4
|
1 # -*- coding: utf-8 -*-
|
|
2 """
|
|
3 Created on Mon Jun 3 19:51:00 2019
|
|
4 @author: Narger
|
|
5 """
|
|
6
|
|
7 import sys
|
|
8 import argparse
|
|
9 import os
|
|
10 import numpy as np
|
|
11 import pandas as pd
|
|
12 from sklearn.datasets import make_blobs
|
|
13 from sklearn.cluster import KMeans, DBSCAN, AgglomerativeClustering
|
|
14 from sklearn.metrics import silhouette_samples, silhouette_score, cluster
|
|
15 import matplotlib
|
|
16 matplotlib.use('agg')
|
|
17 import matplotlib.pyplot as plt
|
|
18 import scipy.cluster.hierarchy as shc
|
|
19 import matplotlib.cm as cm
|
|
20 from typing import Optional, Dict, List
|
|
21
|
|
22 ################################# process args ###############################
|
|
23 def process_args(args :List[str]) -> argparse.Namespace:
|
|
24 """
|
|
25 Processes command-line arguments.
|
|
26
|
|
27 Args:
|
|
28 args (list): List of command-line arguments.
|
|
29
|
|
30 Returns:
|
|
31 Namespace: An object containing parsed arguments.
|
|
32 """
|
|
33 parser = argparse.ArgumentParser(usage = '%(prog)s [options]',
|
|
34 description = 'process some value\'s' +
|
|
35 ' genes to create class.')
|
|
36
|
|
37 parser.add_argument('-ol', '--out_log',
|
|
38 help = "Output log")
|
|
39
|
|
40 parser.add_argument('-in', '--input',
|
|
41 type = str,
|
|
42 help = 'input dataset')
|
|
43
|
|
44 parser.add_argument('-cy', '--cluster_type',
|
|
45 type = str,
|
|
46 choices = ['kmeans', 'dbscan', 'hierarchy'],
|
|
47 default = 'kmeans',
|
|
48 help = 'choose clustering algorythm')
|
|
49
|
|
50 parser.add_argument('-k1', '--k_min',
|
|
51 type = int,
|
|
52 default = 2,
|
|
53 help = 'choose minimun cluster number to be generated')
|
|
54
|
|
55 parser.add_argument('-k2', '--k_max',
|
|
56 type = int,
|
|
57 default = 7,
|
|
58 help = 'choose maximum cluster number to be generated')
|
|
59
|
|
60 parser.add_argument('-el', '--elbow',
|
|
61 type = str,
|
|
62 default = 'false',
|
|
63 choices = ['true', 'false'],
|
|
64 help = 'choose if you want to generate an elbow plot for kmeans')
|
|
65
|
|
66 parser.add_argument('-si', '--silhouette',
|
|
67 type = str,
|
|
68 default = 'false',
|
|
69 choices = ['true', 'false'],
|
|
70 help = 'choose if you want silhouette plots')
|
|
71
|
|
72 parser.add_argument('-td', '--tool_dir',
|
|
73 type = str,
|
|
74 required = True,
|
|
75 help = 'your tool directory')
|
|
76
|
|
77 parser.add_argument('-ms', '--min_samples',
|
|
78 type = float,
|
|
79 help = 'min samples for dbscan (optional)')
|
|
80
|
|
81 parser.add_argument('-ep', '--eps',
|
|
82 type = float,
|
|
83 help = 'eps for dbscan (optional)')
|
|
84
|
|
85 parser.add_argument('-bc', '--best_cluster',
|
|
86 type = str,
|
|
87 help = 'output of best cluster tsv')
|
|
88
|
|
89
|
|
90
|
|
91 args = parser.parse_args()
|
|
92 return args
|
|
93
|
|
94 ########################### warning ###########################################
|
|
95 def warning(s :str) -> None:
|
|
96 """
|
|
97 Log a warning message to an output log file and print it to the console.
|
|
98
|
|
99 Args:
|
|
100 s (str): The warning message to be logged and printed.
