-
Notifications
You must be signed in to change notification settings - Fork 0
/
a31_create_cnn_features_land.py
275 lines (232 loc) · 8.96 KB
/
a31_create_cnn_features_land.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
# -*- coding: utf-8 -*-
__author__ = 'ZFTurbo: https://kaggle.com/zfturbo'
from a00_common_functions import *
GPU_TO_USE = 0
USE_THEANO = 1
# Uncomment if you need to calculate specific fold
# FOLD_TO_CALC = [5]
if USE_THEANO:
os.environ["KERAS_BACKEND"] = "theano"
os.environ["THEANO_FLAGS"] = "device=gpu{},lib.cnmem=0.81".format(GPU_TO_USE, GPU_TO_USE)
else:
os.environ["KERAS_BACKEND"] = "tensorflow"
os.environ["CUDA_VISIBLE_DEVICES"] = "{}".format(GPU_TO_USE)
import random
from a02_zoo import *
random.seed(2016)
np.random.seed(2016)
RESTORE_FROM_LAST_CHECKPOINT = 0
INPUT_PATH = "../input/"
MODELS_PATH = '../models/'
if not os.path.isdir(MODELS_PATH):
os.mkdir(MODELS_PATH)
OUTPUT_PATH = "../subm/"
if not os.path.isdir(OUTPUT_PATH):
os.mkdir(OUTPUT_PATH)
FEATURES_PATH = "../features/"
if not os.path.isdir(FEATURES_PATH):
os.mkdir(FEATURES_PATH)
CODE_COPY_FOLDER = "../models/code/"
if not os.path.isdir(CODE_COPY_FOLDER):
os.mkdir(CODE_COPY_FOLDER)
HISTORY_FOLDER_PATH = "../models/history/"
if not os.path.isdir(HISTORY_FOLDER_PATH):
os.mkdir(HISTORY_FOLDER_PATH)
CACHE_PATH = "../cache/"
if not os.path.isdir(CACHE_PATH):
os.mkdir(CACHE_PATH)
def get_validation_score_land(nfolds, cnn_type):
global FOLD_TO_CALC
from keras.models import load_model
from keras import backend as K
if K.backend() == 'tensorflow':
print('Update dim ordering to "tf"')
K.set_image_dim_ordering('tf')
restore_from_cache = 0
choose = [0, 1, 2, 3, 4, 7, 8, 9, 12, 13, 14, 15, 16]
tbl = pd.read_csv(INPUT_PATH + "train_v2.csv")
labels = tbl['tags'].apply(lambda x: x.split(' '))
counts = defaultdict(int)
for l in labels:
for l2 in l:
counts[l2] += 1
indexes = sorted(list(counts.keys()))
for i in range(len(indexes)):
tbl['label_{}'.format(i)] = 0
files = []
for id in tbl['image_name'].values:
files.append(INPUT_PATH + "/train-jpg/" + id + '.jpg')
files = np.array(files)
lbl = np.zeros((len(labels), len(indexes)))
for j in range(len(labels)):
l = labels[j]
for i in range(len(indexes)):
if indexes[i] in l:
lbl[j][i] = 1
# print(lbl)
print(lbl.shape)
stat = []
kf = KFold(n_splits=nfolds, shuffle=True, random_state=get_random_state(cnn_type))
num_fold = 0
result = np.zeros((len(labels), len(choose)))
for train_ids, valid_ids in kf.split(range(len(files))):
num_fold += 1
if 'FOLD_TO_CALC' in globals():
if num_fold not in FOLD_TO_CALC:
continue
start_time = time.time()
cache_file = CACHE_PATH + '{}_valid_fold_{}_land.pklz'.format(cnn_type, num_fold)
print('Start KFold number {} from {}'.format(num_fold, nfolds))
print('Split train: ', len(train_ids))
print('Split valid: ', len(valid_ids))
valid_files = files[valid_ids]
valid_labels = lbl[valid_ids][:, choose]
if not (os.path.isfile(cache_file) and restore_from_cache):
final_model_path = MODELS_PATH + '{}_fold_{}_land.h5'.format(cnn_type, num_fold)
print('Loading model {}...'.format(final_model_path))
if cnn_type == 'RESNET101' or cnn_type == 'RESNET152' or 'DENSENET' in cnn_type:
model = get_pretrained_model(cnn_type, len(choose), final_layer_activation='sigmoid')
model.load_weights(final_model_path)
weights = model.layers[-1].get_weights()
print(weights[0].shape)
print(weights[1].shape)
else:
c = dict()
c['f2beta_loss'] = f2beta_loss
model = load_model(final_model_path, custom_objects=c)
preds = get_raw_predictions_for_images_v3(model, cnn_type, valid_files)
save_in_file(preds, cache_file)
else:
preds = load_from_file(cache_file)
for i in range(len(valid_ids)):
result[valid_ids[i], :] = preds[i]
print(preds.shape)
print(valid_labels.shape)
best_score = -1
best_thr = -1
for thr1 in range(1, 100):
p = preds.copy()
thr = thr1 / 100
p[p > thr] = 1
p[p <= thr] = 0
score = f2_score(valid_labels, p)
print('THR: {} SCORE: {}'.format(thr, score))
