I have been playing with two dimensional machine learning using pandas (trying to do something like this), and I would like to combine Lat/Long into a single numerical feature -- ideally in a linear fashion. Is there a "best practice" to do this?


A note: for those who've ended here looking for a hashing technique, geohash is likely your best choice.

Representing latitude and longitude in a single linear scale is not possible due to the fact that their domain is inherently a 3D space. Reducing that as per your needs would require a spatial flattening technique that's unheard of to me.


As far as lat/long merging goes, the best of best practices would be to resort to the Haversine formula, which calculates the distance between two points over a spherical surface, and receives those points' coordinates as input.

One way to incorporate that in your use case - where each point should probably have an independent lat/long combination - would be to assume the distance's origin point coordinates to be $(\varphi_1, \lambda_1) = (0, 0)$, which would render

$$d =2r \arcsin\left(\sqrt{\sin^2\left(\frac{\varphi_2 - 0}{2}\right) + \cos(0) \cos(\varphi_2)\sin^2\left(\frac{\lambda_2 - 0}{2}\right)}\right)$$

$$= 2r \arcsin\left(\sqrt{\sin^2\left(\frac{\varphi_2}{2}\right) + \cos(\varphi_2)\sin^2\left(\frac{\lambda_2}{2}\right)}\right)$$

With $r$ being Earth's radius (~6371km) and $(\varphi_2, \lambda_2)$ your point's latitude and longitude, respectively.

However, as stated before, that couldn't possibly give you a linear relation, as you can see by 3d plotting the function: Relativized Haversine 3D Plot


The circumstances imply you're likely to be using pandas, or at least should be. Here's an example implementation of this relativized Haversine formula:

from math import radians, cos, sin, asin, sqrt

def single_pt_haversine(lat, lng, degrees=True):
    'Single-point' Haversine: Calculates the great circle distance
    between a point on Earth and the (0, 0) lat-long coordinate
    r = 6371 # Earth's radius (km). Have r = 3956 if you want miles

    # Convert decimal degrees to radians
    if degrees:
        lat, lng = map(radians, [lat, lng])

    # 'Single-point' Haversine formula
    a = sin(lat/2)**2 + cos(lat) * sin(lng/2)**2
    d = 2 * r * asin(sqrt(a)) 

    return d

Which could be used as in the below minimal example:

>>> import pandas as pd

>>> df = pd.DataFrame([[45.0, 120.0], [60.0, 30.0]], columns=['x', 'y'])
>>> df
      x      y
0  45.0  120.0
1  60.0   30.0

>>> df['harvesine_distance'] = [single_pt_haversine(x, y) for x, y in zip(df.x, df.y)]
>>> df
      x      y  harvesine_distance
0  45.0  120.0        12309.813344
1  60.0   30.0         7154.403197
  • 1
    $\begingroup$ Thank you for this great response. It helps me understand this problem a lot better. Yes, I'm using pandas. $\endgroup$ Apr 30 '19 at 13:36
  • 1
    $\begingroup$ You're welcome! Feel free to accept it if it has solved your question - finding the best practice in dealing with lat-lng distances. $\endgroup$ May 1 '19 at 13:06

The best practice is to not attempt to flatten Earth into a onee dimensional line... Because as you may know, Earth more resembles a sphere than a line. It is much better to treat it as such properly.

There do exist approaches to flatten a k-dimensional space into a one dimensional order though. These are known as space filling curves and are from the 19th century. Their limitations are well understood: for many points they will work quite well - but in other locations they work really badly. As known from complex number theory, you cannot find a good linear order of a plane.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.