Abstract Supervised learning of behavioural modes from body‐worn sensor data, especially accelerometers, has become a transformative research tool in behavioural ecology over the past years. Due to the popularity of the methodology and diverging needs of users, there are a number of software packages dedicated to it, ranging from web based graphical user interfaces to R software packages. In pyecoacc, we aim to augment the functionality of the existing software by integrating recent methodological findings and recommendations. pyecoacc is an open‐source Python package for supervised learning of behavioural modes from accelerometer data. It is designed to work with minimum configuration, while remaining flexible enough to accommodate customization, additions and extensions. The pyecoacc software package includes the common accelerometer feature computations that have become standard in the field, and pipelines for traditional as well as deep learning‐based models. Model selection is facilitated via simple comparison tables with the recommended metrics. The correct computation of behavioural time budgets with the confusion matrix correction is also supported. We demonstrate the software using a dataset of body acceleration of a rodent species (Damaraland mole‐rat, Fukomys damarensis ).
Resheff et al. (Fri,) studied this question.