-
Notifications
You must be signed in to change notification settings - Fork 0
/
training.py
161 lines (126 loc) · 4.88 KB
/
training.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
from keras.layers import Input, Dense, Dropout, concatenate
from keras.models import Model, load_model
from keras.optimizers import Adam
import csv
import numpy as np
from sklearn.model_selection import train_test_split
from random import randint, random
import copy
def read_data(name):
input_data = []
label_data = []
with open('./data/' + name, 'r') as f:
print(name)
reader = csv.reader(f)
for i, row in enumerate(reader):
if i == 0:
labels = row
else:
if 'NULL' in row:
pass
else:
# input_data.append(row[:-2] + [row[-2]])
input_data.append(row[:-2])
label_data.append([row[-1]])
return input_data, label_data
def to_categorical(L):
refined = list(set(L)) # remove duplicated data
encoded = [[0 for _ in range(len(refined))] for _ in range(len(L))] # init
for i, l in enumerate(L):
target = refined.index(l)
encoded[i][target] = 1
return encoded
def fit_transform(L, max_val, min_val):
original = L.reshape(1, -1)[0]
return (original - min_val) / (max_val - min_val)
def build_model(input_shapes, n_output):
input_shape_time, input_shape_id = input_shapes
x1 = Input(shape=input_shape_time)
h1 = Dense(10, activation='relu', kernel_initializer='he_normal')(x1)
# h1 = Dropout(0.5)(h1)
x2 = Input(shape=input_shape_id)
h2 = Dense(10)(x2)
# h2 = Dropout(0.5)(h2)
c = concatenate([h1, h2])
h = Dense(6, activation='relu', kernel_initializer='he_normal')(c)
# h = Dropout(0.5)(h)
y = Dense(n_output, activation='linear', name='y')(h)
return Model(
inputs=[x1, x2],
outputs=y,
name='model')
if __name__ == "__main__":
training = False
"""model"""
input_shape_time = (3, )
input_shape_id = (1, )
input_shapes = [input_shape_time, input_shape_id]
output_shape = 1
model = build_model(input_shapes, output_shape)
# model.summary() # TODO: image
model.compile(loss='mae', optimizer=Adam(), metrics=['mse', 'acc'])
print(">>> compiling model complete")
if training:
"""data"""
file_names = list(map(str, range(201901, 201913)))
postfix = '-refined.csv'
x, y = [], []
for file_name in file_names:
_x, _y = read_data(file_name + postfix)
x += _x
y += _y
x = np.array(x).astype('int32')
y = np.array(y).astype('int32')
max_val = np.max(x[:, 3])
min_val = np.min(x[:, 3])
print('max:', max_val, 'min:', min_val) # using max and min in inference time
x_train, x_test, y_train, y_test = train_test_split(
x,
y,
test_size=0.2,
random_state=950327)
x_train_months = x_train[:, 0].reshape(-1, 1)
x_train_weekdays = x_train[:, 1].reshape(-1, 1)
x_train_hours = x_train[:, 2].reshape(-1, 1)
# x_train_days = x_train[:, 3].reshape(-1, 1)
x_train_times = np.concatenate((x_train_months, x_train_weekdays, x_train_hours), axis=-1)
x_train_ids = fit_transform(x_train[:, 3].reshape(-1, 1), max_val, min_val)
x_test_months = x_test[:, 0].reshape(-1, 1)
x_test_weekdays = x_test[:, 1].reshape(-1, 1)
x_test_hours = x_test[:, 2].reshape(-1, 1)
# x_test_days = x_test[:, 3].reshape(-1, 1)
x_test_times = np.concatenate((x_test_months, x_test_weekdays, x_test_hours), axis=-1)
x_test_ids = fit_transform(x_test[:, 3].reshape(-1, 1), max_val, min_val)
print(">>> loading data complete")
"""learn"""
model.fit(
[x_train_times, x_train_ids],
y_train,
epochs=1,
batch_size=64)
model.save('citibike_DNN_model.h5') # save weights
print(">>> training model complete")
""""eval."""
metrics = model.evaluate(
[x_test_times, x_test_ids],
y_test) # loss(mae), mse, acc
print(">>> evaluating model complete")
print(metrics)
else:
max_val, min_val = 3911, 72 # max: 3911 min: 72
model = load_model('citibike_DNN_model.h5')
print(">>> loading model complete")
"""inference"""
INFERENCE_NUM = 100
rand_times = np.array([[randint(1, 12), randint(0, 6), randint(0, 23)] for _ in range(INFERENCE_NUM)])
rand_ids = np.array([[randint(72, 3911)] for _ in range(INFERENCE_NUM)])
refined_ids = copy.deepcopy(rand_ids)
refined_ids = fit_transform(refined_ids.reshape(-1, 1), max_val, min_val)
pred = model.predict([rand_times, refined_ids]) # month, weekday, hour | id
# TODO: if pred < 0, pred -= 1
# implementing at server.py
from pprint import pprint
pprint(np.around(np.hstack((
rand_times,
rand_ids,
pred)))) # month, weekday, hour, id, pred