Newer
Older
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
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
"""
Effect of lambda: LD
Dataset-1
"""
import numpy as np
import pandas as pd
import utils
import sbm_core
import math
from itertools import combinations
import itertools
from sklearn.metrics.cluster import adjusted_rand_score
# Initilaize
np.random.seed(155)
res = np.zeros((9,5) , dtype=float)
for _itr in range(8,9):
num_roles=2
num_vertices=20
num_segments = 4
num_levels = 2
NO_SAMPLES= 200
nodes = np.arange(num_vertices)
lamda_arr_act = np.zeros((num_roles, num_roles,num_levels) , dtype=float)
H =num_levels
print('k-h levels %d'%(num_levels))
# h-level lambda estimates
lambda_estimates_h = np.random.rand(num_roles, num_roles, H)
lambda_estimates_h = 1e-2*np.random.randint(11,99, size=(num_roles, num_roles, H))
# Make high variant lambdas
yu = (9-_itr)*.1
lambda_estimates_h[0,0,:] = [yu, 0.1]
lambda_estimates_h[0,1,:] = [0.1, yu]
lambda_estimates_h[1,0,:] = lambda_estimates_h[0,1,:]
lambda_estimates_h[1,1,:] = [yu, yu]
l1 =list(range(0, H))
l2 = []
if num_segments > num_levels:
l2 = [np.random.randint(0,H) for i in range(num_segments-H)]
# Mapping from segment to a level
g_mapping= np.array(l1 + l2)
print('g mapping {}'.format(g_mapping))
# initilaize group assignment randomly
group_assignment_arr= np.random.randint(num_roles, size=(num_levels,num_vertices))
# node-group dictionary
group_dic = {}
for i in range(0,num_levels ):
level = i
group_dic_level = {}
keys = nodes
values = list(group_assignment_arr[level])
group_dic_level = dict(zip(keys,values))
group_dic[i] = group_dic_level
print('initial')
# print(group_dic)
for e_h in range(0,num_segments):
g_a = group_dic[g_mapping[e_h]]
list_of_groups= [[] for _ in range(num_roles)]
for idx, val in g_a.items():
list_of_groups[val].append(idx)
print('group assignments {}: {}'.format(e_h,list_of_groups))
# Plotting
#Initialize lamda
lamda_arr = np.zeros((num_roles, num_roles,num_segments) , dtype=float)
for d in range(0, num_segments):
for k in range(0, num_roles):
for g in range(k, num_roles):
lamda_arr[k,g, d]= lambda_estimates_h[k,g,g_mapping[d]]
lamda_arr[g,k, d]= lamda_arr[k,g, d]
change_points_arr = np.zeros((num_roles, num_roles, num_segments+1) , dtype=int)
df_all= None
points= list(range(0, (num_segments+1)*NO_SAMPLES, NO_SAMPLES))
list1 = []
level_seg_mapping = {}
for d in range(num_segments):
level = g_mapping[d]
if level in level_seg_mapping:
level_seg_mapping[level].append(d)
else:
level_seg_mapping[level] = []
level_seg_mapping[level].append(d)
# %%
# Generate piecewise non-homogeneous poisson process
tot_count = np.zeros((num_levels) , dtype=float)
com_len = np.zeros((num_levels) , dtype=float)
# for pair in comb:
for i in range(0,num_levels):
# i = g_mapping[d]
group_assignment = group_assignment_arr[i]
print(group_assignment)
list_of_groups= [[] for _ in range(num_roles)]
for idx, val in enumerate(group_assignment):
list_of_groups[val].append(nodes[idx])
# print(list_of_groups)
size_all_pairs = {}
for kk in range(0, num_roles):
for gg in range(kk, num_roles):
U=list_of_groups[kk]
W=list_of_groups[gg]
if kk == gg:
size_all_pairs[kk,gg] = math.comb(len(U), 2)
if kk != gg:
size_all_pairs[kk,gg] = len(U)*len(W)
for k in range(0, num_roles):
for g in range(k, num_roles):
change_points_arr[k,g,:] = points
lamda_arr[k,g,:] = lamda_arr[g,k,:]
comb = []
if k == g:
comb = list(combinations(list_of_groups[k], 2))
# print(type(comb))
else:
# comb = []
key_data = [list_of_groups[k],list_of_groups[g],]
comb = list(itertools.product(*key_data))
# print(comb)
if len(comb) != size_all_pairs[k,g]:
print('not equal..')
