-
Notifications
You must be signed in to change notification settings - Fork 1
/
evaluation.py
184 lines (136 loc) · 6.68 KB
/
evaluation.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
import tensorflow as tf
from pathlib import Path
import numpy as np
import os
import pickle as pkl
import matplotlib.pyplot as plt
from sklearn.metrics import accuracy_score
import utils
from utils.options import config
import networks
class HNSEvaluator:
def __init__(self, model, weight_dir, debug=False):
self.weight_dir = Path(weight_dir)
self.weights = self.get_weights()
self.target_dir = Path(str(self.weight_dir).replace('weights/', 'results/'))
self.model = model
self.results = {}
self.debug = debug
def complete_evaluation(self, data, steps, batches_to_save=1):
print('Beginning evaluation...')
self.evaluate(data, steps)
print('Saving results...')
if not self.debug and not self.target_dir.is_dir():
os.makedirs(str(self.target_dir))
self.save_results()
print('Generating sample images...')
self.generate_sample_images(data, batches_to_save)
print('Evaluation complete. Results stored in: {}'.format(str(self.target_dir)))
def get_weights(self):
weights = {a: str(self.weight_dir / 'interm_weights_a_{:.2f}.h5'.format(a))
for a in np.arange(0.1, 1.01, 0.05)
if (self.weight_dir / 'interm_weights_a_{:.2f}.h5'.format(a)).exists()}
weights['final'] = self.weight_dir / 'final_weights.h5'
return weights
def evaluate(self, data, steps):
self.results = {}
for identifier, weight in self.weights.items():
if identifier != 'final':
print('Evaluating model for alpha = {:.2f}'.format(identifier))
self.model.load_weights(weight)
accuracy = self.evaluate_model(data, steps)
self.results[identifier] = accuracy
print('Accuracy: {:.2f}%'.format(accuracy * 100))
return self.results
def evaluate_model(self, data, steps):
accuracy = tf.keras.metrics.Mean()
for i, (x, y) in enumerate(data):
preds = self.model(x)['seeker_output']
y_pred = [np.argmax(p) for p in preds]
y_true = [np.argmax(p) for p in y]
accuracy(accuracy_score(y_true, y_pred))
if i == steps:
break
return accuracy.result().numpy()
def save_results(self):
target_file = str(self.target_dir / 'results.pkl')
if not self.debug:
pkl.dump(self.results, open(target_file, 'wb'))
def generate_sample_images(self, data, batches=1):
if not batches:
return
for a in np.arange(0.1, 1.01, 0.05):
directory = str(self.target_dir / 'a_{:.2f}'.format(a))
if not self.debug and not Path(directory).is_dir():
os.makedirs(directory)
self.model.load_weights(self.weights[a])
for b, (x, y) in enumerate(data):
y_ = self.model(x)
ys = y_['seeker_output']
yh = y_['hider_output']
accuracy = np.where(np.argmax(ys, axis=1) == np.argmax(y, axis=1), '_hit', '_miss')
for j in range(len(x)):
i = (b + 1) * j
if not self.debug:
plt.imsave(directory + '/x' + str(i + 1) + '.png', x[i, ..., 0])
plt.imsave(directory + '/y' + str(i + 1) + accuracy[i] + '.png', yh[i, ..., 0])
if (b + 1) == batches:
break
if __name__ == '__main__':
# Experiment identifier
identifier = config['identifier']
# Model configurations
model_id = config['model']
# Debug mode
debug = config['debug']
# Find weight dir
binary_type = config['binary_type']
stochastic_estimator = config['estimator']
slope_increase_rate = config['rate_per_iteration']
sub_dirs = Path(config['config']) / 'hns' / binary_type
if binary_type == 'stochastic':
sub_dirs = sub_dirs / stochastic_estimator
if stochastic_estimator == 'sa':
sub_dirs = sub_dirs / 'rate_{}'.format(str(config['rate']))
weight_dir = str(Path('weights') / sub_dirs / identifier)
# Load dataset
batch_size = config['batch_size']
image_shape = (config['image_size'], ) * 2
test_images = config['test_images']
channels = config['channels']
input_shape = image_shape + (channels,)
num_classes = config['num_classes']
num_trainings = config['num_trainings']
if config['config'] == 'mnist':
train_set = utils.datagen.mnist(batch_size=batch_size, split='train')
test_set = utils.datagen.mnist(batch_size=batch_size, split='test')
elif config['config'] == 'fashion':
train_set = utils.datagen.fashion(batch_size=batch_size, split='train')
test_set = utils.datagen.fashion(batch_size=batch_size, split='test')
elif config['config'] == 'cifar10':
train_set = utils.datagen.cifar10(batch_size=batch_size, split='train', channels=channels)
test_set = utils.datagen.cifar10(batch_size=batch_size, split='test', channels=channels)
elif config['config'] == 'cifar100':
train_set = utils.datagen.cifar100(batch_size=batch_size, split='train', channels=channels)
test_set = utils.datagen.cifar100(batch_size=batch_size, split='test', channels=channels)
else:
data_dir = Path(config['data_dir'])
train_set = utils.datagen.image_generator(data_dir / 'train',batch_size=batch_size, image_shape=image_shape,
channels=channels)
test_set = utils.datagen.image_generator(data_dir / 'test', batch_size=batch_size, image_shape=image_shape,
channels=channels)
# Load model
hns_model = networks.hns.available_models[model_id](input_shape, num_classes, binary_type=binary_type,
stochastic_estimator=stochastic_estimator,
slope_increase_rate=slope_increase_rate)
# Run evaluator
if num_trainings == 1:
evaluator = HNSEvaluator(model=hns_model, weight_dir=weight_dir, debug=debug)
evaluator.complete_evaluation(test_set, steps=test_images//batch_size, batches_to_save=1)
else:
weight_dirs = sorted(Path(weight_dir).glob('*'))
print('Evaluating for {} experiments'.format(len(weight_dirs)))
for i, weight_dir in enumerate(weight_dirs):
evaluator = HNSEvaluator(model=hns_model, weight_dir=weight_dir, debug=debug)
evaluator.complete_evaluation(test_set, steps=test_images // batch_size, batches_to_save=0)
print('completed evaluation {} of {}'.format(i+1, len(weight_dirs)))