|
|
101
|
|
102 Returns:
|
|
103 None
|
|
104 """
|
|
105 args = process_args(sys.argv)
|
|
106 with open(args.out_log, 'a') as log:
|
|
107 log.write(s + "\n\n")
|
|
108 print(s)
|
|
109
|
|
110 ########################## read dataset ######################################
|
|
111 def read_dataset(dataset :str) -> pd.DataFrame:
|
|
112 """
|
|
113 Read dataset from a CSV file and return it as a Pandas DataFrame.
|
|
114
|
|
115 Args:
|
|
116 dataset (str): the path to the dataset to convert into a DataFrame
|
|
117
|
|
118 Returns:
|
|
119 pandas.DataFrame: The dataset loaded as a Pandas DataFrame.
|
|
120
|
|
121 Raises:
|
|
122 pandas.errors.EmptyDataError: If the dataset file is empty.
|
|
123 sys.exit: If the dataset file has the wrong format (e.g., fewer than 2 columns)
|
|
124 """
|
|
125 try:
|
|
126 dataset = pd.read_csv(dataset, sep = '\t', header = 0)
|
|
127 except pd.errors.EmptyDataError:
|
|
128 sys.exit('Execution aborted: wrong format of dataset\n')
|
|
129 if len(dataset.columns) < 2:
|
|
130 sys.exit('Execution aborted: wrong format of dataset\n')
|
|
131 return dataset
|
|
132
|
|
133 ############################ rewrite_input ###################################
|
|
134 def rewrite_input(dataset :pd.DataFrame) -> Dict[str, List[Optional[float]]]:
|
|
135 """
|
|
136 Rewrite the dataset as a dictionary of lists instead of as a dictionary of dictionaries.
|
|
137
|
|
138 Args:
|
|
139 dataset (pandas.DataFrame): The dataset to be rewritten.
|
|
140
|
|
141 Returns:
|
|
142 dict: The rewritten dataset as a dictionary of lists.
|
|
143 """
|
|
144 #Riscrivo il dataset come dizionario di liste,
|
|
145 #non come dizionario di dizionari
|
|
146
|
|
147 dataset.pop('Reactions', None)
|
|
148
|
|
149 for key, val in dataset.items():
|
|
150 l = []
|
|
151 for i in val:
|
|
152 if i == 'None':
|
|
153 l.append(None)
|
|
154 else:
|
|
155 l.append(float(i))
|
|
156
|
|
157 dataset[key] = l
|
|
158
|
|
159 return dataset
|
|
160
|
|
161 ############################## write to csv ##################################
|
|
162 def write_to_csv (dataset :pd.DataFrame, labels :List[str], name :str) -> None:
|
|
163 """
|
|
164 Write dataset and predicted labels to a CSV file.
|
|
165
|
|
166 Args:
|
|
167 dataset (pandas.DataFrame): The dataset to be written.
|
|
168 labels (list): The predicted labels for each data point.
|
|
169 name (str): The name of the output CSV file.
|
|
170
|
|
171 Returns:
|
|
172 None
|
|
173 """
|
|
174 #labels = predict
|
|
175 predict = [x+1 for x in labels]
|
|
176
|
|
177 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
|
|
178
|
|
179 dest = name
|
|
180 classe.to_csv(dest, sep = '\t', index = False,
|
|
181 header = ['Patient_ID', 'Class'])
|
|
182
|
|
183 ########################### trova il massimo in lista ########################
|
|
184 def max_index (lista :List[int]) -> int:
|
|
185 """
|
|
186 Find the index of the maximum value in a list.
|
|
187
|
|
188 Args:
|
|
189 lista (list): The list in which we search for the index of the maximum value.
|
|
190
|
|
191 Returns:
|
|
192 int: The index of the maximum value in the list.
|
|
193 """
|
|
194 best = -1
|
|
195 best_index = 0
|
|
196 for i in range(len(lista)):
|
|
197 if lista[i] > best:
|
|
198 best = lista [i]
|
|
199 best_index = i
|
|
200
|
|
201 return best_index
|
|
202
|
|
203 ################################ kmeans #####################################
|
|
204 def kmeans (k_min: int, k_max: int, dataset: pd.DataFrame, elbow: str, silhouette: str, best_cluster: str) -> None:
|
|
205 """
|
|
206 Perform k-means clustering on the given dataset, which is an algorithm used to partition a dataset into groups (clusters) based on their characteristics.
