-
Notifications
You must be signed in to change notification settings - Fork 0
/
3.Clustering.py
332 lines (280 loc) · 10.4 KB
/
3.Clustering.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
print('----------------------------------------------------------------------')
print('|| PARTITION BASED CLUSTERING USING COSINE, JACCARD AND CORRELATION ||')
print('----------------------------------------------------------------------')
# METHODS
# ----------------------------------------------------------------------------
from nltk.tokenize import word_tokenize as T
from sklearn.metrics.pairwise import pairwise_distances
from scipy.spatial.distance import squareform, pdist
from scipy.stats import rankdata, pearsonr
import pandas as pd, numpy as np, itertools, time
def cosine(A, B):
return pdist([A, B], metric='cosine')[0]
def correlation(A, B):
return pdist([A, B], metric='correlation')[0]
def jaccard(A, B):
return 1 - sum(np.minimum(A, B)) / sum(np.maximum(A, B))
def link(A, B):
return np.dot(A, B)
def FindInitCent(data, k):
nplus = k
disbank = pairwise_distances(data, metric=meas)
simbank = 1 - disbank
neghbrs = np.where(simbank >= teta, 1, 0)
neghsum = neghbrs.sum(0)
idxcand = np.argsort(neghsum, kind='mergesort')[-(k + nplus):][::-1]
linkcnd = [];
simicnd = [];
combine = []
for i, j in itertools.combinations(idxcand, 2):
linkcnd.append(link(neghbrs[i], neghbrs[j]))
simicnd.append(simbank[i][j])
combine.append((i, j))
ranklnk = rankdata(linkcnd, method='dense')
ranksim = rankdata(simicnd, method='dense')
ranksum = ranklnk + ranksim
rankcom = []
for i in itertools.combinations(idxcand, k):
temp = 0
for j in itertools.combinations(i, 2):
temp += ranksum[combine.index(j)]
rankcom.append([[temp], list(i)])
return rankcom[np.argmin(np.transpose(rankcom)[0])][1], simbank, neghbrs
def FindCluster(data, cent, ocent):
if sim != 'CORRELATION':
lmax = len(data)
ncent = []
for i in cent:
ncent.append([1 - pdist([i, j], metric=meas)[0] for j in data])
ncent = np.where(np.array(ncent) >= teta, 1, 0)
cluster = []
for enu1, i in enumerate(data):
simi = []
for enu2, j in enumerate(cent):
if sim == 'CORRELATION':
simi.append(1 - meas(i, j))
else:
simi.append(alfa * (link(ncent[enu2], ocent[enu1]) / lmax) + (1 - alfa) * (1 - meas(i, j)))
cluster.append(np.argmax(simi))
return cluster
def FindCentroid(data, clus, k):
newc = []
for i in range(k):
idx = np.isin(clus, i)
newc.append(np.mean(data[idx], 0))
return np.array(newc)
def FindSplit(data, cent, clus):
avg = []
for i, j in enumerate(cent):
temp = []
for k in data[np.isin(clus, i)]:
temp.append(meas(j, k))
avg.append(np.average(temp))
return avg
def getSim(sim):
sim = str.lower(sim)
if sim == 'cosine':
return eval(sim)
elif sim == 'correlation':
return eval(sim)
return eval(sim)
def getMeth(data, k, meth):
if meth == 'K_MEANS':
return eval(meth)(data, k, [], True, [])
elif meth == 'BKMEANS':
return eval(meth)(np.arange(len(data)), np.zeros(len(data)), 0, [])
return eval(meth)(data, k)
def ExpressionText(cent, clus, dicti):
N = 10
idx = [np.argsort(i)[-(N + 10):][::-1] for i in cent]
val = [i[np.argsort(i)[-(N + 10):][::-1]] for i in cent]
for i, j in itertools.combinations(np.arange(len(idx)), 2):
inter = list(set(idx[i]) & set(idx[j]))
if len(inter) == 0:
continue
for k in inter:
idx1 = list(idx[i]).index(k)
idx2 = list(idx[j]).index(k)
if val[i][idx1] < val[j][idx2]:
idx[i] = np.delete(idx[i], idx1)
elif val[i][idx1] > val[j][idx2]:
idx[j] = np.delete(idx[j], idx2)
text = []
for enu, i in enumerate(idx):
print(' CLUSTER', enu + 1, ':', clus.count(enu), 'Doc >> ', end='')
text.append(', '.join(np.array(dicti)[i[0:N]]))
print(text[enu])
return text
def PrintResult(var):
var = np.array(var)
X = [[i] for i in np.average(var, 1)]
Y = np.hstack((np.average(var, 0), 0))
Z = np.hstack((var, X))
Z = np.vstack((Z, Y))
n1 = [' ' + i[0:3] for i in name1] + [' AVG']
n2 = name2 + ['AVERAGE']
df = pd.DataFrame({n2[0]: [round(i, 6) for i in Z[0]],
n2[1]: [round(i, 6) for i in Z[1]],
n2[2]: [round(i, 6) for i in Z[2]],
n2[3]: [round(i, 6) for i in Z[3]]},
index=n1)
print(df)
indx1 = np.argmax(np.average(var, 1))
indx2 = np.argmax(np.average(var, 0))
print(' BEST USING', name2[indx1], name1[indx2], '\n')
return np.array(var)
# READ DATA
# --------------------------------------------------------------------------
with open('DataWeight.csv', newline='') as f:
reader = f.readlines()
reader = [x.replace(',', ' ') for x in reader]
tfidf = []
for row in reader:
