annotate Marea/marea_cluster.py @ 33:abf0bfe01c78 draft

Uploaded
author bimib
date Wed, 16 Oct 2019 16:25:56 -0400
parents 944e15aa970a
children 1a97d1537623
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
1 # -*- coding: utf-8 -*-
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
2 """
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
3 Created on Mon Jun 3 19:51:00 2019
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
4 @author: Narger
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
5 """
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
6
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
7 import sys
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
8 import argparse
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
9 import os
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
10 from sklearn.datasets import make_blobs
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
11 from sklearn.cluster import KMeans, DBSCAN, AgglomerativeClustering
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
12 from sklearn.metrics import silhouette_samples, silhouette_score, davies_bouldin_score, cluster
25
9992eba50cfb fixed pyplot backend
bimib
parents: 23
diff changeset
13 import matplotlib
9992eba50cfb fixed pyplot backend
bimib
parents: 23
diff changeset
14 matplotlib.use('agg')
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
15 import matplotlib.pyplot as plt
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
16 import scipy.cluster.hierarchy as shc
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
17 import matplotlib.cm as cm
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
18 import numpy as np
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
19 import pandas as pd
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
20
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
21 ################################# process args ###############################
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
22
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
23 def process_args(args):
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
24 parser = argparse.ArgumentParser(usage = '%(prog)s [options]',
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
25 description = 'process some value\'s' +
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
26 ' genes to create class.')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
27
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
28 parser.add_argument('-ol', '--out_log',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
29 help = "Output log")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
30
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
31 parser.add_argument('-in', '--input',
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
32 type = str,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
33 help = 'input dataset')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
34
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
35 parser.add_argument('-cy', '--cluster_type',
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
36 type = str,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
37 choices = ['kmeans', 'meanshift', 'dbscan', 'hierarchy'],
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
38 default = 'kmeans',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
39 help = 'choose clustering algorythm')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
40
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
41 parser.add_argument('-k1', '--k_min',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
42 type = int,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
43 default = 2,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
44 help = 'choose minimun cluster number to be generated')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
45
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
46 parser.add_argument('-k2', '--k_max',
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
47 type = int,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
48 default = 7,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
49 help = 'choose maximum cluster number to be generated')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
50
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
51 parser.add_argument('-el', '--elbow',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
52 type = str,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
53 default = 'false',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
54 choices = ['true', 'false'],
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
55 help = 'choose if you want to generate an elbow plot for kmeans')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
56
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
57 parser.add_argument('-si', '--silhouette',
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
58 type = str,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
59 default = 'false',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
60 choices = ['true', 'false'],
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
61 help = 'choose if you want silhouette plots')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
62
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
63 parser.add_argument('-db', '--davies',
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
64 type = str,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
65 default = 'false',
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
66 choices = ['true', 'false'],
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
67 help = 'choose if you want davies bouldin scores')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
68
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
69 parser.add_argument('-td', '--tool_dir',
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
70 type = str,
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
71 required = True,
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
72 help = 'your tool directory')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
73
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
74 parser.add_argument('-ms', '--min_samples',
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
75 type = float,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
76 help = 'min samples for dbscan (optional)')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
77
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
78 parser.add_argument('-ep', '--eps',
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
79 type = float,
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
80 help = 'eps for dbscan (optional)')
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
81
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
82 parser.add_argument('-bc', '--best_cluster',
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
83 type = str,
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
84 help = 'output of best cluster tsv')
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
85
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
86
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
87
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
88 args = parser.parse_args()
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
89 return args
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
90
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
91 ########################### warning ###########################################
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
92
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
93 def warning(s):
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
94 args = process_args(sys.argv)
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
95 with open(args.out_log, 'a') as log:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
96 log.write(s + "\n\n")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
97 print(s)
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
98
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
99 ########################## read dataset ######################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
100
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
101 def read_dataset(dataset):
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
102 try:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
103 dataset = pd.read_csv(dataset, sep = '\t', header = 0)
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
104 except pd.errors.EmptyDataError:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
105 sys.exit('Execution aborted: wrong format of dataset\n')
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
106 if len(dataset.columns) < 2:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
107 sys.exit('Execution aborted: wrong format of dataset\n')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
108 return dataset
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
109
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
