pip install -i https://test.pypi.org/simple/ Lifting
For developer mode clone the repo and do
pip install -e .
See here: https://test.pypi.org/project/Lifting/
The lifting scheme is an adaptation of wavelet-based image compression to irregularly-shaped (i.e., non-square or -rectangular) domains. For a given input grid, it iteratively performs 3 steps till the grid is compressed to a single value called the scaling coefficient. The three steps are:
- Split: split the grid into pairs of 2 (and in case of an odd number of grid cells one triplet will exist).
- Predict: Assign x-y values to each pair/triple and then store the differences of x-y as the wavelet coefficients
- Update: update each pair/triple group's values with that of their mean.
Our adaption of the lifting scheme to spatio-temporal frameworks means that we can compress irregularly-shaped images into a single value across many different time-steps/samples. In modelling complex phenomena such as weather and climate, this allows one to extract a regional signal from which to model large-scale responses.
For example, one can compress the image of India as shown below for a single time step
And again for the Mediterranean
Amongst others, the lifting scheme has been used for climate model emulation. It's compression of multiple climate fields allows efficient joint, multivariate emulation (see MERCURY).