-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
153 lines (123 loc) · 5.98 KB
/
main.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
import code
import numpy as np
import utils
import data_creator
import ransac_model
import os
import pickle
import test
import csv
def main(main_folder, noise, n_images, noise_file=None):
verbose = False
plot_figures = False
video = False
#Create base objects and their presence or not in the image
objects = ['square',
'triangle',
'square',
]
# Create image and save all data parameters
data_parameters = {'is_constrained': True,
'noise_type': 'template',
'std_noise': noise,
}
# Models to test with different initial values of lambda
models = 6 * ['basic_model'] + 6 * ['sinkhorn_model_annealing_stop']
model_modules = list(map(__import__, models))
lambda_0 = 2 * [50, 100, 500, 1000, 2000, 5000]
#Load CCAE images dataset
if noise_file is not None:
with open(noise_file, 'rb') as f:
loaded_data = pickle.load(f)
sims = 500
restarts = 5
r_mnk = {key: [] for key in range(n_images)}
for ii in range(n_images):
# Get image from CCAE dataset or generate one from the generator
if noise_file is not None:
data_model = data_creator.load_image(loaded_data, objects, ii)
else:
# Generate a random visible objects vector, removing blank images
visible_objects = np.random.randint(0, 2, len(objects))
while not sum(visible_objects):
visible_objects = np.random.randint(0, 2, len(objects))
# Create image
data_model = data_creator.create_image(objects, visible_objects, data_parameters)
# Continue if the image is blank
if not sum(data_model['visible_objects']):
print('Figure ' + str(ii) + ' empty, skipping')
continue
####################
# Run RANSAC and save results
X_obj_est, assignment, X_transformed = ransac_model.run(data_model)
# Plot the evolution of the VBEM
save_folder = main_folder + '/Figures' + str(ii) + '/RANSAC/'
if plot_figures:
utils.save_figures_ransac(save_folder, data_model, X_transformed, data_parameters['is_constrained'])
#Save data model
ransac_model.save_results(save_folder, data_model, X_obj_est, X_transformed, assignment)
######################
# Get all test multiple initializations for all methods before start.
# All methods use the same random initialization function
for rr in range(restarts):
r_mnk[ii].append(model_modules[0].r_mnk_initialization(data_model))
# Compute GCM-perm for the same image with a different initialization of r_mnk
#Compare different models under same initialization
for mm, model in enumerate(models):
#Test different restarts
final_ELBO = -np.inf
for rr in range(restarts):
# Set model hyperparameters
hyper_params = model_modules[mm].hyperparams_initialization(data_model, lambda_0=lambda_0[mm])
# Initialize dict with all variables necessary for the inference
params = model_modules[mm].params_initialization(r_mnk[ii][rr], data_model, hyper_params)
# VBEM
new_ELBO = -np.inf
ELBO_epoch = [new_ELBO]
results = []
for ss in range(sims):
#Update parameters and get ELBO
params, ELBO_terms, score = model_modules[mm].params_update(data_model, params, hyper_params)
ELBO = np.sum(ELBO_terms)
# Save results
results.append([ELBO] + ELBO_terms + [score])
if verbose and not ss % 50:
print("Init {} - Iter {} - ELBO: {:4f}, log_x: {:4f}, KL_Y: {:4f}, KL_Z: {:4f}, Score: {:4f}"
.format(ii,ss,ELBO,ELBO_terms[0],-ELBO_terms[1],-ELBO_terms[2],score))
#stopping criteria
ELBO_epoch += [ELBO]
if model_modules[mm].stop(ELBO_epoch, new_ELBO, hyper_params, params, data_model, lambda_0[mm]):
break
else:
new_ELBO = ELBO
# code.interact(local=dict(globals(), **locals()))
# Check whether the current restart is better than the previous one
if ELBO > final_ELBO:
final_ELBO = ELBO
final_params = params
final_results = results
# Exit if solution obtained (not necessary, but speeds up computation)
if utils.is_correct(params, data_model):
break
print('Figure ' + str(ii) + ' ' + model + ', Lambda ' + str(lambda_0[mm]) + \
' - Restarts: ' + str(rr) + ': ' + str(utils.is_correct(final_params, data_model)))
# code.interact(local=dict(globals(), **locals()))
#Saving results
save_folder = main_folder + '/Figures' + str(ii) + '/' + model + '_' + str(lambda_0[mm]) + '/'
utils.save_results(final_results, data_model, final_params, hyper_params, save_folder)
#Plot the evolution of the VBEM
if plot_figures:
utils.save_figures(save_folder, data_model, final_params, data_parameters['is_constrained'])
# Create video with the inference evolution
if video:
utils.video_creation(save_folder, np.array(final_results), final_params, data_model, frames=1000)
# code.interact(local=dict(globals(), **locals()))
if __name__ == '__main__':
# Run this several times
n_images = 512
# noise_levels = [0.01, 0.05, 0.1, 0.5]
noise_levels = [0.25]
for noise in noise_levels:
main_folder = 'Test_noise_025'
# main(main_folder, noise, n_images)
main(main_folder, noise, n_images, 'valid_report_vals_025.pkl')