-
Notifications
You must be signed in to change notification settings - Fork 13
/
model_train.py
155 lines (106 loc) · 3.51 KB
/
model_train.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
# coding: utf-8
import json
import pandas as pd
import os
import numpy as np
import pickle
from glob import glob
from tqdm import tqdm
from sklearn.decomposition import PCA
from sklearn.preprocessing import normalize
from sklearn.ensemble import ExtraTreesRegressor
from keras.applications.vgg16 import VGG16
from keras.preprocessing import image
from keras.applications.vgg16 import preprocess_input
from keras import backend as K
# extract features from images
model = VGG16(weights='imagenet')
prev_layer = model.layers[-2]
dense = K.function(model.inputs, [prev_layer.output])
imgs = glob('./photos/*.jpg')
res = {}
for img_path in tqdm(imgs):
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
features, = dense([x])
res[img_path] = features[0]
keys = list(res.keys())
imgs = [res[k] for k in keys]
imgs = np.array(imgs)
codes = [k[9:-4] for k in keys]
pca = PCA(n_components=128, random_state=1)
imgs_pca = pca.fit_transform(imgs)
imgs_pca = normalize(imgs_pca)
with open('pca.bin', 'wb') as f:
pickle.dump(pca, f)
# read the training data
def unwrap(d):
res = {}
res['updated'] = d['updated']
ann = d['annotations']
entities = {}
for e in ann.get('webDetection', {}).get('webEntities', []):
if 'description' not in e:
continue
if 'score' not in e:
continue
desc = e['description'].lower()
score = e['score']
entities[desc] = score
res['ann_entities'] = entities
insta = d['instagram']
res['insta_code'] = insta['code']
res['insta_dimensions_h'] = insta['dimensions']['height']
res['insta_dimensions_w'] = insta['dimensions']['width']
res['insta_caption'] = insta.get('caption', '').lower()
res['insta_comments_disabled'] = insta['comments_disabled']
res['insta_comments'] = insta['comments']['count']
res['insta_date'] = insta['date']
res['insta_likes'] = insta['likes']['count']
res['insta_owner'] = insta['owner']['id']
res['insta_thumbnail_src'] = insta['thumbnail_src']
res['insta_is_video'] = insta['is_video']
res['insta_id'] = insta['id']
res['insta_display_src'] = insta['display_src']
return res
with open('dataset.json') as f:
data = json.load(f)
df_data = []
for u in data:
df_user = pd.DataFrame([unwrap(d) for d in u['posts']])
df_user['username'] = u['username']
df_data.append(df_user)
df_data = pd.concat(df_data).reset_index(drop=1)
codes_idx = {c: i for (i, c) in enumerate(codes)}
df_data['img_idx'] = df_data.insta_code.apply(lambda c: codes_idx.get(c, -1))
df_data = df_data[df_data.img_idx != -1].reset_index(drop=1)
# normalization of target
means = df_data.groupby('username').insta_likes.mean()
stds = df_data.groupby('username').insta_likes.std()
with open('means_stds.bin', 'wb') as f:
md = means.to_dict()
sd = stds.to_dict()
pickle.dump((md, sd), f)
means_series = means[df_data.username].reset_index(drop=1)
stds_series = stds[df_data.username].reset_index(drop=1)
y_norm = (df_data.insta_likes - means_series) / stds_series
# train a model
X = imgs_pca[df_data.img_idx.values]
y = y_norm.values
et_params = dict(
n_estimators=200,
criterion='mse',
max_depth=30,
min_samples_split=6,
min_samples_leaf=6,
max_features=50,
bootstrap=False,
n_jobs=-1,
random_state=1
)
et = ExtraTreesRegressor(**et_params)
et.fit(X, y)
with open('et.bin', 'wb') as f:
pickle.dump(et, f)