solves unsupervised learning quiz

Author: anishLearnsToCode

README.md

@@ -108,6 +108,7 @@ __Instructor__: Andrew Ng.
 
 ## Week 8
 ### Quizzes
+- [Unsupervised Learning](week8/unsupervised-learning-quiz.md)
 
 ### Programming Exercises
 

week8/assets/unsupervised-learning/question-1.PNG

@@ -0,0 +1,7 @@
+# Unsupervised Learning Quiz
+
+![Question 1](assets/unsupervised-learning/question-1.PNG)
+![Question 2](assets/unsupervised-learning/question-2.PNG)
+![Question 3](assets/unsupervised-learning/question-3.PNG)
+![Question 4](assets/unsupervised-learning/question-4.PNG)
+![Question 5](assets/unsupervised-learning/question-5.PNG)

week8/assets/unsupervised-learning/question-2.PNG

@@ -0,0 +1,88 @@
+clc;
+clear;
+
+function costs = clusterVariationCosts(X)
+  K = size(X, 1);
+  for i = 1:K
+    [~, minCost] = multipleKMeans(X, i, 5);
+    costs(i) =  minCost; 
+  endfor
+endfunction
+
+function [kMeans, minCost] = multipleKMeans(X, K, iters)
+  n = size(X, 2);
+  kMeans = zeros(iters, K, n);
+  cost = zeros(iters, 1);
+  for i = 1:iters
+    [clusters, J] = kMeansClustering(X, K);
+    kMeans(i, :, :) = clusters;
+    cost(i) = J;
+  endfor
+  [minCost, index] = min(cost);
+  kMeans = reshape(kMeans(index, :, :), K, n);
+endfunction
+
+function [clusters, minCost] = kMeansClustering(X, K)
+  m = size(X, 1);
+  n = size(X, 2);
+  
+  % randomly generate K clusters and positions
+  randomizedDataSet = randperm(m);
+  clusters = X(randomizedDataSet(1:K), :);
+  
+  % Run the K-means clustering algorithm
+  for iter = 1:10
+    pointGroup = distanceFromClusterPoints(X, K, clusters);
+    [minDistace, index] = min(pointGroup, [], 2);
+    cost = (1 / m) * sum(minDistace);
+    clusters = centroids(X, K, index, clusters);
+  endfor
+  minCost = cost;
+endfunction
+
+function mat = distanceFromClusterPoints(X, K, clusters)
+  m = size(X, 1);
+  mat = zeros(m, K);
+  for i = 1:K
+    mat(:, i) = euclideanDistanceSquare(X, clusters(i, :));
+  endfor
+endfunction
+
+function c = centroids(X, K, clusterData, clusters)
+  n = size(X, 2);
+  m = size(X, 1);
+  c = zeros(K, n);
+  frequency = zeros(K, 1);
+  for i = 1:m
+    frequency(clusterData(i))++;
+    c(clusterData(i), :) += X(i, :);
+  endfor
+  mask = c == 0;
+  frequency = maskZeroAsOne(frequency);
+  c = c ./ frequency;
+  c += mask .*  clusters;
+endfunction
+
+function mat = maskZeroAsOne(mat)
+  mask = mat == 0;
+  mat += mask;
+endfunction
+
+function d = euclideanDistanceSquare(X, cluster)
+  trainingDataSize = size(X, 1); 
+  similarityMatrix = repelem(cluster, trainingDataSize, 1);
+  difference = X - similarityMatrix;
+  d = sum(difference .^ 2, 2);
+endfunction
+
+labels = 5;
+data = [-10 1 ; 11 2 ; 45 3 ; 4 4 ; 7 5 ; 100 100 ; 5 -8 ; -89 23];
+% disp('K means clustering - single');
+% disp(kMeansClustering(data, labels));
+
+disp('multiple k means');
+[kMeans, minCost] = multipleKMeans(data, labels, 10);
+disp('min cost'); disp(minCost);
+disp('k means'); disp(kMeans);
+
+plot(clusterVariationCosts(data));