110 ############################ rewrite_input ###################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
111
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
112 def rewrite_input(dataset):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
113 #Riscrivo il dataset come dizionario di liste,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
114 #non come dizionario di dizionari
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
115
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
116 for key, val in dataset.items():
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
117 l = []
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
118 for i in val:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
119 if i == 'None':
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
120 l.append(None)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
121 else:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
122 l.append(float(i))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
123
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
124 dataset[key] = l
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
125
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
126 return dataset
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
127
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
128 ############################## write to csv ##################################
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
129
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
130 def write_to_csv (dataset, labels, name):
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
131 #labels = predict
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
132 predict = [x+1 for x in labels]
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
133
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
134 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
135
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
136 dest = name
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
137 classe.to_csv(dest, sep = '\t', index = False,
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
138 header = ['Patient_ID', 'Class'])
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
139
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
140 ########################### trova il massimo in lista ########################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
141 def max_index (lista):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
142 best = -1
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
143 best_index = 0
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
144 for i in range(len(lista)):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
145 if lista[i] > best:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
146 best = lista [i]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
147 best_index = i
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
148
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
149 return best_index
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
150
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
151 ################################ kmeans #####################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
152
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
153 def kmeans (k_min, k_max, dataset, elbow, silhouette, davies, best_cluster):
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
154 if not os.path.exists('clustering'):
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
155 os.makedirs('clustering')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
156
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
157
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
158 if elbow == 'true':
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
159 elbow = True
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
160 else:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
161 elbow = False
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
162
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
163 if silhouette == 'true':
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
164 silhouette = True
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
165 else:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
166 silhouette = False
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
167
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
168 if davies == 'true':
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
169 davies = True
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
170 else:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
171 davies = False
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
172
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
173
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
174 range_n_clusters = [i for i in range(k_min, k_max+1)]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
175 distortions = []
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
176 scores = []
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
177 all_labels = []
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
178
31
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
179 clusterer = KMeans(n_clusters=1, random_state=10)
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
180 distortions.append(clusterer.fit(dataset).inertia_)
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
181
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
182
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
183 for n_clusters in range_n_clusters:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
184 clusterer = KMeans(n_clusters=n_clusters, random_state=10)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
185 cluster_labels = clusterer.fit_predict(dataset)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
186
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
187 all_labels.append(cluster_labels)
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
188 if n_clusters == 1:
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
189 silhouette_avg = 0
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
190 else:
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
191 silhouette_avg = silhouette_score(dataset, cluster_labels)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
192 scores.append(silhouette_avg)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
193 distortions.append(clusterer.fit(dataset).inertia_)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
194
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
195 best = max_index(scores) + k_min
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
196
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
197 for i in range(len(all_labels)):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
198 prefix = ''
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
199 if (i + k_min == best):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
200 prefix = '_BEST'
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
201
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
202 write_to_csv(dataset, all_labels[i], 'clustering/kmeans_with_' + str(i + k_min) + prefix + '_clusters.tsv')
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
203
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
204
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
205 if (prefix == '_BEST'):
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
206 labels = all_labels[i]
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
207 predict = [x+1 for x in labels]
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
208 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
209 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
210
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
211
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
212 if davies:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
213 with np.errstate(divide='ignore', invalid='ignore'):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
214 davies_bouldin = davies_bouldin_score(dataset, all_labels[i])
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
215 warning("\nFor n_clusters = " + str(i + k_min) +
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
216 " The average davies bouldin score is: " + str(davies_bouldin))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
217
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
218
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
219 if silhouette:
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
220 silihouette_draw(dataset, all_labels[i], i + k_min, 'clustering/silhouette_with_' + str(i + k_min) + prefix + '_clusters.png')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
221
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
222
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
223 if elbow:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
224 elbow_plot(distortions, k_min,k_max)
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
