Ameyanagi/spdist

spdist: Simple metrics for comparing the distance between two curves.

spdist is a simple metrics for comparing the distance between two given curves. The curves can be passed in as a numpy array with discrete values. It will interpolate between the values and calculate the minimum distance between each points in the curve and reference curve.

pip install spdist

How to use.

Currently spdist has only one function spdist.

import spdist
import numpy as np

x = np.linspace(0, 10, 100)
y = 2*x

x_ref = x
y_ref = 2*x + 1

distance = spdist.spdist(x,y,x_ref,y_ref)

print(f"{distance}")

Example plot

Following example plot a straight line $y = 2x$ and a line with a constant offset. The distance between the two lines is calculated using the spdist function. The normal vector is calculated as $(2/\sqrt{5}, -1/\sqrt{5})$. Normal vector scaled by the distance is plot green line.

Algorithm

The algorithm of the caculation is somewhat brute force and the time complexity is $O(n^2)$. The algorithm is as follows:

distance = 0

for i in zip(x,y):

  tmp_distance = 0
  for j in zip(x_ref, y_ref):
    if (x_ref == x_ref_next) and (y_ref == y_ref_next):
        # point to point distance
        tmp_distance = min(tmp_distance, ((x - x_ref)**2 + (y - y_ref)**2)**0.5)
        continue
    # point to line distance
    # https://en.wikipedia.org/wiki/Distance_from_a_point_to_a_line#Line_defined_by_two_points
    tmp_distance = min(tmp_distance, ((x_ref_next - x_ref) * (y_ref - y) - (x_ref - x) * (y_ref_next - y_ref)) / ((x_ref_next - x_ref)**2 + (y_ref_next - y_ref)**2)**0.5)

  distance += tmp_distance

distance /= len(x)

Although the algorithm itself is not optimized, the whole library is written in Rust Rust with parallel processing. (Thanks to Rust's borrow checker and rayon rayon, which is both great work.) Therefore, the calculation is fast enough.

Where to use

This library was originally developed to calculate the distance between the measured spectra and the reference spectra of XAS. The metrics is useful to quantify and capture the features of the spectra.