|
| 1 | +importnumpy |
| 2 | +importmatplotlib.pyplot |
| 3 | +importpygad |
| 4 | + |
| 5 | +cluster1_num_samples=20 |
| 6 | +cluster1_x1_start=0 |
| 7 | +cluster1_x1_end=5 |
| 8 | +cluster1_x2_start=2 |
| 9 | +cluster1_x2_end=6 |
| 10 | +cluster1_x1=numpy.random.random(size=(cluster1_num_samples)) |
| 11 | +cluster1_x1=cluster1_x1* (cluster1_x1_end-cluster1_x1_start)+cluster1_x1_start |
| 12 | +cluster1_x2=numpy.random.random(size=(cluster1_num_samples)) |
| 13 | +cluster1_x2=cluster1_x2* (cluster1_x2_end-cluster1_x2_start)+cluster1_x2_start |
| 14 | + |
| 15 | +cluster2_num_samples=20 |
| 16 | +cluster2_x1_start=4 |
| 17 | +cluster2_x1_end=12 |
| 18 | +cluster2_x2_start=14 |
| 19 | +cluster2_x2_end=18 |
| 20 | +cluster2_x1=numpy.random.random(size=(cluster2_num_samples)) |
| 21 | +cluster2_x1=cluster2_x1* (cluster2_x1_end-cluster2_x1_start)+cluster2_x1_start |
| 22 | +cluster2_x2=numpy.random.random(size=(cluster2_num_samples)) |
| 23 | +cluster2_x2=cluster2_x2* (cluster2_x2_end-cluster2_x2_start)+cluster2_x2_start |
| 24 | + |
| 25 | +cluster3_num_samples=20 |
| 26 | +cluster3_x1_start=12 |
| 27 | +cluster3_x1_end=18 |
| 28 | +cluster3_x2_start=8 |
| 29 | +cluster3_x2_end=11 |
| 30 | +cluster3_x1=numpy.random.random(size=(cluster3_num_samples)) |
| 31 | +cluster3_x1=cluster3_x1* (cluster3_x1_end-cluster3_x1_start)+cluster3_x1_start |
| 32 | +cluster3_x2=numpy.random.random(size=(cluster3_num_samples)) |
| 33 | +cluster3_x2=cluster3_x2* (cluster3_x2_end-cluster3_x2_start)+cluster3_x2_start |
| 34 | + |
| 35 | +c1=numpy.array([cluster1_x1,cluster1_x2]).T |
| 36 | +c2=numpy.array([cluster2_x1,cluster2_x2]).T |
| 37 | +c3=numpy.array([cluster3_x1,cluster3_x2]).T |
| 38 | + |
| 39 | +data=numpy.concatenate((c1,c2,c3),axis=0) |
| 40 | + |
| 41 | +matplotlib.pyplot.scatter(cluster1_x1,cluster1_x2) |
| 42 | +matplotlib.pyplot.scatter(cluster2_x1,cluster2_x2) |
| 43 | +matplotlib.pyplot.scatter(cluster3_x1,cluster3_x2) |
| 44 | +matplotlib.pyplot.title("Optimal Clustering") |
| 45 | +matplotlib.pyplot.show() |
| 46 | + |
| 47 | +defeuclidean_distance(X,Y): |
| 48 | +""" |
| 49 | + Calculate the euclidean distance between X and Y. It accepts: |
| 50 | + :X should be a matrix of size (N, f) where N is the number of samples and f is the number of features for each sample. |
| 51 | + :Y should be of size f. In other words, it is a single sample. |
| 52 | +
|
| 53 | + Returns a vector of N elements with the distances between the N samples and the Y. |
| 54 | + """ |
| 55 | + |
| 56 | +returnnumpy.sqrt(numpy.sum(numpy.power(X-Y,2),axis=1)) |
| 57 | + |
| 58 | +defcluster_data(solution,solution_idx): |
| 59 | +""" |
| 60 | + Clusters the data based on the current solution. |
| 61 | + """ |
| 62 | + |
| 63 | +globalnum_clusters,feature_vector_length,data |
| 64 | +cluster_centers= []# A list of size (C, f) where C is the number of clusters and f is the number of features representing each sample. |
| 65 | +all_clusters_dists= []# A list of size (C, N) where C is the number of clusters and N is the number of data samples. It holds the distances between each cluster center and all the data samples. |
| 66 | +clusters= []# A list with C elements where each element holds the indices of the samples within a cluster. |
| 67 | +clusters_sum_dist= []# A list with C elements where each element represents the sum of distances of the samples with a cluster. |
| 68 | + |
