sup3r.preprocessing.derivers.methods.SpatioTemporalEncoding

sup3r.preprocessing.derivers.methods.SpatioTemporalEncoding#

class SpatioTemporalEncoding[source]#

Bases: DerivedFeature

General positional or temporal encoding.

References

[cong2022]

Cong, Yezhen, et al. “Satmae: Pre-training transformers for

temporal and multi-spectral satellite imagery.” Advances in Neural Information Processing Systems 35 (2022): 197-211.

Note

We use the referenced paper for inspiration but modify the argument of the trig functions to be 2πi * k / d so the periodicity matches the underlying feature (e.g. hour of day repeats every 24 hours).

Methods

compute(data, **kwargs)

Compute method for derived feature.

Attributes

inputs

abstract classmethod compute(data: Sup3rX | Sup3rDataset, **kwargs)#

Compute method for derived feature. This can use any of the features contained in the xr.Dataset data and the attributes (e.g. .lat_lon, .time_index accessed through Sup3rX accessor).

Parameters:

  • data (Union[Sup3rX, Sup3rDataset]) – Initialized and standardized through a Loader with a specific spatiotemporal extent rasterized for the features contained using a Rasterizer.

  • kwargs (dict) – Optional keyword arguments used in derivation. height is a typical example. Could also be pressure.

Contents