Hopper). This supports the theory that hyperparameters should be dynamically adjusted during training and shows the potential for more insights on AutoRL problems that can be gained through landscape analysis. Our code can be found at \url{https://anon-github.automl.cc/r/autorl_landscape-F04D}.

Cite this Paper

BibTeX
@InProceedings{pmlr-v224-mohan23a, title = {AutoRL Hyperparameter Landscapes}, author = {Mohan, Aditya and Benjamins, Carolin and Wienecke, Konrad and Dockhorn, Alexander and Lindauer, Marius}, booktitle = {Proceedings of the Second International Conference on Automated Machine Learning}, pages = {13/1--27}, year = {2023}, editor = {Faust, Aleksandra and Garnett, Roman and White, Colin and Hutter, Frank and Gardner, Jacob R.}, volume = {224}, series = {Proceedings of Machine Learning Research}, month = {12--15 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v224/mohan23a/mohan23a.pdf}, url = {https://proceedings.mlr.press/v224/mohan23a.html}, abstract = {Although Reinforcement Learning (RL) has shown to be capable of producing impressive results, its use is limited by the impact of its hyperparameters on performance. This often makes it difficult to achieve good results in practice. Automated RL (AutoRL) addresses this difficulty, yet little is known about the dynamics of the hyperparameter landscapes that hyperparameter optimization (HPO) methods traverse in search of optimal configurations. In view of existing AutoRL approaches dynamically adjusting hyperparameter configurations, we propose an approach to build and analyze these hyperparameter landscapes not just for one point in time but at multiple points in time throughout training. Addressing an important open question on the legitimacy of such dynamic AutoRL approaches, we provide thorough empirical evidence that the hyperparameter landscapes strongly vary over time across representative algorithms from RL literature (DQN, PPO and SAC) in different kinds of environments (Cartpole, Bipedal Walker and Hopper). This supports the theory that hyperparameters should be dynamically adjusted during training and shows the potential for more insights on AutoRL problems that can be gained through landscape analysis. Our code can be found at \url{https://anon-github.automl.cc/r/autorl_landscape-F04D}.} }
Endnote
%0 Conference Paper %T AutoRL Hyperparameter Landscapes %A Aditya Mohan %A Carolin Benjamins %A Konrad Wienecke %A Alexander Dockhorn %A Marius Lindauer %B Proceedings of the Second International Conference on Automated Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Aleksandra Faust %E Roman Garnett %E Colin White %E Frank Hutter %E Jacob R. Gardner %F pmlr-v224-mohan23a %I PMLR %P 13/1--27 %U https://proceedings.mlr.press/v224/mohan23a.html %V 224 %X Although Reinforcement Learning (RL) has shown to be capable of producing impressive results, its use is limited by the impact of its hyperparameters on performance. This often makes it difficult to achieve good results in practice. Automated RL (AutoRL) addresses this difficulty, yet little is known about the dynamics of the hyperparameter landscapes that hyperparameter optimization (HPO) methods traverse in search of optimal configurations. In view of existing AutoRL approaches dynamically adjusting hyperparameter configurations, we propose an approach to build and analyze these hyperparameter landscapes not just for one point in time but at multiple points in time throughout training. Addressing an important open question on the legitimacy of such dynamic AutoRL approaches, we provide thorough empirical evidence that the hyperparameter landscapes strongly vary over time across representative algorithms from RL literature (DQN, PPO and SAC) in different kinds of environments (Cartpole, Bipedal Walker and Hopper). This supports the theory that hyperparameters should be dynamically adjusted during training and shows the potential for more insights on AutoRL problems that can be gained through landscape analysis. Our code can be found at \url{https://anon-github.automl.cc/r/autorl_landscape-F04D}.
APA
Mohan, A., Benjamins, C., Wienecke, K., Dockhorn, A. & Lindauer, M.. (2023). AutoRL Hyperparameter Landscapes. Proceedings of the Second International Conference on Automated Machine Learning, in Proceedings of Machine Learning Research 224:13/1-27 Available from https://proceedings.mlr.press/v224/mohan23a.html.

Related Material