-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiment.py
328 lines (261 loc) · 11.1 KB
/
experiment.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
import random
from scipy.stats import uniform
from dataclasses import dataclass
import numpy as np
from .one_d import AnnotatorSet, random_assignment, SimpleNormalAnnotatorPopulation, sample_tokyo_latitudes,FunctionNormalAnnotatorPopulation
@dataclass
class CGPInstance:
t: int # Number of points to geolocate
w: int # Number of workers
a: int # Number of annotations
t_A: np.ndarray # Task for each annotation
w_A: np.ndarray # Worker for each annotation
ann: np.ndarray # Value for each annotation
class TooManyAnnotationsPerIndividual(Exception):
pass
class TooManyAnnotations(Exception):
pass
class ActiveAnnotationContest:
def __init__(self, points: np.ndarray, ans: AnnotatorSet,
max_total_annotations: int = 0,
max_annotations_per_individual: int = 1e200):
self.points = points
self.annotator_set = ans
self.max_total_annotations = max_total_annotations
self.max_annotations_per_individual = max_annotations_per_individual
self.reset()
@property
def n_points(self):
return len(self.points)
@property
def n_annotators(self):
return self.annotator_set.n_annotators
def reset(self):
self.annotations_per_individual = np.zeros(self.annotator_set.n_annotators)
def batch_request(self, t_A, w_A):
u, u_counts = np.unique(w_A, return_counts=True)
self.annotations_per_individual[u] += u_counts
if np.max(self.annotations_per_individual) > self.max_annotations_per_individual:
self.annotations_per_individual[u] -= u_counts
raise TooManyAnnotationsPerIndividual()
if np.sum(self.annotations_per_individual) > self.max_total_annotations:
print(np.sum(self.annotations_per_individual))
self.annotations_per_individual[u] -= u_counts
raise TooManyAnnotations(np.sum(self.annotations_per_individual))
ann = self.annotator_set.batch_annotation(t_A, w_A, self.points)
return ann
def request(self, t: int, w: int):
if self.annotations_per_individual[w] >= self.max_annotations_per_individual:
raise TooManyAnnotationsPerIndividual()
if np.sum(self.annotations_per_individual) >= self.max_total_annotations:
raise TooManyAnnotations()
self.annotations_per_individual[w] += 1
return self.annotator_set[w].annotate(self.points[t:t+1])
class ActiveAnnotationMethod:
def run(self, exp: ActiveAnnotationContest):
return None #points
def max_redundancy(exp):
max_total_annotations = min(exp.max_annotations_per_individual * exp.n_points, exp.max_total_annotations)
k = max_total_annotations // exp.n_points
return k
def random_annotation(exp, batch_start=0, batch_size=None, k=None):
if batch_size is None:
batch_size = exp.n_points
if k is None:
k = max_redundancy(exp)
t, w = random_assignment(batch_size, exp.n_annotators, k)
t += batch_start
ann = exp.batch_request(t, w)
return t, w, ann
def softmax_stable(x):
return (np.exp(x - np.max(x)) / np.exp(x - np.max(x)).sum())
# Add normalization of the weights
def normalize(tab):
norm=[]
norm=[(k-np.min(tab))/(np.max(tab)-np.min(tab)) for k in tab]
return(np.array(norm))
# +
def sigma_assignment(batch_size, sigmas, k, greediness, previous_anns=None, batch_start=0):
n_annotators = len(sigmas)
if previous_anns is not None:
for i in np.sort(previous_anns)[::-1]:
sigmas = np.delete(sigmas, i)
weights = (sigmas ** (-2))
sum_weights = np.sum(weights)
max_weight = np.max(weights)
min_weight = np.min(weights)
if greediness < 1.00000001:
greediness_factor = 0
else:
greediness_factor = np.log(greediness) / (max_weight - min_weight)
#print(greediness_factor)
p = softmax_stable(greediness_factor * weights)
if previous_anns is not None:
for i in np.sort(previous_anns):
p = np.insert(p, i, 0)
#print("probabilities: ",p)
# print("p=", p, np.sum(p))
# pm = softmax_stable(-greedyness * (sigmas ** (-2)))
# print("pm=", pm, np.sum(pm))
t_A = np.zeros(batch_size * k, dtype=int)
w_A = np.zeros(batch_size * k, dtype=int)
selected_annotators_indexes = np.argsort(p[None, :] * np.random.rand(batch_size, n_annotators), axis=1)[:, -k:]
# print("sel = ", selected_annotators_indexes)
total_annotations = 0
for j in range(n_annotators):
j_point_indices = np.argwhere(selected_annotators_indexes == j)[:, 0]
end_annotations = total_annotations + len(j_point_indices)
t_A[total_annotations:end_annotations] = j_point_indices
w_A[total_annotations:end_annotations] = j
total_annotations = end_annotations
# print("w=", w_A)
return t_A+batch_start, w_A
# +
def sigma_annotation(exp, sigmas, previous_anns=None, greediness=1., batch_start=0, batch_size=None, k=None):
if batch_size is None:
batch_size = exp.n_points
if k is None:
k = max_redundancy(exp)
t, w = sigma_assignment(batch_size, sigmas, k, greediness, previous_anns, batch_start=batch_start)
ann = exp.batch_request(t, w)
return t, w, ann
# -
def mse(points, predictions):
return np.average(np.square(predictions-points))
def mean_norm_error(points, predictions, ord=None):
return (np.linalg.norm(predictions-points, ord=ord) / len(predictions))
#
def mean_location_norm_error(exp, predictions, ord=1):