225
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
226
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
227
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
228
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
229
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
230 ############################## elbow_plot ####################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
231
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
232 def elbow_plot (distortions, k_min, k_max):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
233 plt.figure(0)
31
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
234 x = list(range(k_min, k_max + 1))
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
235 x.insert(0, 1)
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
236 plt.plot(x, distortions, marker = 'o')
944e15aa970a Uploaded
bimib
parents: 28
diff changeset
237 plt.xlabel('Number of clusters (k)')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
238 plt.ylabel('Distortion')
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
239 s = 'clustering/elbow_plot.png'
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
240 fig = plt.gcf()
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
241 fig.set_size_inches(18.5, 10.5, forward = True)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
242 fig.savefig(s, dpi=100)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
243
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
244
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
245 ############################## silhouette plot ###############################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
246 def silihouette_draw(dataset, labels, n_clusters, path):
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
247 if n_clusters == 1:
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
248 return None
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
249
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
250 silhouette_avg = silhouette_score(dataset, labels)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
251 warning("For n_clusters = " + str(n_clusters) +
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
252 " The average silhouette_score is: " + str(silhouette_avg))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
253
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
254 plt.close('all')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
255 # Create a subplot with 1 row and 2 columns
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
256 fig, (ax1) = plt.subplots(1, 1)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
257
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
258 fig.set_size_inches(18, 7)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
259
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
260 # The 1st subplot is the silhouette plot
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
261 # The silhouette coefficient can range from -1, 1 but in this example all
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
262 # lie within [-0.1, 1]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
263 ax1.set_xlim([-1, 1])
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
264 # The (n_clusters+1)*10 is for inserting blank space between silhouette
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
265 # plots of individual clusters, to demarcate them clearly.
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
266 ax1.set_ylim([0, len(dataset) + (n_clusters + 1) * 10])
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
267
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
268 # Compute the silhouette scores for each sample
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
269 sample_silhouette_values = silhouette_samples(dataset, labels)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
270
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
271 y_lower = 10
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
272 for i in range(n_clusters):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
273 # Aggregate the silhouette scores for samples belonging to
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
274 # cluster i, and sort them
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
275 ith_cluster_silhouette_values = \
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
276 sample_silhouette_values[labels == i]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
277
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
278 ith_cluster_silhouette_values.sort()
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
279
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
280 size_cluster_i = ith_cluster_silhouette_values.shape[0]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
281 y_upper = y_lower + size_cluster_i
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
282
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
283 color = cm.nipy_spectral(float(i) / n_clusters)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
284 ax1.fill_betweenx(np.arange(y_lower, y_upper),
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
285 0, ith_cluster_silhouette_values,
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
286 facecolor=color, edgecolor=color, alpha=0.7)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
287
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
288 # Label the silhouette plots with their cluster numbers at the middle
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
289 ax1.text(-0.05, y_lower + 0.5 * size_cluster_i, str(i))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
290
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
291 # Compute the new y_lower for next plot
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
292 y_lower = y_upper + 10 # 10 for the 0 samples
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
293
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
294 ax1.set_title("The silhouette plot for the various clusters.")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
295 ax1.set_xlabel("The silhouette coefficient values")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
296 ax1.set_ylabel("Cluster label")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
297
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
298 # The vertical line for average silhouette score of all the values
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
299 ax1.axvline(x=silhouette_avg, color="red", linestyle="--")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
300
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
301 ax1.set_yticks([]) # Clear the yaxis labels / ticks
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
302 ax1.set_xticks([-0.1, 0, 0.2, 0.4, 0.6, 0.8, 1])
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
303
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
304
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
305 plt.suptitle(("Silhouette analysis for clustering on sample data "
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
306 "with n_clusters = " + str(n_clusters) + "\nAverage silhouette_score = " + str(silhouette_avg)), fontsize=12, fontweight='bold')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
307
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
308
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
309 plt.savefig(path, bbox_inches='tight')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
310
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
311 ######################## dbscan ##############################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
312
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
313 def dbscan(dataset, eps, min_samples, best_cluster):
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
314 if not os.path.exists('clustering'):
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
315 os.makedirs('clustering')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
316
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
317 if eps is not None:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
318 clusterer = DBSCAN(eps = eps, min_samples = min_samples)
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
319 else:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
320 clusterer = DBSCAN()
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
321
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
322 clustering = clusterer.fit(dataset)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
323
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
324 core_samples_mask = np.zeros_like(clustering.labels_, dtype=bool)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
325 core_samples_mask[clustering.core_sample_indices_] = True
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
326 labels = clustering.labels_
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
327
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
328 # Number of clusters in labels, ignoring noise if present.