| 69 | +forclust_idxinrange(num_clusters): |
| 70 | +# Return the current cluster center. |
| 71 | +cluster_centers.append(solution[feature_vector_length*clust_idx:feature_vector_length*(clust_idx+1)]) |
| 72 | +# Calculate the distance (e.g. euclidean) between the current cluster center and all samples. |
| 73 | +cluster_center_dists=euclidean_distance(data,cluster_centers[clust_idx]) |
| 74 | +all_clusters_dists.append(numpy.array(cluster_center_dists)) |
| 75 | + |
| 76 | +cluster_centers=numpy.array(cluster_centers) |
| 77 | +all_clusters_dists=numpy.array(all_clusters_dists) |
| 78 | + |
| 79 | +# A 1D array that, for each sample, holds the index of the cluster with the smallest distance. |
| 80 | +# In other words, the array holds the sample's cluster index. |
| 81 | +cluster_indices=numpy.argmin(all_clusters_dists,axis=0) |
| 82 | +forclust_idxinrange(num_clusters): |
| 83 | +clusters.append(numpy.where(cluster_indices==clust_idx)[0]) |
| 84 | +# Calculate the sum of distances for the cluster. |
| 85 | +iflen(clusters[clust_idx])==0: |
| 86 | +# In case the cluster is empty (i.e. has zero samples). |
| 87 | +clusters_sum_dist.append(0) |
| 88 | +else: |
| 89 | +# When the cluster is not empty (i.e. has at least 1 sample). |
| 90 | +clusters_sum_dist.append(numpy.sum(all_clusters_dists[clust_idx,clusters[clust_idx]])) |
| 91 | +# clusters_sum_dist.append(numpy.sum(euclidean_distance(data[clusters[clust_idx], :], cluster_centers[clust_idx]))) |
| 92 | + |
| 93 | +clusters_sum_dist=numpy.array(clusters_sum_dist) |
| 94 | + |
| 95 | +returncluster_centers,all_clusters_dists,cluster_indices,clusters,clusters_sum_dist |
| 96 | + |
| 97 | +deffitness_func(solution,solution_idx): |
| 98 | +_,_,_,_,clusters_sum_dist=cluster_data(solution,solution_idx) |
| 99 | + |
| 100 | +# The tiny value 0.00000001 is added to the denominator in case the average distance is 0. |
| 101 | +fitness=1.0/ (numpy.sum(clusters_sum_dist)+0.00000001) |
| 102 | + |
| 103 | +returnfitness |
| 104 | + |
| 105 | +num_clusters=3 |
| 106 | +feature_vector_length=data.shape[1] |
| 107 | +num_genes=num_clusters*feature_vector_length |
| 108 | + |
| 109 | +ga_instance=pygad.GA(num_generations=100, |
| 110 | +sol_per_pop=10, |
| 111 | +init_range_low=0, |
| 112 | +init_range_high=20, |
| 113 | +num_parents_mating=5, |
| 114 | +keep_parents=2, |
| 115 | +num_genes=num_genes, |
| 116 | +fitness_func=fitness_func, |
| 117 | +suppress_warnings=True) |
| 118 | + |
| 119 | +ga_instance.run() |
| 120 | + |
| 121 | +best_solution,best_solution_fitness,best_solution_idx=ga_instance.best_solution() |
| 122 | +print("Best solution is {bs}".format(bs=best_solution)) |
| 123 | +print("Fitness of the best solution is {bsf}".format(bsf=best_solution_fitness)) |
| 124 | +print("Best solution found after {gen} generations".format(gen=ga_instance.best_solution_generation)) |
| 125 | + |
| 126 | +cluster_centers,all_clusters_dists,cluster_indices,clusters,clusters_sum_dist=cluster_data(best_solution,best_solution_idx) |
| 127 | + |
| 128 | +forcluster_idxinrange(num_clusters): |
| 129 | +cluster_x=data[clusters[cluster_idx],0] |
| 130 | +cluster_y=data[clusters[cluster_idx],1] |
| 131 | +matplotlib.pyplot.scatter(cluster_x,cluster_y) |
| 132 | +matplotlib.pyplot.scatter(cluster_centers[cluster_idx,0],cluster_centers[cluster_idx,1],linewidths=5) |
| 133 | +matplotlib.pyplot.title("Clustering using PyGAD") |
| 134 | +matplotlib.pyplot.show() |