/
distance_comparaison.py
333 lines (298 loc) · 14.4 KB
/
distance_comparaison.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
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
matplotlib.rcParams['pdf.fonttype'] = 42
matplotlib.rcParams['ps.fonttype'] = 42
import GROMOV_personal as gromov
import pickle
import time
import ot
import networkx as nx
import argparse
def main(entropy=True,
epsilons=[0, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 5, 10],
points=[20, 50, 100, 500]):
if entropy:
nb_iter = 1
np.random.seed(12345)
n_samples_s = 50
n_samples_t = n_samples_s
clique_size_s = 20
clique_size_t = clique_size_s
variance_s = 5
variance_t = variance_s
p_in_s = 0.5
p_in_t = p_in_s
ps_out_s = 0.1
ps_out_t = ps_out_s
# generate synthetic graph
Gs = nx.gaussian_random_partition_graph(n=n_samples_s, s=clique_size_s,
v=variance_s,
p_in=p_in_t, p_out=ps_out_s,
directed=False,
seed=np.random.randint(10000))
Gt = nx.gaussian_random_partition_graph(n=n_samples_t, s=clique_size_t,
v=variance_t,
p_in=p_in_t, p_out=ps_out_t,
directed=False,
seed=np.random.randint(10000))
Cs, Ct = nx.adjacency_matrix(Gs).toarray(), nx.adjacency_matrix(Gt).toarray()
Cs = Cs * np.random.uniform(0.25, 1.75, Cs.shape)
Ct = Ct * np.random.uniform(0.25, 1.75, Ct.shape)
Cs += Cs.T
Ct += Ct.T
def loss_fun(C1, C2):
return np.abs(C1 - C2)
distance = {}
# epsilon_list = [0.1, 0.5, 1]
for epsilon in epsilons:
print("\nepsilon :", epsilon)
T = gromov.entropic_gromov_wasserstein(C1=Cs,
C2=Ct,
p=ot.unif(n_samples_s),
q=ot.unif(n_samples_t),
loss_fun=loss_fun,
epsilon=epsilon)
std = None
t = []
for i in range(nb_iter):
np.random.seed(123 + i)
time_init = time.time()
d = gromov.compute_distance(T=T, C1=Cs, C2=Ct, loss=loss_fun)
t.append(time.time() - time_init)
distance[str(epsilon) + "real" + "distance"] = d
distance[str(epsilon) + "real" + "std"] = std
distance[str(epsilon) + "real" + "time"] = t
std = None
t = []
for i in range(nb_iter):
np.random.seed(123 + i)
time_init = time.time()
d, std, std_total = gromov.compute_distance_sampling_both(T=T, C1=Cs, C2=Ct,
loss_fun=loss_fun,
std=True,
std_total=True)
t.append(time.time() - time_init)
distance[str(epsilon) + "sample" + "distance"] = d
distance[str(epsilon) + "sample" + "std"] = std
distance[str(epsilon) + "sample" + "std_total"] = std_total
distance[str(epsilon) + "sample" + "time"] = t
std = None
t = []
for i in range(nb_iter):
np.random.seed(123 + i)
time_init = time.time()
T_ = gromov.sparsify_T(T)
d = gromov.compute_distance_sparse(Cs, Ct, loss_fun,
T_, dim_T=T.shape)
t.append(time.time() - time_init)
distance[str(epsilon) + "sparse" + "distance"] = d
distance[str(epsilon) + "sparse" + "std"] = std
distance[str(epsilon) + "sparse" + "time"] = t
print(distance)
with open("pickle_distance/entropy_distance.pickle", 'wb') as handle:
pickle.dump(distance, handle)
else:
nb_iter = 10
def loss_fun(C1, C2):
return (C1 - C2) ** 2
distance = {}
epsilon = 0.05
for p, point in enumerate(points):
print("\npoint", point)
np.random.seed(123456 + p)
n_samples_s = point
n_samples_t = n_samples_s
clique_size_s = int(point / 4)
clique_size_t = clique_size_s
variance_s = 5
variance_t = variance_s
p_in_s = 0.5
p_in_t = p_in_s
ps_out_s = 0.1
ps_out_t = ps_out_s
# generate synthetic graph
Gs = nx.gaussian_random_partition_graph(n=n_samples_s, s=clique_size_s,
v=variance_s,
p_in=p_in_t, p_out=ps_out_s,
directed=False,
seed=np.random.randint(10000))
Gt = nx.gaussian_random_partition_graph(n=n_samples_t, s=clique_size_t,
v=variance_t,
p_in=p_in_t, p_out=ps_out_t,
directed=False,
seed=np.random.randint(10000))
Cs, Ct = nx.adjacency_matrix(Gs).toarray(), nx.adjacency_matrix(Gt).toarray()
Cs = Cs * np.random.uniform(0.25, 1.75, Cs.shape)
Ct = Ct * np.random.uniform(0.25, 1.75, Ct.shape)
Cs += Cs.T
Ct += Ct.T
def Cs_(array_ij, array_kl=None):
a = Cs[array_ij[0], array_kl]
return a.reshape(len(array_ij[0]), -1, 1)
def Ct_(array_ij, array_kl=None):
a = Ct[array_ij[0], array_kl]
return a.reshape(len(array_ij[0]), 1, -1)
T = gromov.Generalisation_OT(C1=Cs_,
C2=Ct_,
loss_fun=loss_fun,
T=np.outer(ot.unif(n_samples_s), ot.unif(n_samples_t)),
epsilon_min=epsilon, # swap between the epsilon !
