haversine_pytorch

The pairwise distance calculation between two sets of points with Haversine formula for CPU/GPU vectorized in PyTorch.

The function calculates the pairwise distances between point in two lists in meters with >99.5% accuracy. PyTorch vectorized implementation allows usage in CPU and GPU.

Comaprison

haversine_torch

Wall time ~5.9s:
    - CPU 
    - 568m pairs
    - test sample 8,154*69,600

Wall time ~10.7s:
     - CPU
     - 1.1b
     - test sample 8,154*132,000

sklearn.metrics.pairwise.haversine_distances

Wall time ~37.4s: 
     - CPU
     - 568m pairs
     - test sample 8,154*69,600

Example

l = [[1.01, 1.01], [2.01, 2.01]]
r = [[2.01, 2.01], [3.01, 3.01]]
dist_martix = haversine_torch(l,r)

Parameters

left: List[List[float]] or List[Tuple[float]]
    List of lists or tuples with coordinates with [latitude, longitude]
right: List[List[float]] or List[Tuple[float]]
    List of lists or tuples with coordinates with [latitude, longitude]
matrix: bool
    If you want to receive the matrix of left*right or tensor of length left*right. Default True.
R: int
    Radius of planet where you wnat to measure distance. Default value = 6371000 according to the radius of the Earth.