print('d val {}'.format( d))
com_len[i] = len(comb)
# print('comb len {}'.format( com_len[d]))
tot_count[i] = 0
for pair in comb:
s = np.random.poisson(lamda_arr[k,g,d], NO_SAMPLES)
# print(np.count_nonzero(s))
tot_count[i] += np.count_nonzero(s)
list_org=[i for i, e in enumerate(s) if e != 0]
if len(list_org) == 0:
print('zero')
for d in level_seg_mapping[i]:
list1 = [x+points[d] for x in list_org]
df= None
df = pd.DataFrame(data=list1)
df.columns =['timestamp']
# print(list1)
# if max(list1) > 799:
# print('{} {}'.format(d, max(list1)))
N= df.size
# print(pair)
# print(pair[0])
list_start_stations =[pair[0]] * N
list_end_stations =[pair[1]] * N
df['source'] = list_start_stations
df['target'] = list_end_stations
df_all=pd.concat([df_all, df], ignore_index=True)
# for dd in level_seg_mapping:
# dd = d
lamda_arr_act[k,g,i] = round(((tot_count[i])/(NO_SAMPLES*com_len[i])),1)
lamda_arr_act[g,k,i] = lamda_arr_act[k,g,i]
# print('tot count')
# print(tot_count[dd])
# print(' {} {} {} {} : k g d :lamb'.format(k,g,d,lamda_arr_act[g,k,dd]))
print(' {} {} {} {} : k g d :lamb'.format(k,g,i,lamda_arr_act[g,k,i]))
# Remove self loops
df_all = df_all[((df_all['source'] ) != (df_all['target']))]
#sort
df_all=df_all.sort_values('timestamp')
df_all = df_all[['target', 'timestamp','source']]
# Save as .csv file
# df_all.to_csv('./Data/synthetic_ground_truth_g1.csv')
df= None
df=df_all
dest_folder='./Results/synthetic/3'
t_df = df['timestamp']
nodes_arr = np.union1d(df['target'],df['source']).astype(int)
# list of nodes
nodes = nodes_arr.tolist()
num_vertices = len(nodes)
# create a new dictionary - key: node-pair , value: list of timestamps
dic=df.groupby(['source','target'])['timestamp'].apply(list).to_dict()
print('{} {} {} '.format(group_dic, lamda_arr_act,change_points_arr))
# liklihood_sum = sbm_core.mm_compute_cost(group_dic,lamda_arr_act,change_points_arr,num_roles,num_segments,dic,g_mapping)
# print(' Initial Actual likelihood .......%f'%liklihood_sum)
def _swap (row):
if row['source'] > row['target']:
row['source'] , row['target'] =row['target'] , row['source']
return row
# Undirected graph
df=df.apply(lambda row: _swap(row), axis=1)
#scale timestamps for zeroth reference point
refValue = df['timestamp'].min()
df['timestamp'] -= refValue
chg_points = change_points_arr[0,0,:]
ranges_arr = [ [chg_points[s]+1,chg_points[s+1]] for s in range(0,len(chg_points)-1)]
ranges_arr[0][0] = 0
list_time_stamps = list(t_df)
# iterate over timestamps list
gt_arr = list()
for item in list_time_stamps:
# find the segment which the timestamp belongs
# (is dependent on which groups the two nodes belong)
d = sbm_core._findSegment(ranges_arr, len(ranges_arr) , int(item))
gt_arr.append(d)
# Experiment
import experiment
# User parameters
# num_roles=2
# num_segments=10
# num_levels=5# Optional arg
algo_ver=4
dest_folder='./Results/synthetic/'
# tuning parameters
theta = 1e-7
eta = 1
tuning_params= {'theta':theta,'eta':eta}
exp_obj = experiment.Experiment(df,num_roles,num_segments,algo_ver,dest_folder,tuning_params,num_levels,refValue)
# [likelihood_f,group_dic_f] = exp_obj.execute()
[it,ll1,group_dic_d,lambda_estimates,change_points_arr_d]= exp_obj.execute()
# SEGMENTATION ACCURACY
t_df = sorted(t_df)
chg_points = change_points_arr_d[0,0,:]
ranges_arr = [ [chg_points[s]+1,chg_points[s+1]] for s in range(0,len(chg_points)-1)]
ranges_arr[0][0] = 0
list_time_stamps = list(t_df)
# iterate over timestamps list
dis_arr = list()
for item in list_time_stamps:
# find the segment which the timestamp belongs
# (is dependent on which groups the two nodes belong)
d = sbm_core._findSegment(ranges_arr, len(ranges_arr) , int(item))
dis_arr.append(d)
gt_arr= np.array(gt_arr, dtype=np.float64)
dis_arr= np.array(dis_arr, dtype=np.float64)
ind_seg = adjusted_rand_score(gt_arr,dis_arr)
print('ind {} : {}'.format(_itr, ind_seg))
liklihood_sum = sbm_core.mm_compute_cost(group_dic,lamda_arr_act,change_points_arr,num_roles,num_segments,dic,g_mapping)
print(' Initial Actual likelihood .......%f'%liklihood_sum)
print('g mapping {}'.format(g_mapping))
for e_h in range(0,num_segments):
g_a = group_dic[g_mapping[e_h]]
list_of_groups= [[] for _ in range(num_roles)]
for idx, val in g_a.items():
list_of_groups[val].append(idx)
print('group assignments {}: {}'.format(e_h,list_of_groups))
#group ass, of level 1
list_of_groups_1= [[] for _ in range(num_roles)]
#group ass, of level 2
list_of_groups_2= [[] for _ in range(num_roles)]
g1= group_dic_d[0]
g2= group_dic_d[1]
found_cont = 0
for i_h in range(0,num_levels):
# i_h level
grp = group_dic_d[i_h]
list_of_groups_d= [[] for _ in range(num_roles)]
for idx, val in grp.items():
list_of_groups_d[val].append(idx)
ds= list(group_dic_d[i_h].values() )
gt1 = list(g1.values())
gt2 = list(g2.values())
ind1=adjusted_rand_score(ds,gt1)
ind2=adjusted_rand_score(ds,gt2)
d_in = max(ind1,ind2)
found_cont += d_in
ind = found_cont/2
res[_itr][1] = ind
res[_itr][4] = ind_seg
print('end')
# 0.989349
# 0.9899235585218414
# 0.9887209171780673
# 0.9900141929986654
# 0.9900915114849232
# 0.9895393785077311
# 0.9890441642420313
# 0.5056343918828786
# 0.489279
# 2
# 3
# 2
# 3
# 3
# 3
# 4
# 3
# 3