nb_iter_batch=1000,
epsilon_init=0,
KL=1)
std = None
t = []
if point <= 500:
for i in range(nb_iter):
np.random.seed(12345 * p + i)
time_init = time.time()
d = gromov.compute_distance(T=T, C1=Cs, C2=Ct, loss=loss_fun)
t.append(time.time() - time_init)
else:
t = d = None
distance[str(point) + "real" + "distance"] = d
distance[str(point) + "real" + "std"] = std
distance[str(point) + "real" + "time"] = t
std = None
t = []
for i in range(nb_iter):
np.random.seed(12345 * p + i)
time_init = time.time()
d, std, std_total = gromov.compute_distance_sampling_both(T=T, C1=Cs, C2=Ct,
loss_fun=loss_fun,
std=True,
std_total=True)
t.append(time.time() - time_init)
distance[str(point) + "sample" + "distance"] = d
distance[str(point) + "sample" + "std"] = std
distance[str(point) + "sample" + "std_total"] = std_total
distance[str(point) + "sample" + "time"] = t
std = None
t = []
for i in range(nb_iter):
np.random.seed(12345 * p + i)
time_init = time.time()
T_ = gromov.sparsify_T(T)
d = gromov.compute_distance_sparse(Cs, Ct, loss_fun, T_, dim_T=T.shape)
t.append(time.time() - time_init)
distance[str(point) + "sparse" + "distance"] = d
distance[str(point) + "sparse" + "std"] = std
distance[str(point) + "sparse" + "time"] = t
with open("pickle_distance/points_distance.pickle", 'wb') as handle:
pickle.dump(distance, handle)
def plot(log_scale=True, entropy=True, epsilon_range=None, figsize1=10, figsize2=10, marker=["o", "x", "^"]):
if entropy:
with open("pickle_distance/entropy_distance.pickle", 'rb') as handle:
distance = pickle.load(handle)
if epsilon_range is None:
epsilon_range = [0.0, 0.005, 0.01, 0.05, 0.1, 0.5, 1.0] # , 5, 10]
else:
with open("pickle_distance/points_distance.pickle", 'rb') as handle:
distance = pickle.load(handle)
if epsilon_range is None:
epsilon_range = [10, 20, 50, 100, 500]
name_algo2 = ["Real", "Sampled", "Sparse"]
for n_i, name_algo in enumerate(["real", "sample", "sparse"]):
time_list = []
distance_list = []
std_list = []
std_total_list = []
for epsilon in epsilon_range:
if distance[str(epsilon) + name_algo + "time"] is not None:
time_list.append(distance[str(epsilon) + name_algo + "time"])
distance_list.append(distance[str(epsilon) + name_algo + "distance"])
if name_algo == "sample":
std_total_list.append(distance[str(epsilon) + name_algo + "std_total"])
std_list.append(distance[str(epsilon) + name_algo + "std"])
time_list = np.array(time_list)
distance_list = np.array(distance_list)
std_list = np.array(std_list)
std_total_list = np.array(std_total_list)
plt.figure(1, figsize=(figsize1, figsize2))
plt.plot(range(time_list.shape[0]), np.mean(time_list, axis=1), marker[n_i], label=name_algo2[n_i],
linestyle="-", markersize=8)
# plt.plot(range(time_list.shape[0]), np.mean(time_list, axis=1), marker[n_i], label=name_algo2[n_i])
plt.fill_between(range(time_list.shape[0]),
np.mean(time_list, axis=1) - np.std(time_list, axis=1),
np.mean(time_list, axis=1) + np.std(time_list, axis=1),
alpha=0.3)
# name_algo2[1] = "Sampled mean"
plt.figure(2, figsize=(figsize1, figsize2))
