Repository navigation
Using xarray to make transformations less ambiguous #47
Description
Cost/benefit analysis of using xarray to annotate the dimensions of vectors (see code examples below).
Pro:
- would make transformations less ambiguous
- possibility to apply transformations to points of any shapes, for instance (n, 2) and (2, n).
Cons:
- extra verbosity for the use
Current approach: initially I am using numpy arrays for the vectors describing points and performing a shape check (e.g. 2D translations can't be applied to a set of points stored as a numpy array of shape (2, n), but only with a shape (n, 2) (excluding edge cases like n=2).
Later we can discuss the preferred approach.
Code examples from @ivirshup of using xarray.DataArray to annotate the dimensions of vectors.
import xarray as xr
da = xr.DataArray([1,2,3], coords={"dim": ["x", "y", "z"]})
da.sel({"dim": "x"})
<xarray.DataArray ()>
array(1)
Coordinates:
dim <U1 'x'
import xarray as xr, numpy as np
N_POINTS = 10
points = xr.DataArray(
np.random.random((N_POINTS, 3)),
coords={
"points": range(N_POINTS),
"dim": ["x", "y", "z"],
},
)
points
<xarray.DataArray (points: 10, dim: 3)>
array([[0.97379558, 0.29238395, 0.37392129],
[0.62459118, 0.13214354, 0.0514482 ],
[0.41742495, 0.59828447, 0.79762354],
[0.87583248, 0.71631053, 0.30032732],
[0.93012066, 0.73074886, 0.46370001],
[0.74505332, 0.84503157, 0.52606046],
[0.85789939, 0.44050558, 0.23522551],
[0.35954281, 0.11675383, 0.23321456],
[0.99866969, 0.83124205, 0.55161158],
[0.25951685, 0.09205387, 0.00736066]])
Coordinates:
* points (points) int64 0 1 2 3 4 5 6 7 8 9
* dim (dim) <U1 'x' 'y' 'z'
translation = xr.DataArray([0, 1, 0], coords={"dim": ["x", "y", "z"]})
points + translation
Activity
ivirshup commentedon Dec 12, 2022
How could this interact with shapely/ geopandas? It looks like geopandas and shapely effectively hardcode x, y, z
LucaMarconato commentedon Dec 12, 2022
Good catch. I'll think about it. I think that #39) is also related (more details in the linked issue).
LucaMarconato commentedon Feb 13, 2023
The transformations are now using this approach internally to apply the transformations.