|
|
207 The goal is to divide the data into homogeneous groups, where the elements within each group are similar to each other and different from the elements in other groups.
|
|
208
|
|
209 Args:
|
|
210 k_min (int): The minimum number of clusters to consider.
|
|
211 k_max (int): The maximum number of clusters to consider.
|
|
212 dataset (pandas.DataFrame): The dataset to perform clustering on.
|
|
213 elbow (str): Whether to generate an elbow plot for kmeans ('true' or 'false').
|
|
214 silhouette (str): Whether to generate silhouette plots ('true' or 'false').
|
|
215 best_cluster (str): The file path to save the output of the best cluster.
|
|
216
|
|
217 Returns:
|
|
218 None
|
|
219 """
|
|
220 if not os.path.exists('clustering'):
|
|
221 os.makedirs('clustering')
|
|
222
|
|
223
|
|
224 if elbow == 'true':
|
|
225 elbow = True
|
|
226 else:
|
|
227 elbow = False
|
|
228
|
|
229 if silhouette == 'true':
|
|
230 silhouette = True
|
|
231 else:
|
|
232 silhouette = False
|
|
233
|
|
234 range_n_clusters = [i for i in range(k_min, k_max+1)]
|
|
235 distortions = []
|
|
236 scores = []
|
|
237 all_labels = []
|
|
238
|
|
239 clusterer = KMeans(n_clusters=1, random_state=10)
|
|
240 distortions.append(clusterer.fit(dataset).inertia_)
|
|
241
|
|
242
|
|
243 for n_clusters in range_n_clusters:
|
|
244 clusterer = KMeans(n_clusters=n_clusters, random_state=10)
|
|
245 cluster_labels = clusterer.fit_predict(dataset)
|
|
246
|
|
247 all_labels.append(cluster_labels)
|
|
248 if n_clusters == 1:
|
|
249 silhouette_avg = 0
|
|
250 else:
|
|
251 silhouette_avg = silhouette_score(dataset, cluster_labels)
|
|
252 scores.append(silhouette_avg)
|
|
253 distortions.append(clusterer.fit(dataset).inertia_)
|
|
254
|
|
255 best = max_index(scores) + k_min
|
|
256
|
|
257 for i in range(len(all_labels)):
|
|
258 prefix = ''
|
|
259 if (i + k_min == best):
|
|
260 prefix = '_BEST'
|
|
261
|
|
262 write_to_csv(dataset, all_labels[i], 'clustering/kmeans_with_' + str(i + k_min) + prefix + '_clusters.tsv')
|
|
263
|
|
264
|
|
265 if (prefix == '_BEST'):
|
|
266 labels = all_labels[i]
|
|
267 predict = [x+1 for x in labels]
|
|
268 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
|
|
269 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
|
|
270
|
|
271
|
|
272
|
|
273
|
|
274 if silhouette:
|
|
275 silhouette_draw(dataset, all_labels[i], i + k_min, 'clustering/silhouette_with_' + str(i + k_min) + prefix + '_clusters.png')
|
|
276
|
|
277
|
|
278 if elbow:
|
|
279 elbow_plot(distortions, k_min,k_max)
|
|
280
|
|
281
|
|
282
|
|
283
|
|
284
|
|
285 ############################## elbow_plot ####################################
|
|
286 def elbow_plot (distortions: List[float], k_min: int, k_max: int) -> None:
|
|
287 """
|
|
288 Generate an elbow plot to visualize the distortion for different numbers of clusters.
|
|
289 The elbow plot is a graphical tool used in clustering analysis to help identifying the appropriate number of clusters by looking for the point where the rate of decrease
|
|
290 in distortion sharply decreases, indicating the optimal balance between model complexity and clustering quality.