plt.plot(range(time_list.shape[0]), distance_list, marker[n_i], label=name_algo2[n_i], linestyle="-",
markersize=8)
if std_list[0] is not None and n_i == 1:
plt.fill_between(range(time_list.shape[0]),
distance_list - 2 * std_list,
distance_list + 2 * std_list,
alpha=0.3, color="orange")#, label="2 Standard deviation")
# alpha = 0.3, label = "Std with stratification")
if name_algo == "sample" and False:
plt.fill_between(range(time_list.shape[0]),
distance_list - 0.25 * std_total_list,
distance_list + 0.25 * std_total_list,
alpha=0.1, color="black", label="Std without stratification")
plt.figure(1)
handles, labels = plt.gca().get_legend_handles_labels()
order = [0, 1, 2]
plt.legend([handles[idx] for idx in order], [labels[idx] for idx in order], prop={'size': 12})
# plt.title("Time needed to compute T and Wasserstein")
if entropy:
plt.xlabel("Entropy: $\epsilon$", fontsize=14)
else:
plt.xlabel("Number of points: $N$", fontsize=14)
plt.ylabel("Computational time (s)", fontsize=14)
plt.xticks(range(len(epsilon_range)), epsilon_range, fontsize=10)
plt.yticks(fontsize=10)
if not entropy:
plt.yscale("log")
if entropy:
plt.savefig("./figure/epsilon_time.pdf", bbox_inches="tight")
# plt.savefig("./figure/epsilon_time.png", bbox_inches="tight")
else:
plt.savefig("./figure/point_time.pdf", bbox_inches="tight")
# plt.savefig("./figure/point_time.png", bbox_inches="tight")
plt.figure(2)
handles, labels = plt.gca().get_legend_handles_labels()
order = [0, 1, 2]
# plt.title("Wasserstein Distance")
if entropy:
plt.xlabel("Entropy: $\epsilon$", fontsize=14)
plt.legend([handles[idx] for idx in order], [labels[idx] for idx in order], prop={'size': 12})
else:
plt.xlabel("Number of points: $N$", fontsize=14)
plt.legend([handles[idx] for idx in order], [labels[idx] for idx in order],
loc=2, prop={'size': 12})
plt.ylabel("GW value : $\mathcal{E}$ (T)", fontsize=14)
plt.xticks(range(len(epsilon_range)), epsilon_range, fontsize=10)
plt.yticks(fontsize=10)
if entropy:
plt.savefig("./figure/epsilon_distance.pdf", bbox_inches="tight")
# plt.savefig("./figure/epsilon_distance.eps", bbox_inches="tight")
else:
plt.savefig("./figure/point_distance.pdf", bbox_inches="tight")
# plt.savefig("./figure/point_distance.eps", bbox_inches="tight")
plt.plot()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='GROMAP')
parser.add_argument("-p", "--plot", action="store_true", help="Plot the result")
parser.add_argument("-e", "--entropy", action="store_true", help="Use entropy experiment")
parser.add_argument("--epsilons", default="0.0,0.005,0.01,0.05,0.1,0.5,1.0", help="epsilons values")
parser.add_argument("--points", default="10,20,50,100,500", help="number of points")
args = parser.parse_args()
if args.plot:
if args.entropy:
list_range = args.epsilons
else:
list_range = args.points
plot(entropy=args.entropy, epsilon_range=list_range)
else:
epsilons = args.epsilons.split(",")
points = args.points.split(",")
for i in range(len(epsilons)):
epsilons[i] = float(epsilons[i])
for i in range(len(points)):
points[i] = int(points[i])
main(entropy=args.entropy,
epsilons=epsilons,
points=points)