if score > best_score:
best_score = score
best_thr = thr
stat.append((best_score, best_thr))
print('Best score: {} THR: {}'.format(best_score, best_thr))
print('Fold time: {} seconds'.format(time.time() - start_time))
# if num_fold == 1:
# exit()
best_score = -1
best_thr = -1
for thr1 in range(1, 100):
p = result.copy()
thr = thr1 / 100
p[p > thr] = 1
p[p <= thr] = 0
score = f2_score(lbl[:, choose], p)
print('THR: {} SCORE: {}'.format(thr, score))
if score > best_score:
best_score = score
best_thr = thr
print('Best overall score: {} THR: {}'.format(best_score, best_thr))
for i in range(len(stat)):
print('Best score fold {}: {} THR: {}'.format(i+1, stat[i][0], stat[i][1]))
print(result.sum(axis=1))
# Save validation file
out = open(FEATURES_PATH + "valid_{}_score_{}_thr_{}_land.csv".format(cnn_type, best_score, best_thr), "w")
# out = open(FEATURES_PATH + "valid_{}_land.csv".format(cnn_type), "w")
out.write("image_name")
for i in choose:
out.write("," + indexes[i])
out.write("\n")
ids = tbl['image_name'].values
for i in range(len(result)):
out.write(ids[i])
for j in range(len(choose)):
out.write("," + str(result[i][j]))
out.write("\n")
out.close()
return best_score, best_thr
def process_test_land(nfolds, cnn_type, score, thr):
global FOLD_TO_CALC
from keras.models import load_model
from keras import backend as K
if K.backend() == 'tensorflow':
print('Update dim ordering to "tf"')
K.set_image_dim_ordering('tf')
restore_from_cache = 1
choose = [0, 1, 2, 3, 4, 7, 8, 9, 12, 13, 14, 15, 16]
tbl = pd.read_csv(INPUT_PATH + "sample_submission_v2.csv")
indexes = get_indexes()
ids = tbl['image_name'].values
files = []
for id in ids:
files.append("../input/test-jpg/" + id + '.jpg')
files = np.array(files)
# files = files[:100]
preds = []
for num_fold in range(1, nfolds+1):
if 'FOLD_TO_CALC' in globals():
if num_fold not in FOLD_TO_CALC:
continue
cache_file = CACHE_PATH + '{}_test_fold_{}_land'.format(cnn_type, num_fold)
print('Start KFold number {} from {}'.format(num_fold, nfolds))
if os.path.isfile(cache_file) and restore_from_cache:
print('Restore from cache...')
p = load_from_file(cache_file)
else:
final_model_path = MODELS_PATH + '{}_fold_{}_land.h5'.format(cnn_type, num_fold)
print('Loading model {}...'.format(final_model_path))
if cnn_type == 'RESNET101' or cnn_type == 'RESNET152' or 'DENSENET' in cnn_type:
model = get_pretrained_model(cnn_type, len(choose), final_layer_activation='sigmoid')
model.load_weights(final_model_path)
else:
c = dict()
c['f2beta_loss'] = f2beta_loss
model = load_model(final_model_path, custom_objects=c)
p = get_raw_predictions_for_images_v3(model, cnn_type, files)
save_in_file(p, cache_file)
preds.append(p)
preds = np.array(preds)
print(preds.shape)
preds = np.mean(preds, axis=0)
# Save raw file
out = open(FEATURES_PATH + "test_{}_score_{}_thr_{}_land.csv".format(cnn_type, score, thr), "w")
out.write("image_name")
for i in choose:
out.write("," + indexes[i])
out.write("\n")
ids = tbl['image_name'].values
for i in range(len(preds)):
out.write(ids[i])
for j in range(len(choose)):
out.write("," + str(preds[i][j]))
out.write("\n")
out.close()
# Create submission
out = open(OUTPUT_PATH + "subm_{}_score_{}_thr_{}_land.csv".format(cnn_type, score, thr), "w")
out.write("image_name,tags\n")
for i in range(len(files)):
out.write(ids[i] + ',')
for j in range(len(choose)):
if preds[i][j] > thr:
out.write(indexes[j] + ' ')
out.write("\n")
out.close()
if __name__ == '__main__':
num_folds = 5
for cnn in ['DENSENET_121']:
best_score, best_thr = get_validation_score_land(num_folds, cnn)
process_test_land(num_folds, cnn, best_score, best_thr)
'''
Validation result:
Best overall score: 0.8783776613969391 THR: 0.19
Best score fold 1: 0.8717132327773861 THR: 0.21
Best score fold 2: 0.8810646984891833 THR: 0.22
Best score fold 3: 0.8828568139672164 THR: 0.19
Best score fold 4: 0.8775218783366753 THR: 0.21
Best score fold 5: 0.8796787233198925 THR: 0.19
'''