sequential.py

import math
import csv
import time
import numpy as np
import collections
from sklearn.cluster import KMeans
from sklearn.metrics.cluster import adjusted_rand_score
import matplotlib.pyplot as plt
import matplotlib.colors as colors
#===================================================Euclidian distance=====================
def eucl_distance(point_one, point_two): #define function tomeasure Euclidian distance between point_one and point_two
if(len(point_one) != len(point_two)): #check for an error
(len(1st point)is not equal len(2nd point))
raise Exception("Error: non comparable points") #check for an error
sum_diff = 0.0 #set variable sum_diff
for i in range(len(point_one)): #iterating take difference between each elem from 1st and 2nd point
diff = pow((float(point_one[i]) - float(point_two[i])), 2) #and square this difference.
sum_diff += diff #As the result summing all these differencess and equating to sum_diff
final = math.sqrt(sum_diff) #sqrt from sum_diff will be Euqlidian dist betw(1st and 2nd points)
return final
#===================================================Devide data set to further scattering=====================
global dimensions, num_clusters, num_points,dimensions,data,flag 
#turn on a timer which allows us to estimate performane of this algorithm
#===============================================reading and preparing data set======================
print("Enter the number of clusters you want to make: ")
num_clusters = input()
num_clusters = int(num_clusters)
start_time = time.time()
with open('3D_spatial_network.csv','r') as f:
    reader = csv.reader(f)
    data = list(reader)
data.pop(0)
for i in range (len(data)):
    data[i].pop(0)
data=np.array(data).astype(np.float)
data=data[0:10000]
#Print(data)
kmeans = KMeans(n_clusters=num_clusters, random_state=0).fit(data).labels_
# print('data',[ data[i] for i in [indices] ])
# data=np.array ([[2,10],[2,5],[8,4],[5,8],[7,5],[6,4],[1,2],[4,9]])
#====================================================================================================

#================================================Initialize centroids matrix=========================
initial=[]
for i in range(num_clusters):
 initial.append(data[i])
initial=np.vstack(initial)
#====================================================================================================
num_points = len(data) #number of rows
dimensions = len(data[0]) #number of columns 
#chunks = [ [] for _ in range(size) ]
#for i, chunk in enumerate(data):
# chunks[i % size].append(chunk)

#====================================================================================================
flag= True
while flag==True:
cluster=[]
#print str(rank) + ': ' + str(data)
#===================================Calculating dist matrix in each process==============================
dist =np.zeros((len(data),len(initial)))
for j in range(len(initial)):
for i in range(len(data)):
dist[i][j]=np.linalg.norm(initial[j]-data[i])
#print('dist',dist)
#===================================Initilize lable for each sample in each process======================
for i in range (len(dist)):
#iterable take each raw in dist matrix and
cluster.append(np.argmin(dist[i])+1) #find column index of min value (this index is number of centorud)
#print('clust vect',cluster)
#===================================Calculating the number of samples in each cluster====================
Q_clusts=collections.Counter(cluster)
#Q_clusts=np.array((collections.Counter(clusters).keys(),collections.Counter(clusters).values()))# what is the labels was faced in each process and their frequency
#=========================================================================================================
#====================================From each worker we gather cluster vector and join them ============
#==========================================================================================================
#if rank==0:
# cluster=[item for sublist in cluster for item in sublist]
# data = [item.tolist() for item in data]
# data=[item for sublist in data for item in sublist]
# print(data)
# print(cluster)
centroid=np.zeros((len(initial),len(initial[0])))
for k in range (1,num_clusters+1):
indices = [i for i, j in enumerate(cluster) if j == k]
#print('ind',indices)
#print(k, [data[i] for i in indices] )
#print('sum',np.sum([data[i] for i in indices], axis=0 ))
#print('div',np.divide((np.sum([data[i] for i in indices],axis=0)).astype(np.float),totcounter[k]))centroid[k-1]=np.divide((np.sum([data[i] for i in indices],axis=0)).astype(np.float),Q_clusts[k])
print('centroid',centroid)
print ('initial', initial)
print('centroid',centroid)
if np.all(centroid==initial):
flag=False
print ("Execution time %s seconds" % (time.time() - start_time))
else:
#print ('initial after', initial)
initial= centroid
# print('final clust vect',cluster)
print('libkms clust vect',kmeans)
print('adjusted_rand_score',adjusted_rand_score(kmeans,cluster))
# fig = plt.figure()
# scatter = fig.scatter([x],[y],c=[Cluster],s=50)