DeepSensor Roadmap

This page contains a list of new features that we would like to add to DeepSensor in the future. Some of these have been raised as issues on the GitHub issue tracker with further details.

Note

We will soon create a GitHub project board to track progress on these items, which will provide a more up-to-date view of the roadmap.

Note

We are unable to provide a timetable for the roadmap due to maintainer time constraints. If you are interested in contributing to the project, check out our Contributing to DeepSensor page.

• Patch-wise training and inference

• Saving a TaskLoader when instantiated with raw xarray/pandas objects

• Non-Gaussian likelihoods

• Spatial-only modelling

• Continuous time measurements (i.e. not just discrete, uniformly sampled data on the same time grid)

• Improve forecasting functionality

• Test the framework with other models (e.g. GPs)

• Add simple baselines models (e.g. linear interpolation, GPs)

• Test and extend support for using dask in the DataProcessor and TaskLoader

• Infer linked context-target sets from the TaskLoader entries, don’t require user to explicitly specify links kwarg

• Improve unit test suite, increase coverage, test more edge cases, etc