|
|
291
|
|
292 Args:
|
|
293 distortions (list): List of distortion values for different numbers of clusters.
|
|
294 k_min (int): The minimum number of clusters considered.
|
|
295 k_max (int): The maximum number of clusters considered.
|
|
296
|
|
297 Returns:
|
|
298 None
|
|
299 """
|
|
300 plt.figure(0)
|
|
301 x = list(range(k_min, k_max + 1))
|
|
302 x.insert(0, 1)
|
|
303 plt.plot(x, distortions, marker = 'o')
|
|
304 plt.xlabel('Number of clusters (k)')
|
|
305 plt.ylabel('Distortion')
|
|
306 s = 'clustering/elbow_plot.png'
|
|
307 fig = plt.gcf()
|
|
308 fig.set_size_inches(18.5, 10.5, forward = True)
|
|
309 fig.savefig(s, dpi=100)
|
|
310
|
|
311
|
|
312 ############################## silhouette plot ###############################
|
|
313 def silhouette_draw(dataset: pd.DataFrame, labels: List[str], n_clusters: int, path:str) -> None:
|
|
314 """
|
|
315 Generate a silhouette plot for the clustering results.
|
|
316 The silhouette coefficient is a measure used to evaluate the quality of clusters obtained from a clustering algorithmand it quantifies how similar an object is to its own cluster compared to other clusters.
|
|
317 The silhouette coefficient ranges from -1 to 1, where:
|
|
318 - A value close to +1 indicates that the object is well matched to its own cluster and poorly matched to neighboring clusters. This implies that the object is in a dense, well-separated cluster.
|
|
319 - A value close to 0 indicates that the object is close to the decision boundary between two neighboring clusters.
|
|
320 - A value close to -1 indicates that the object may have been assigned to the wrong cluster.
|
|
321
|
|
322 Args:
|
|
323 dataset (pandas.DataFrame): The dataset used for clustering.
|
|
324 labels (list): The cluster labels assigned to each data point.
|
|
325 n_clusters (int): The number of clusters.
|
|
326 path (str): The path to save the silhouette plot image.
|
|
327
|
|
328 Returns:
|
|
329 None
|
|
330 """
|
|
331 if n_clusters == 1:
|
|
332 return None
|
|
333
|
|
334 silhouette_avg = silhouette_score(dataset, labels)
|
|
335 warning("For n_clusters = " + str(n_clusters) +
|
|
336 " The average silhouette_score is: " + str(silhouette_avg))
|
|
337
|
|
338 plt.close('all')
|
|
339 # Create a subplot with 1 row and 2 columns
|
|
340 fig, (ax1) = plt.subplots(1, 1)
|
|
341
|
|
342 fig.set_size_inches(18, 7)
|
|
343
|
|
344 # The 1st subplot is the silhouette plot
|
|
345 # The silhouette coefficient can range from -1, 1 but in this example all
|
|
346 # lie within [-0.1, 1]
|
|
347 ax1.set_xlim([-1, 1])
|
|
348 # The (n_clusters+1)*10 is for inserting blank space between silhouette
|
|
349 # plots of individual clusters, to demarcate them clearly.
|
|
350 ax1.set_ylim([0, len(dataset) + (n_clusters + 1) * 10])
|
|
351
|
|
352 # Compute the silhouette scores for each sample
|
|
353 sample_silhouette_values = silhouette_samples(dataset, labels)
|
|
354
|
|
355 y_lower = 10
|
|
356 for i in range(n_clusters):
|
|
357 # Aggregate the silhouette scores for samples belonging to
|
|
358 # cluster i, and sort them
|
|
359 ith_cluster_silhouette_values = \
|
|
360 sample_silhouette_values[labels == i]
|
|
361
|
|
362 ith_cluster_silhouette_values.sort()
|
|
363
|
|
364 size_cluster_i = ith_cluster_silhouette_values.shape[0]
|
|
365 y_upper = y_lower + size_cluster_i
|
|
366
|
|
367 color = cm.nipy_spectral(float(i) / n_clusters)
|
|
368 ax1.fill_betweenx(np.arange(y_lower, y_upper),
|
|
369 0, ith_cluster_silhouette_values,
|
|
370 facecolor=color, edgecolor=color, alpha=0.7)
|
|
371
|
|
372 # Label the silhouette plots with their cluster numbers at the middle
|
|
373 ax1.text(-0.05, y_lower + 0.5 * size_cluster_i, str(i))
|
|
374
|
|
375 # Compute the new y_lower for next plot
|
|
376 y_lower = y_upper + 10 # 10 for the 0 samples
|
|
377
|
|
378 ax1.set_title("The silhouette plot for the various clusters.")