#print(predictions["locations"] - exp.points)
return (np.linalg.norm(predictions["locations"] - exp.points, ord=ord) / len(exp.points))
#
# def mean_sigma_norm_error(exp, predictions, ord=None):
# sigmas = np.array([a.sigma(0) for a in exp.annotator_set.annotators])
# return (np.linalg.norm(predictions["sigmas"] - sigmas, ord=ord) / len(predictions))
#def mean_location_norm_error(exp, predictions, ord=None):
# return (np.linalg.norm(np.log(np.abs(predictions["locations"] - exp.points)), ord=ord) / len(predictions))
def mean_sigma_norm_error(exp, predictions, ord=None):
sigmas = np.array([a.sigma(0) for a in exp.annotator_set.annotators])
return np.linalg.norm(np.log(predictions["sigmas"]/sigmas), ord=ord) / len(sigmas)
# saving experiments setup
import pickle
#tokyo latitude sampling
def tok(n):
t = np.array(sample_tokyo_latitudes(n)) #we sample the n points
tok_norm=(t-np.min(t))/(np.max(t)-np.min(t)) #we normalize
return(tok_norm)
def save_experiment_setup(params):
# Create a dictionary containing experiment data
np.random.seed(1234) # we set a seed to generate each time the same points/sigmas
random.seed(1234)
n_points, n_annotators, redundancy = (params[0], params[1], params[2]) # choice of general parameters
sig_distr = params[3] # choice of the sigma distrib
if sig_distr == 'uniform':
annotator_population = SimpleNormalAnnotatorPopulation(
uniform(scale=0.1)) # Which value there ?? 0.1 at the beginning
if sig_distr == 'beta':
annotator_population = SimpleNormalAnnotatorPopulation()
point_distr = params[4] # choice of point distrib
if point_distr == 'uniform':
point_distribution = uniform()
points = point_distribution.rvs(n_points)
else:
points = tok(n_points)
list_tru_sig = []
ann_set = annotator_population.sample(n_annotators)
list_true_sig = [ann_set.annotators[k]._sigma for k in range(len(ann_set.annotators))]
# print(list_true_sig)
experiment_data = {
"nb_points": params[0],
"nb_annotators": params[1],
"redundancy": params[2],
"sigma_distrib": params[3],
"point_distrib": params[4],
"points": points,
"sigmas": list_true_sig,
"random_seed": np.random.randint(0, 10000)
}
filename = f"np_{params[0]}_na_{params[1]}_rd_{params[2]}_sd_{params[3]}_pd_{params[4]}_setup.pkl"
# Save the experiment data to a file using pickle
with open(filename, "wb") as f:
pickle.dump(experiment_data, f, protocol=5)
def save_experiment_setup_2(params):
# Create a dictionary containing experiment data
np.random.seed(1234) # we set a seed to generate each time the same points/sigmas
random.seed(1234)
n_points, n_annotators, redundancy = (params[0], params[1], params[2]) # choice of general parameters
sig_distr = params[3] # choice of the sigma distrib
if sig_distr == 'uniform':
annotator_population = FunctionNormalAnnotatorPopulation()
# Which value there ?? 0.1 at the beginning
if sig_distr == 'beta':
annotator_population = FunctionNormalAnnotatorPopulation()
point_distr = params[4] # choice of point distrib
if point_distr == 'uniform':
point_distribution = uniform()
points = point_distribution.rvs(n_points)
else:
points = tok(n_points)
list_tru_sig = []
allx = np.arange(1000) / 1000.
ann_set = annotator_population.sample(n_annotators)
list_true_sig = [ann_set.annotators[k].sigma(allx) for k in range(len(ann_set.annotators))]
# print(list_true_sig)
experiment_data = {
"nb_points": params[0],
"nb_annotators": params[1],
"redundancy": params[2],
"sigma_distrib": params[3],
"point_distrib": params[4],
"points": points,
"sigmas": list_true_sig,
"random_seed": np.random.randint(0, 10000)
}
filename = f"np_{params[0]}_na_{params[1]}_rd_{params[2]}_sd_{params[3]}_pd_{params[4]}_setup_not_constant.pkl"
# Save the experiment data to a file using pickle
with open(filename, "wb") as f:
pickle.dump(experiment_data, f, protocol=5)
# saving experiments setup (only the setup, not te results)
import pickle
def save_experiment(list_param, results):
# Create a dictionary containing experiment data
experiment_data = {
"nb_points": list_param[0],
"nb_annotators": list_param[1],
"redundancy": list_param[2],
"sigma_distrib": list_param[3],
"point_distrib": list_param[4],
"points": results[0],
"sigmas": results[1],
"method_results": {
"name": list_param[5],
"points": results[2][0],
"sigmas": results[3][0]
}
}
# ,
# "conservative2": {
# "points":results[2][1] ,
# "sigmas": results[3][1]
# },
# "10shot": {
# "points": results[2][2],
# "sigmas": results[3][2]
# }
# }
# }
filename = f"np_{list_param[0]}_na_{list_param[1]}_rd_{list_param[2]}_sd_{list_param[3]}_pd_{list_param[4]}.pkl"
# Save the experiment data to a file using pickle
with open(filename, "wb") as f:
pickle.dump(experiment_data, f, protocol=5)
def load_experiment(path):
# Load the pickled file
with open(path, 'rb') as f:
data = pickle.load(f)
return(data)
# load an experiment for a pickle file, return list of params
def load_experiment_setup(path):
# Load the pickled file
with open(path, 'rb') as f:
data = pickle.load(f)
return([data["nb_points"],data["nb_annotators"],data["redundancy"],data["sigma_distrib"],data["point_distrib"],data["points"],data["sigmas"],data["random_seed"]])