You have full access to this open access article. Neural architecture search NAS emerged as a way to automatically optimize neural networks for a specific task and dataset. Despite an abundance of research on NAS for images and natural language applications, similar studies for time series data are lacking.
Among NAS search spaces, chain-structured are the simplest and most applicable to small datasets like time series. We compare three popular NAS strategies on chain-structured search spaces: Bayesian optimization specifically Tree-structured Parzen Estimator , the hyperband method, and reinforcement learning in the context of financial time series forecasting.
We find Bayesian optimization and the hyperband method performing best among the strategies, and RNN and 1D CNN best among the architectures, but all methods were very close to each other with a high variance due to the difficulty of working with financial datasets.
We discuss our approach to overcome the variance and provide implementation recommendations for future users and researchers. Deep neural networks have been very successful in a wide variety of tasks over the last two decades. In large part their success is attributed to their ability to perform very well without major manual feature engineering required when compared to more classical techniques [ 1 ]. However, the exact architecture of the neural network still has to be prescribed manually by the user.
This led to the development of so-called auto-ML techniques that aim to automate this process. In the context of deep neural networks, auto-ML has a very large overlap with neural architecture search NAS , itself having a large overlap with hyperparameter optimization. A lot of research has been done in NAS in recent years, see [ 2 ] for an overview and insights from over papers. However, most work focused on computer vision or natural language applications, with less investigation into architectures for analyzing time series data.
