Kmeans python file

import csv

import numpy as np

import matplotlib.pyplot as plt

# Computes the distance between two data points

def calc_distance(X1, X2):

return(np.sum((X1 - X2)**2))**0.5

# Function to read data from the csv files

# Hint 1: Remember that CSV files are comma separated, so you should use a "," as a delimiter.

# Hint 2: Ensure you are reading the csv file in the correct mode.

def read_csv(filename):

with open(filename) as fh:

rd = csv.DictReader(fh, delimiter=',')

data =[]

for row in rd:

data.append(row)

values = [list(dictionary.values())[1:] for dictionary in data]

# Convert the data into float for processing

values = [list(map(float,value)) for value in values]

return values

# Define a function that finds the closest centroid to each point out of all the centroids

# Hint: This function should call the function you implemented that computes the distance between two data points.

# Hint: Numpy has a useful method that allows you to find the index of the smallest value in an array.

def closest_centroid(data,centroids):

# assign centroid to data points by using a dictionary with centroid index as key and its data points as value

assigned_centroid = {}

for j in range(len(centroids)):

assigned_centroid[j] = []

for i in data:

distance =[]

# calculate distance to all centroids

for j in centroids:

distance.append(calc_distance(np.array(i),np.array(j)))

# Use numpy argmin the usefull method to find index of smallest value in distance array which corresponds to index of centroid

assigned_centroid[np.argmin(distance)].append(i)

return assigned_centroid

#Make a function to visualise the clusters. (optional, but useful to see the changes and if your algorithm is working)

def plot_clusters(assign_data,centroids,iter):

plt.figure()

plt.scatter(np.array(centroids)[:, 0], np.array(centroids)[:, 1], color='black')

for i in range(len(centroids)):

plt.scatter(np.array(assign_data[i])[:,0],np.array(assign_data[i])[:,1],alpha = 0.2)

# plt.scatter(np.array(data)[:, 0], np.array(data)[:, 1], alpha=0.1)

plt.xlabel("Birth rate")

plt.ylabel("Life Expectancy")

plt.title(f"Centroids and clusters in iteration = {iter+1}")

plt.savefig(f"Cluster_{iter}.png")

print(f"Plot has been saved to Cluster_{iter}.png")

# plt.show()

# Make the initialisation procedure

# Dataset you want to analyse

filename = 'dataBoth.csv'

# Define number of clusters k

numClusters = 3

while True:

try:

numClusters = int(input("Please enter the number of Clusters "))

break

except ValueError:

print("Please enter a valid number of clusters")

# No of iterations

max_iter =6

while True:

try:

max_iter = int(input("Please enter the number of iterations "))

break

except ValueError:

print("Please enter a valid number of clusters")

# get data and initalize centroids

X = read_csv(filename)

centroids_idx = np.random.choice(len(X),numClusters)

centroids =[]

for idx in centroids_idx:

centroids.append(X[idx])

# Implement the k-means algorithm, using appropriate looping for the number of iterations

# --- find the closest centroid to each point and assign the point to that centroid's cluster

# --- calculate the new mean of all points in that cluster

# --- visualize (optional, but useful to see the changes)

#---- repeat

for i in range(max_iter):

# Closest centroid to each points

assign_centroids = closest_centroid(X,centroids)

# Calculate new mean and update centroids_idx

new_centroids =[np.mean(assign_centroids[centroid],axis=0) for centroid in assign_centroids.keys()]

# Visualize cluster for each iterations

plot_clusters(closest_centroid(X,new_centroids),new_centroids,i)

centroids = new_centroids

# ====

# Print out the results for questions

#1) The number of countries belonging to each cluster

#2) The list of countries belonging to each cluster

#3) The mean Life Expectancy and Birth Rate for each cluster

# Final Data and their clusters

final_cluster_data = closest_centroid(X,centroids)

# Print the required in the question

for i in range(len(centroids)):

print(f"Number of countries in cluster with centroid {centroids[i]} = {len(final_cluster_data[i])}")

# Get countries data

data =[]

with open(filename) as fh:

rd = csv.DictReader(fh, delimiter=',')

for row in rd:

data.append(list(row.values()))

for i in range(len(centroids)):

print(f" Countries in cluster with centroid {centroids[i]} ")

for j in final_cluster_data[i]:

country = [values[0] for values in data if values[1]==str(j[0]) and values[2] == str(j[1])]

if len(country)>0:

print(country[0])

for i in range(len(centroids)):

print(f"The mean Life Expectancy and Birth Rate for cluster with centroid {centroids[i]} = {round(np.mean(final_cluster_data[i],axis=0)[1],3)},{round(np.mean(final_cluster_data[i],axis=0)[0],3)}")

Datas

data 1 1953: https://docs.google.com/spreadsheets/d/1YwXeOPB_NGu3Eq7-TDmaeX4L-yGJYOq9C8TyQsrj_9s/edit?usp=sharing

data 2 2008: https://docs.google.com/spreadsheets/d/1USFlYnM0UkOHJTMwpSMu4g88bpsGWspGs7-WAIR2qMw/edit?usp=sharing

data both: https://docs.google.com/spreadsheets/d/1KjL1rdbrZIbKZQEfALRZ9-utJdI6V6GFVqfr6z7sNo4/edit?usp=sharing

To change the dataset, search for 'dataBoth.csv' and change the filename variable in the initialization section (on Line 58).

cluster 3 - centroids and clusters in iteration = 4

cluster 4 = centroids and clusters in iteration = 5

Compulsory Task 2 Follow these steps: - Create a text file called interpretation.txt and fill in your answers to the questions below. - Run your kmeans.py file using 3 clusters on the 1953 and 2008 datasets separately. Document your observations in your text file. Take note of how the clusters change from 1953 to 2008 . You will need to pay attention not only to which countries are in clusters together but also to the Life Expectancy and BirthRates for those clusters. - Next, run the algorithm with 4 clusters on dataBoth.csv. Note any observations in your text file. Which countries are moving up clusters? How does the data from 2008 compare to the data from 1958 ? Are there any 2008 countries that are in a cluster that is made up mostly of 1953 countries? Try and explain why. Are there any 1953 countries that are in a cluster that is made up of mostly 2008 countries? Try and explain why in your text file