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
329 n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
330
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
331
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
332 ##TODO: PLOT SU DBSCAN (no centers) e HIERARCHICAL
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
333
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
334 labels = labels
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
335 predict = [x+1 for x in labels]
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
336 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
337 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
338
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
339
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
340 ########################## hierachical #######################################
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
341
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
342 def hierachical_agglomerative(dataset, k_min, k_max, best_cluster):
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
343
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
344 if not os.path.exists('clustering'):
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
345 os.makedirs('clustering')
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
346
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
347 plt.figure(figsize=(10, 7))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
348 plt.title("Customer Dendograms")
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
349 shc.dendrogram(shc.linkage(dataset, method='ward'))
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
350 fig = plt.gcf()
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
351 fig.savefig('clustering/dendogram.png', dpi=200)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
352
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
353 range_n_clusters = [i for i in range(k_min, k_max+1)]
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
354
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
355 scores = []
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
356 labels = []
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
357 for n_clusters in range_n_clusters:
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
358 cluster = AgglomerativeClustering(n_clusters=n_clusters, affinity='euclidean', linkage='ward')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
359 cluster.fit_predict(dataset)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
360 cluster_labels = cluster.labels_
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
361 labels.append(cluster_labels)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
362 silhouette_avg = silhouette_score(dataset, cluster_labels)
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
363 write_to_csv(dataset, cluster_labels, 'clustering/hierarchical_with_' + str(n_clusters) + '_clusters.tsv')
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
364 scores.append(silhouette_avg)
23
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
365 #warning("For n_clusters =", n_clusters,
a8825e66c3a0 Uploaded
bimib
parents: 16
diff changeset
366 #"The average silhouette_score is :", silhouette_avg)
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
367
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
368 best = max_index(scores) + k_min
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
369
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
370 for i in range(len(labels)):
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
371 if (i + k_min == best):
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
372 labels = labels[i]
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
373 predict = [x+1 for x in labels]
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
374 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
375 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
376
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
377
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
378
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
379
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
380
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
381
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
382 ############################# main ###########################################
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
383
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
384
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
385 def main():
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
386 if not os.path.exists('clustering'):
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
387 os.makedirs('clustering')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
388
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
389 args = process_args(sys.argv)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
390
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
391 #Data read
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
392
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
393 X = read_dataset(args.input)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
394 X = pd.DataFrame.to_dict(X, orient='list')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
395 X = rewrite_input(X)
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
396 X = pd.DataFrame.from_dict(X, orient = 'index')
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
397
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
398 for i in X.columns:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
399 tmp = X[i][0]
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
400 if tmp == None:
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
401 X = X.drop(columns=[i])
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
402
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
403
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
404 if args.cluster_type == 'kmeans':
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
405 kmeans(args.k_min, args.k_max, X, args.elbow, args.silhouette, args.davies, args.best_cluster)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
406
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
407 if args.cluster_type == 'dbscan':
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
408 dbscan(X, args.eps, args.min_samples, args.best_cluster)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
409
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
410 if args.cluster_type == 'hierarchy':
33
abf0bfe01c78 Uploaded
bimib
parents: 31
diff changeset
411 hierachical_agglomerative(X, args.k_min, args.k_max, args.best_cluster)
16
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
412
c71ac0bb12de Uploaded
bimib
parents: 14
diff changeset
413 ##############################################################################
0
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
414
23ac9cf12788 Uploaded
bimib
parents:
diff changeset
415 if __name__ == "__main__":
28
e6831924df01 small fixes (elbow plot and output managment)
bimib
parents: 25
diff changeset
416 main()