tfidf.append([float(j) for j in T(row)])
tfidf = np.array(tfidf)
with open('Dictionary.csv') as f:
dictionary = f.readlines()
dictionary = [x.strip() for x in dictionary]
with open('SetSyns.csv') as f:
reader = f.readlines()
reader = [x.replace(',', ' ') for x in reader]
setsyn = []
for row in reader:
setsyn.append(' '.join([word for word in T(row)]))
for enu, i in enumerate(dictionary):
if i.startswith('syn'):
idx = int(i[3::]) - 1
dictionary[enu] = '-'.join([x for x in T(setsyn[idx])])
nclas = [20, 19, 102, 73, 38, 8, 26]
count, clas = 0, []
for i in nclas:
for j in range(i):
clas.append(count)
count += 1
# CLUSTERING METHODS
# --------------------------------------------------------------------------
def K_MEANS(data, k, initc, opt, negh):
if opt:
initc, temp, negh = FindInitCent(data, k)
cent1 = data[initc]
clus1 = FindCluster(data, cent1, negh)
convergent = False
while not convergent:
cent2 = FindCentroid(data, clus1, k)
clus2 = FindCluster(data, cent2, negh)
if clus1 == clus2:
convergent = True
else:
clus1 = clus2
if not opt:
return clus2, FindSplit(data, cent2, clus2)
return clus2
def BKMEANS(idx, citer, k, nglob):
initc, temp, negh = FindInitCent(tfidf[idx], 2)
subc, avg = K_MEANS(tfidf[idx], 2, initc, False, negh)
for i, j in enumerate(idx):
citer[j] = subc[i] + k
nglob = nglob + avg
indxmax = np.argmax(nglob)
if k + 2 == K:
return [int(i) for i in list(citer)]
citer = np.where(citer == indxmax, -1, np.where(citer > indxmax, citer - 1, citer))
idxsplt = np.arange(len(citer))[np.isin(citer, -1)]
k += 1
nglob.remove(nglob[indxmax])
return BKMEANS(idxsplt, citer, k, nglob)
def KMEDOID(data, k):
M, simbank, negh = FindInitCent(data, k)
if sim != 'CORRELATION':
lmax = len(data)
lnkbank = pairwise_distances(negh, metric=link)
simbank = alfa * (lnkbank / lmax) + (1 - alfa) * simbank
O = [i for i in range(len(data)) if i not in M]
costmax = sum(np.max(simbank[M], 0))
convergent = False
while not convergent:
Msave = [i for i in M]
for i in range(len(O)):
for j in range(len(M)):
O[i], M[j] = M[j], O[i]
costiter = sum(np.max(simbank[M], 0))
if costiter > costmax:
costmax = costiter
else:
O[i], M[j] = M[j], O[i]
if len(set(Msave) - set(M)) == 0:
clus = FindCluster(data, data[M], negh)
convergent = True
return clus
def F_Purity(clus, clas, k):
n = len(clus)
FM, PR = [], []
for i in range(k):
niF = clas.count(i)
njP = clus.count(i)
idF1 = np.arange(n)[np.isin(clas, i)]
idP1 = np.arange(n)[np.isin(clus, i)]
Fij, nijP = [], []
for j in range(k):
njF = clus.count(j)
idF2 = np.arange(n)[np.isin(clus, j)]
idP2 = np.arange(n)[np.isin(clas, j)]
nijF = len(np.intersect1d(idF1, idF2))
nijP.append(len(np.intersect1d(idP1, idP2)))
if nijF == 0:
Fij.append(0)
else:
Pij = nijF / njF
Rij = nijF / niF
Fij.append((2 * Pij * Rij) / (Pij + Rij))
Pj = (1 / njP) * max(nijP)
PR.append((njP / n) * Pj)
FM.append((niF / n) * max(Fij))
return round(sum(FM), 6), round(sum(PR), 6)
# MAIN PROGRAM
# --------------------------------------------------------------------------
K = 7
alfa = 0.9
name1 = ['COSINE', 'CORRELATION', 'JACCARD']
name2 = ['K_MEANS', 'BKMEANS', 'KMEDOID']
teta = 0.0005
TN = 0.1
accur = 0.0005
print('EXPERIMENT USING TETA =', teta, '-', TN, 'WITH', accur, 'ACCURACY')
print('AVERAGE PAIRWISE SIMILARITY:')
for i in name1:
meas = getSim(i)
print('>> ', str.upper(i), ':', np.average(1 - pairwise_distances(tfidf, metric=meas)))
FMPure, N = [], 0
while round(teta, 6) <= TN:
print('\n>> FOR TETA =', teta)
FP = []
for sim in name1:
meas = getSim(sim)
for meth in name2:
# print(' <> USING',sim,meth)
# print(' Average Similarity:',np.average(1-pairwise_distances(tfidf,metric=meas)))
clus = getMeth(tfidf, K, meth)
cent = FindCentroid(tfidf, clus, K)
FP = np.append(FP, F_Purity(clus, clas, K))
# ExpressionText(cent,clus,dictionary)
FP = np.array(FP).reshape(9, 2)
print('\n RESULT')
print(' ------------------------------------------')
print(' F-MEASURE')
temp = PrintResult(np.transpose(FP.reshape(3, 3, 2))[0])
print(' PURITY')
temp = PrintResult(np.transpose(FP.reshape(3, 3, 2))[1])
FMPure = np.append(FMPure, FP)
teta += accur;
N += 1
FMPure = np.reshape(FMPure, (N, 9, 2))
Rerata = np.average(FMPure, 0).reshape(3, 3, 2)
print('\nAVERAGE', N, 'TRIALS')
print('-------------------------------------------')
print('F-MEASURE')
Fdata = PrintResult(np.transpose(Rerata)[0])
print('PURITY')
Pdata = PrintResult(np.transpose(Rerata)[1])
df = pd.DataFrame(np.vstack((Fdata, Pdata)))
df.to_csv('FMPURITY.csv', index=False, header=False)