-
-
Save dexhunter/04719fd58fd45ea78f8aa9407dc330d3 to your computer and use it in GitHub Desktop.
Computing the geometric median in Python by Daniel J Lewis
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
copied from a poorly-formatted PDF: | |
http://danieljlewis.org/files/2010/07/geomedian.pdf | |
http://danieljlewis.org/2010/07/09/computing-the-geometric-median-in-python/ | |
and fixed the formatting for others to use | |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import math | |
from dbfpy import dbf | |
import matplotlib.pyplot as plt | |
#create a plot | |
fig = plt.figure(1, figsize = [10,10], dpi=90) | |
axScatter = plt.subplot(111) | |
def candMedian(dataPoints): | |
#Calculate the first candidate median as the geometric mean | |
tempX = 0.0 | |
tempY = 0.0 | |
for i in range(0,len(dataPoints)): | |
tempX += dataPoints[i][0] | |
tempY += dataPoints[i][1] | |
return [tempX/len(dataPoints),tempY/len(dataPoints)] | |
def numersum(testMedian,dataPoint): | |
# Provides the denominator of the weiszfeld algorithm depending on whether you are adjusting the candidate x or y | |
return 1/math.sqrt((testMedian[0]-dataPoint[0])**2 + (testMedian[1]-dataPoint[1])**2) | |
def denomsum(testMedian, dataPoints): | |
# Provides the denominator of the weiszfeld algorithm | |
temp = 0.0 | |
for i in range(0,len(dataPoints)): | |
temp += 1/math.sqrt((testMedian[0] - dataPoints[i][0])**2 + (testMedian[1] - dataPoints[i][1])**2) | |
return temp | |
def objfunc(testMedian, dataPoints): | |
# This function calculates the sum of linear distances from the current candidate median to all points | |
# in the data set, as such it is the objective function we are minimising. | |
temp = 0.0 | |
for i in range(0,len(dataPoints)): | |
temp += math.sqrt((testMedian[0]-dataPoints[i][0])**2 + (testMedian[1]-dataPoints[i][1])**2) | |
return temp | |
# Use the above functions to calculate the median | |
# Test Data - later to be read from a file | |
#dataPoints = [[3,4],[3,3],[6,8],[9,3],[3,5],[1,2],[6,3]] | |
# Data read from dbf file exported, and randomly offset, from ArcGIS 9.3 | |
dataPoints = [] | |
filestring = r"C:\Users\Daniel\Documents\AddressLayerSwk\AddressDBFs\TestMediansBradford.dbf" | |
db = dbf.Dbf(filestring) | |
for i in range(0,len(db)): | |
rec = db[i] | |
x = float(rec['easting']) | |
y = float(rec['northing']) | |
dataPoints.append([x,y]) | |
# add data points to scatter plot | |
axScatter.scatter(x,y) | |
axScatter.set_aspect(1.) | |
# Create a starting 'median' | |
testMedian = candMedian(dataPoints) | |
print testMedian | |
#add mean to scatter plot | |
axScatter.scatter(testMedian[0],testMedian[1],s=150,color='green', marker='x') | |
# numIter depends on how long it take to get a suitable convergence of objFunc | |
numIter = 50 | |
#minimise the objective function. | |
for x in range(0,numIter): | |
print objfunc(testMedian,dataPoints) | |
denom = denomsum(testMedian,dataPoints) | |
nextx = 0.0 | |
nexty = 0.0 | |
for y in range(0,len(dataPoints)): | |
nextx += (dataPoints[y][0] * numersum(testMedian,dataPoints[y]))/denom | |
nexty += (dataPoints[y][1] * numersum(testMedian,dataPoints[y]))/denom | |
testMedian = [nextx,nexty] | |
# add final median to scatter plot (to see progression add this line into the loop) | |
axScatter.scatter(testMedian[0],testMedian[1],s=150,color='red', marker='x') | |
#create a legend for plot | |
leg = axScatter.legend(('Mean Centre','Median Centre','Data Points'),scatterpoints = 1) | |
print testMedian | |
plt.show() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment