-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval.py
444 lines (386 loc) · 14.5 KB
/
eval.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
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
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
import os
import shutil
import argparse
import setproctitle
import scipy.stats
import numpy as np
from collections import Counter
from math import radians, cos, sin, asin, sqrt
from utils import get_gps, read_data_from_file, read_logs_from_file, map_ids_to_tokens_py
import pdb
import pickle
leng = 168
def geodistance(lng1,lat1,lng2,lat2):
lng1, lat1, lng2, lat2 = map(radians, [float(lng1), float(lat1), float(lng2), float(lat2)])
dlon=lng2-lng1
dlat=lat2-lat1
a=sin(dlat/2)**2 + cos(lat1) * cos(lat2) * sin(dlon/2)**2
distance=2*asin(sqrt(a))*6371*1000
distance=round(distance/1000,3)
return distance
class EvalUtils(object):
"""
some commonly-used evaluation tools and functions
"""
@staticmethod
def filter_zero(arr):
"""
remove zero values from an array
:param arr: np.array, input array
:return: np.array, output array
"""
arr = np.array(arr)
filtered_arr = np.array(list(filter(lambda x: x != 0., arr)))
return filtered_arr
@staticmethod
def arr_to_distribution(arr, min, max, bins):
"""
convert an array to a probability distribution
:param arr: np.array, input array
:param min: float, minimum of converted value
:param max: float, maximum of converted value
:param bins: int, number of bins between min and max 区间个数
:return: np.array, output distribution array
"""
distribution, base = np.histogram(
arr, np.arange(
min, max, float(
max - min) / bins))
return distribution, base[:-1]
@staticmethod
def norm_arr_to_distribution(arr, bins=100):
"""
normalize an array and convert it to distribution
:param arr: np.array, input array
:param bins: int, number of bins in [0, 1]
:return: np.array, np.array
"""
arr = (arr - arr.min()) / (arr.max() - arr.min())
arr = EvalUtils.filter_zero(arr)
distribution, base = np.histogram(arr, np.arange(0, 1, 1. / bins))
return distribution, base[:-1]
@staticmethod
def log_arr_to_distribution(arr, min=-30., bins=100):
"""
calculate the logarithmic value of an array and convert it to a distribution
:param arr: np.array, input array
:param bins: int, number of bins between min and max
:return: np.array,
"""
arr = (arr - arr.min()) / (arr.max() - arr.min())
arr = EvalUtils.filter_zero(arr)
arr = np.log(arr)
distribution, base = np.histogram(arr, np.arange(min, 0., 1./bins))
ret_dist, ret_base = [], []
for i in range(bins):
if int(distribution[i]) == 0:
continue
else:
ret_dist.append(distribution[i])
ret_base.append(base[i])
return np.array(ret_dist), np.array(ret_base)
@staticmethod
def get_js_divergence(p1, p2):
"""
calculate the Jensen-Shanon Divergence of two probability distributions
:param p1:
:param p2:
:return:
"""
# normalize
p1 = p1 / (p1.sum()+1e-14)
p2 = p2 / (p2.sum()+1e-14)
m = (p1 + p2) / 2
js = 0.5 * scipy.stats.entropy(p1, m) + \
0.5 * scipy.stats.entropy(p2, m)
return js
class IndividualEval(object):
def __init__(self, data):
self.max_locs = 32400
self.max_distance = 252.007
max_grid = 180
GRID_SIZE = 1000
lon_l, lon_r, lat_b, lat_u = 115.43, 117.52, 39.44, 41.05 # Beijing
earth_radius = 6378137.0
pi = 3.1415926535897932384626
meter_per_degree = earth_radius * pi / 180.0
lat_step = GRID_SIZE * (1.0 / meter_per_degree)
ratio = np.cos((lat_b + lat_u) * np.pi / 360)
lon_step = lat_step / ratio
self.X = []
self.Y = []
for grid_id in range(1,self.max_locs+1):
x = grid_id // max_grid #经度
y = grid_id - x*max_grid - 1 #纬度
self.X.append((x+0.5)*lon_step + lon_l)
self.Y.append((y+0.5)*lat_step + lat_b)
def get_topk_visits(self,trajs, k):
topk_visits_loc = []
topk_visits_freq = []
for traj in trajs:
topk = Counter(traj).most_common(k)
for i in range(len(topk), k):
# supplement with (loc=-1, freq=0)
topk += [(-1, 0)]
loc = [l for l, _ in topk]
freq = [f for _, f in topk]
loc = np.array(loc, dtype=int)
freq = np.array(freq, dtype=float) / trajs.shape[1]
topk_visits_loc.append(loc)
topk_visits_freq.append(freq)
topk_visits_loc = np.array(topk_visits_loc, dtype=int)
topk_visits_freq = np.array(topk_visits_freq, dtype=float)
return topk_visits_loc, topk_visits_freq
def get_overall_topk_visits_freq(self, trajs, k):
_, topk_visits_freq = self.get_topk_visits(trajs, k)
mn = np.mean(topk_visits_freq, axis=0)
return mn / np.sum(mn)
def get_overall_topk_visits_loc_freq_arr(self, trajs, k=1):
topk_visits_loc, _ = self.get_topk_visits(trajs, k)
k_top = np.zeros(self.max_locs, dtype=float)
for i in range(k):
cur_k_visits = topk_visits_loc[:, i]
for ckv in cur_k_visits:
index = int(ckv)
if index == -1:
continue
k_top[index] += 1
k_top = k_top / np.sum(k_top)
return k_top
def get_overall_topk_visits_loc_freq_dict(self, trajs, k):
topk_visits_loc, _ = self.get_topk_visits(trajs, k)
k_top = {}
for i in range(k):
cur_k_visits = topk_visits_loc[:, i]
for ckv in cur_k_visits:
index = int(ckv)
if index in k_top:
k_top[int(ckv)] += 1
else:
k_top[int(ckv)] = 1
return k_top
def get_overall_topk_visits_loc_freq_sorted(self, trajs, k):
k_top = self.get_overall_topk_visits_loc_freq_dict(trajs, k)
k_top_list = list(k_top.items())
k_top_list.sort(reverse=True, key=lambda k: k[1])
return np.array(k_top_list)
def get_geodistances(self, trajs):
distances = []
seq_len = leng
for traj in trajs:
for i in range(seq_len - 1):
lng1 = self.X[traj[i]]
lat1 = self.Y[traj[i]]
lng2 = self.X[traj[i + 1]]
lat2 = self.Y[traj[i + 1]]
distances.append(geodistance(lng1,lat1,lng2,lat2))
distances = np.array(distances, dtype=float)
return distances
def get_distances(self, trajs):
distances = []
seq_len = leng
for traj in trajs:
# trajs:2500,leng
# traj:leng,
for i in range(seq_len - 1):
dx = self.X[traj[i]] - self.X[traj[i + 1]]
dy = self.Y[traj[i]] - self.Y[traj[i + 1]]
# 一个格点id -> 转换成经纬度dx,dy
distances.append(dx**2 + dy**2)
distances = np.array(distances, dtype=float)
return distances
def get_durations(self, trajs):
d = []
for traj in trajs:
num = 1
for i, lc in enumerate(traj[1:]):
if lc == traj[i]:
num += 1
else:
d.append(num)
num = 1
return np.array(d)/leng
def get_gradius(self, trajs):
"""
get the std of the distances of all points away from center as `gyration radius`
:param trajs:
:return:
"""
gradius = []
seq_len = leng
for traj in trajs:
xs = np.array([self.X[t] for t in traj])
ys = np.array([self.Y[t] for t in traj])
xcenter, ycenter = np.mean(xs), np.mean(ys)
dxs = xs - xcenter
dys = ys - ycenter
rad = [dxs[i]**2 + dys[i]**2 for i in range(seq_len)]
rad = np.mean(np.array(rad, dtype=float))
gradius.append(rad)
gradius = np.array(gradius, dtype=float)
return gradius
def get_periodicity(self, trajs):
"""
stat how many repetitions within a single trajectory
:param trajs:
:return:
"""
reps = []
for traj in trajs:
reps.append(float(len(set(traj)))/leng)
reps = np.array(reps, dtype=float)
return reps
def get_timewise_periodicity(self, trajs):
"""
stat how many repetitions of different times
:param trajs:
:return:
"""
pass
def get_geogradius(self, trajs):
"""
get the std of the distances of all points away from center as `gyration radius`
:param trajs:
:return:
"""
gradius = []
for traj in trajs:
xs = np.array([self.X[t] for t in traj])
ys = np.array([self.Y[t] for t in traj])
lng1, lat1 = np.mean(xs), np.mean(ys)
rad = []
for i in range(len(xs)):
lng2 = xs[i]
lat2 = ys[i]
distance = geodistance(lng1,lat1,lng2,lat2)
rad.append(distance)
rad = np.mean(np.array(rad, dtype=float))
gradius.append(rad)
gradius = np.array(gradius, dtype=float)
return gradius
def get_individual_jsds(self, t1, t2):
"""
get jsd scores of individual evaluation metrics
:param t1: test_data
:param t2: gene_data
:return:
"""
t1 = t1-1
t2 = t2-1
# travel distance
d1 = self.get_distances(t1)
d2 = self.get_distances(t2) # shape = 天数,47
# max_d = max(d1[d1.argmax()],d2[d2.argmax()])
d1_dist, _ = EvalUtils.arr_to_distribution(
d1, 0, self.max_distance, 10000)
d2_dist, _ = EvalUtils.arr_to_distribution(
d2, 0, self.max_distance, 10000)
d_jsd = EvalUtils.get_js_divergence(d1_dist, d2_dist)
# gyration radius
g1 = self.get_geogradius(t1)
g2 = self.get_geogradius(t2)
# max_d = max(g1[g1.argmax()],g2[g2.argmax()])
g1_dist, _ = EvalUtils.arr_to_distribution(
g1, 0, self.max_distance, 10000) # 上界?TODO
g2_dist, _ = EvalUtils.arr_to_distribution(
g2, 0, self.max_distance, 10000)
g_jsd = EvalUtils.get_js_divergence(g1_dist, g2_dist)
# stay duration
du1 = self.get_durations(t1)
du2 = self.get_durations(t2)
du1_dist, _ = EvalUtils.arr_to_distribution(du1, 0, 1, leng)
du2_dist, _ = EvalUtils.arr_to_distribution(du2, 0, 1, leng)
du_jsd = EvalUtils.get_js_divergence(du1_dist, du2_dist)
p1 = self.get_periodicity(t1)
p2 = self.get_periodicity(t2)
p1_dist, _ = EvalUtils.arr_to_distribution(p1, 0, 1, leng)
p2_dist, _ = EvalUtils.arr_to_distribution(p2, 0, 1, leng)
p_jsd = EvalUtils.get_js_divergence(p1_dist, p2_dist)
l1 = CollectiveEval.get_visits(t1,self.max_locs)
l2 = CollectiveEval.get_visits(t2,self.max_locs)
l1_dist, _ = CollectiveEval.get_topk_visits(l1, 100)
l2_dist, _ = CollectiveEval.get_topk_visits(l2, 100)
l1_dist, _ = EvalUtils.arr_to_distribution(l1_dist,0,1,500)
l2_dist, _ = EvalUtils.arr_to_distribution(l2_dist,0,1,500)
l_jsd = EvalUtils.get_js_divergence(l1_dist, l2_dist)
f1 = self.get_overall_topk_visits_freq(t1, 100)
f2 = self.get_overall_topk_visits_freq(t2, 100)
f1_dist, _ = EvalUtils.arr_to_distribution(f1,0,1,500) # TODO
f2_dist, _ = EvalUtils.arr_to_distribution(f2,0,1,500)
f_jsd = EvalUtils.get_js_divergence(f1_dist, f2_dist)
return d_jsd, g_jsd, du_jsd, p_jsd, l_jsd, f_jsd
class CollectiveEval(object):
"""
collective evaluation metrics
"""
@staticmethod
def get_visits(trajs,max_locs):
"""
get probability distribution of visiting all locations
:param trajs:
:return:
"""
visits = np.zeros(shape=(max_locs), dtype=float)
for traj in trajs:
for t in traj:
visits[t] += 1
visits = visits / np.sum(visits)
return visits
# 连续呆在一个地方:算一次
@staticmethod
def get_timewise_visits(trajs):
"""
stat how many visits of a certain location in a certain time
:param trajs:
:return:
"""
pass
@staticmethod
def get_topk_visits(visits, K):
"""
get top-k visits and the corresponding locations
:param trajs:
:param K:
:return:
"""
locs_visits = [[i, visits[i]] for i in range(visits.shape[0])]
locs_visits.sort(reverse=True, key=lambda d: d[1])
topk_locs = [locs_visits[i][0] for i in range(K)]
topk_probs = [locs_visits[i][1] for i in range(K)]
return np.array(topk_probs), topk_locs
@staticmethod
def get_topk_accuracy(v1, v2, K):
"""
get the accuracy of top-k visiting locations
:param v1:
:param v2:
:param K:
:return:
"""
_, tl1 = CollectiveEval.get_topk_visits(v1, K)
_, tl2 = CollectiveEval.get_topk_visits(v2, K)
coml = set(tl1) & set(tl2)
return len(coml) / K
def evaluate(datasets):
if datasets == 'telecom':
individualEval = IndividualEval(data='mobile')
start_point = np.load('../data/mobile/start.npy')
else:
individualEval = IndividualEval(data='geolife')
# start_point = np.load('../data/geolife/start.npy')
for num in range(200,300,20):
print("epoch=",num)
gen_path =\
f"../results/results_KL2.1_exp12/checkpoint_epoch{num}/gen_data_epoch{num}.pkl"
test_data = pickle.load(open('../dataset_generation/test_data.pkl', 'rb'))
gene_data = pickle.load(open(gen_path, 'rb'))
print(individualEval.get_individual_jsds(test_data,gene_data))
if __name__ == "__main__":
# global
parser = argparse.ArgumentParser()
parser.add_argument('--task',default='default', type=str)
parser.add_argument('--cuda',default=0,type=int)
parser.add_argument('--datasets',default='geolife',type=str)
opt = parser.parse_args()
max_locs = 32400
evaluate(opt.datasets)