timesift: Learn Predictive Representations of Time-Varying Data

Finds the time scale at which a long time-stamped record best predicts a response. timesift() builds the record at each grain or lookback named in sift, fits every learner on every representation it can read, scores them all on the same held-out folds and cells, and chooses or stacks them inside each training fold. The arrays themselves come from grain_matrix() and lookback_matrix() without the fitting layer. The shipped default is presence-absence with a joint multi-label head, scored by AUC with the true skill statistic beside it.

Details

Start with vignette("timesift"), or the articles at https://gillescolling.com/timesift/.

See also

Useful links:

• https://gillescolling.com/timesift/

• https://github.com/gcol33/timesift

• Report bugs at https://github.com/gcol33/timesift/issues

Author

Maintainer: Gilles Colling gilles.colling051@gmail.com (ORCID) [copyright holder]

Authors:

• Gilles Colling gilles.colling051@gmail.com (ORCID) [copyright holder]