|
|
379 ax1.set_xlabel("The silhouette coefficient values")
|
|
380 ax1.set_ylabel("Cluster label")
|
|
381
|
|
382 # The vertical line for average silhouette score of all the values
|
|
383 ax1.axvline(x=silhouette_avg, color="red", linestyle="--")
|
|
384
|
|
385 ax1.set_yticks([]) # Clear the yaxis labels / ticks
|
|
386 ax1.set_xticks([-0.1, 0, 0.2, 0.4, 0.6, 0.8, 1])
|
|
387
|
|
388
|
|
389 plt.suptitle(("Silhouette analysis for clustering on sample data "
|
|
390 "with n_clusters = " + str(n_clusters) + "\nAverage silhouette_score = " + str(silhouette_avg)), fontsize=12, fontweight='bold')
|
|
391
|
|
392
|
|
393 plt.savefig(path, bbox_inches='tight')
|
|
394
|
|
395 ######################## dbscan ##############################################
|
|
396 def dbscan(dataset: pd.DataFrame, eps: float, min_samples: float, best_cluster: str) -> None:
|
|
397 """
|
|
398 Perform DBSCAN clustering on the given dataset, which is a clustering algorithm that groups together closely packed points based on the notion of density.
|
|
399
|
|
400 Args:
|
|
401 dataset (pandas.DataFrame): The dataset to be clustered.
|
|
402 eps (float): The maximum distance between two samples for one to be considered as in the neighborhood of the other.
|
|
403 min_samples (float): The number of samples in a neighborhood for a point to be considered as a core point.
|
|
404 best_cluster (str): The file path to save the output of the best cluster.
|
|
405
|
|
406 Returns:
|
|
407 None
|
|
408 """
|
|
409 if not os.path.exists('clustering'):
|
|
410 os.makedirs('clustering')
|
|
411
|
|
412 if eps is not None:
|
|
413 clusterer = DBSCAN(eps = eps, min_samples = min_samples)
|
|
414 else:
|
|
415 clusterer = DBSCAN()
|
|
416
|
|
417 clustering = clusterer.fit(dataset)
|
|
418
|
|
419 core_samples_mask = np.zeros_like(clustering.labels_, dtype=bool)
|
|
420 core_samples_mask[clustering.core_sample_indices_] = True
|
|
421 labels = clustering.labels_
|
|
422
|
|
423 # Number of clusters in labels, ignoring noise if present.
|
|
424 n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
|
|
425
|
|
426
|
|
427 labels = labels
|
|
428 predict = [x+1 for x in labels]
|
|
429 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
|
|
430 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
|
|
431
|
|
432
|
|
433 ########################## hierachical #######################################
|
|
434 def hierachical_agglomerative(dataset: pd.DataFrame, k_min: int, k_max: int, best_cluster: str, silhouette: str) -> None:
|
|
435 """
|
|
436 Perform hierarchical agglomerative clustering on the given dataset.
|
|
437
|
|
438 Args:
|
|
439 dataset (pandas.DataFrame): The dataset to be clustered.
|
|
440 k_min (int): The minimum number of clusters to consider.
|
|
441 k_max (int): The maximum number of clusters to consider.
|
|
442 best_cluster (str): The file path to save the output of the best cluster.
|
|
443 silhouette (str): Whether to generate silhouette plots ('true' or 'false').
