poplar: Accelerating selection bias modelling with machine learning

poplar is a lightweight package for performing selection bias modelling with machine learning. It is fully implemented with pytorch.

It is best-suited to problems where the selection process can only be modelled at a high computational cost, and is efficient and accurate even at high dimensionality. It has been applied to the modelling of gravitational wave selection biases, specifically extreme mass ratio inspiral (EMRI) sources observable by the Laser Interferometer Space Antenna (LISA) detector.

If you find poplar useful in your work, please cite both Chapman-Bird et al. (2023) and the package doi.

Changes in v0.2.0:

• Fixed a bug that prevented loading of a saved LinearModel on a machine with no GPU available in the slot that the model was originally saved on. Now, models are always moved to CPU prior to pickling.
• Fixed a bug when using the IdentityRescaler that prevented moving of models to GPU
• Some documentation changes and other minor bug fixes.