Rapidae: A Python Library for Creation, Experimentation, and Benchmarking of Autoencoder Models

Autoencoders have proven successful across diverse applications such as data reconstruction, anomaly detection, and feature extraction, however, these advancements remain largely dispersed among various fields, lacking a cohesive tool that unifies them. Herein, we present Rapidae, an open-source Python library designed to ease the use, development, and benchmarking of autoencoder models. Rapidae is oriented towards precise and imprecise data, mainly in the form of time series, and accommodates a broad range of user preferences and existing workflows as is backend-agnostic, supporting a seamless transition between TensorFlow, PyTorch, and JAX. In addition to its computational flexibility, the library provides a user-friendly graphical interface, lowering the entry barrier for newcomers and making it accessible for educational purposes. This paper delves into the core features of the library, underlining its user-centric design, and evaluates its impact on streamlining research processes. Rapidae is available at https://github.com/NahuelCostaCortez/rapidae and welcomes original contributions from other autoencoder research works.