|
|
444
|
|
445 Returns:
|
|
446 None
|
|
447 """
|
|
448 if not os.path.exists('clustering'):
|
|
449 os.makedirs('clustering')
|
|
450
|
|
451 plt.figure(figsize=(10, 7))
|
|
452 plt.title("Customer Dendograms")
|
|
453 shc.dendrogram(shc.linkage(dataset, method='ward'), labels=dataset.index.values.tolist())
|
|
454 fig = plt.gcf()
|
|
455 fig.savefig('clustering/dendogram.png', dpi=200)
|
|
456
|
|
457 range_n_clusters = [i for i in range(k_min, k_max+1)]
|
|
458
|
|
459 scores = []
|
|
460 labels = []
|
|
461
|
|
462 n_classi = dataset.shape[0]
|
|
463
|
|
464 for n_clusters in range_n_clusters:
|
|
465 cluster = AgglomerativeClustering(n_clusters=n_clusters, affinity='euclidean', linkage='ward')
|
|
466 cluster.fit_predict(dataset)
|
|
467 cluster_labels = cluster.labels_
|
|
468 labels.append(cluster_labels)
|
|
469 write_to_csv(dataset, cluster_labels, 'clustering/hierarchical_with_' + str(n_clusters) + '_clusters.tsv')
|
|
470
|
|
471 best = max_index(scores) + k_min
|
|
472
|
|
473 for i in range(len(labels)):
|
|
474 prefix = ''
|
|
475 if (i + k_min == best):
|
|
476 prefix = '_BEST'
|
|
477 if silhouette == 'true':
|
|
478 silhouette_draw(dataset, labels[i], i + k_min, 'clustering/silhouette_with_' + str(i + k_min) + prefix + '_clusters.png')
|
|
479
|
|
480 for i in range(len(labels)):
|
|
481 if (i + k_min == best):
|
|
482 labels = labels[i]
|
|
483 predict = [x+1 for x in labels]
|
|
484 classe = (pd.DataFrame(list(zip(dataset.index, predict)))).astype(str)
|
|
485 classe.to_csv(best_cluster, sep = '\t', index = False, header = ['Patient_ID', 'Class'])
|
|
486
|
|
487
|
|
488 ############################# main ###########################################
|
|
489 def main() -> None:
|
|
490 """
|
|
491 Initializes everything and sets the program in motion based on the fronted input arguments.
|
|
492
|
|
493 Returns:
|
|
494 None
|
|
495 """
|
|
496 if not os.path.exists('clustering'):
|
|
497 os.makedirs('clustering')
|
|
498
|
|
499 args = process_args(sys.argv)
|
|
500
|
|
501 #Data read
|
|
502
|
|
503 X = read_dataset(args.input)
|
|
504 X = pd.DataFrame.to_dict(X, orient='list')
|
|
505 X = rewrite_input(X)
|
|
506 X = pd.DataFrame.from_dict(X, orient = 'index')
|
|
507
|
|
508 for i in X.columns:
|
|
509 tmp = X[i][0]
|
|
510 if tmp == None:
|
|
511 X = X.drop(columns=[i])
|
|
512
|
|
513 ## NAN TO HANLDE
|
|
514
|
|
515 if args.k_max != None:
|
|
516 numero_classi = X.shape[0]
|
|
517 while args.k_max >= numero_classi:
|
|
518 err = 'Skipping k = ' + str(args.k_max) + ' since it is >= number of classes of dataset'
|
|
519 warning(err)
|
|
520 args.k_max = args.k_max - 1
|
|
521
|
|
522
|
|
523 if args.cluster_type == 'kmeans':
|
|
524 kmeans(args.k_min, args.k_max, X, args.elbow, args.silhouette, args.best_cluster)
|
|
525
|
|
526 if args.cluster_type == 'dbscan':
|
|
527 dbscan(X, args.eps, args.min_samples, args.best_cluster)
|
|
528
|
|
529 if args.cluster_type == 'hierarchy':
|
|
530 hierachical_agglomerative(X, args.k_min, args.k_max, args.best_cluster, args.silhouette)
|
|
531
|
|
532 ##############################################################################
|
|
533 if __name__ == "__main__":
|
|
534 main()
|