Signature
Parameters
callable
required
A function that transforms an (N, 2) or (N, 3) NumPy array of
float64 coordinates into an array of the same shape and dtype.
The function must preserve coordinate order and must be serializable
for execution on Spark workers.
bool
default:"False"
If False, discard the third dimension. If True, include the third
dimension in the coordinates passed to the function. Coordinates of
2D geometries have NaN in the third column.
Returns
GeoSeries
Transformed geometries with the original index and CRS. Null and
empty geometries are preserved.
Requires Spark 3.5 or newer and Shapely 2.0 or newer, including on the
workers. Execution is lazy; exceptions raised by the callback or invalid
returned arrays surface when Spark evaluates the result.Unlike local GeoPandas, the callback receives coordinates from one
Spark batch at a time, not from the entire GeoSeries. Batch sizes and
boundaries are unspecified, and Spark may execute the function multiple
times. Use deterministic, coordinate-wise transformations whose results
do not depend on batching or mutable state. For example,
lambda coords: coords + [1, 2] is independent of batching, while
lambda coords: coords - coords.mean(axis=0) centers each batch
separately. Compute global statistics first and pass the resulting
constants to the callback when a transformation needs them.This restriction is a caller responsibility; arbitrary Python callbacks
cannot be checked for dependence on batch boundaries. The result keeps
the input CRS; use to_crs for CRS-aware reprojection.