-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_lstm.py
206 lines (176 loc) · 6.63 KB
/
train_lstm.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
import json
import numpy as np
import dateutil.parser as dp
import argparse
ap = argparse.ArgumentParser()
ap.add_argument("-d", "--debug", default=0, help="Show debug message 0:no message, 1:show predict, 2:show train+predict, 3: show plot")
ap.add_argument("-r", "--repeat", default=1, help="training times")
ap.add_argument("-s", "--save_config", action='store_true', help="save plot to file")
args = vars(ap.parse_args())
DEBUG = int(args['debug'])
code = '0050'
TRAIN_TEST = 0.8
# code = '2317'
# TRAIN_TEST = 0.8
name = 'tw{0}'.format(code)
FIELDS = []
# FIELDS.append('date')
# FIELDS.append('low')
# FIELDS.append('high')
# FIELDS.append('open')
FIELDS.append('close')
# FIELDS.append('change')
# FIELDS.append('transaction')
# FIELDS.append('turnover')
# FIELDS.append('capacity')
BLOCK_SIZE = 5
BATCH_SIZE = 64
PREDICT_PERIOD = 1
FIELD_LEN=len(FIELDS)
EPOCHS = 200
NEURONS_1 = 10
NEURONS_2 = 20
SHUFFLE_ALL = True
def json2data(node):
data=[]
for k in FIELDS:
data.append(node[k])
return data
def getTimestamp(date):
return dp.parse(date)
def moving_average(values, index, q):
t = values[index-q+1, index+1]
return np.average(t)
# load data
def loadData():
data=[]
with open('./data/{0}.json'.format(name)) as data_file:
data = json.loads(data_file.read())
col = FIELD_LEN * BLOCK_SIZE
row = len(data)- BLOCK_SIZE - PREDICT_PERIOD + 1
ub = len(data) - PREDICT_PERIOD - 1
all_data= np.zeros((row, col+1))
x_data = np.zeros((row, col))
y_data = np.zeros(row)
for i, d in enumerate(data):
if i < BLOCK_SIZE-1:
continue
if i > ub:
break
idx = i-BLOCK_SIZE+1
temp=[]
for pre in range(BLOCK_SIZE-1, -1, -1):
temp += json2data(data[i-pre])
# x_data[i-BLOCK_SIZE] = temp
all_data[idx, :col] = temp
try:
# y_data[i-BLOCK_SIZE] = data[i+PREDICT_PERIOD]['close']
all_data[idx, col] = data[i+PREDICT_PERIOD]['close']
except IndexError as e:
print('i={0}, BLOCK_SIZE={1}, PREDICT_PERIOD={2}'.format(i, BLOCK_SIZE, PREDICT_PERIOD))
all_data = np.asarray(all_data)
if SHUFFLE_ALL:
np.random.shuffle(all_data)
[x_data, y_data] = np.split(all_data, [col], axis=1)
TRAINING = int(len(x_data) * TRAIN_TEST)
x_train_data = x_data[0:TRAINING, :]
x_test_data = x_data[TRAINING:, :]
mean = np.mean(x_train_data)
x_train_data -= mean
std = x_train_data.std()
x_train_data /= std
x_test_data -= mean
x_test_data /= std
y_train_data = y_data[0:TRAINING]
y_test_data = y_data[TRAINING:]
x_train_data= np.reshape(x_train_data, (x_train_data.shape[0], 1, x_train_data.shape[1]))
x_test_data = np.reshape(x_test_data, (x_test_data.shape[0], 1, x_test_data.shape[1]))
# y_train_data = np.reshape(y_train_data, (y_train_data.shape[0], 1, 1))
# y_test_data = np.reshape(y_test_data, (y_test_data.shape[0], 1, 1))
return (x_train_data, y_train_data), (x_test_data, y_test_data)
from keras import models, layers, optimizers
import tensorflow as tf
import keras.backend.tensorflow_backend as ktf
config = tf.ConfigProto(inter_op_parallelism_threads = 4, intra_op_parallelism_threads = 4)
sess = tf.Session(config = config)
ktf.set_session(sess)
def build_lstm(input_shape):
model = models.Sequential()
# model.add(layers.GRU(512, return_sequences=True, kernel_initializer='Orthogonal', name='gru1', input_shape=input_shape))
model.add(layers.LSTM(NEURONS_1, input_shape=input_shape, return_sequences=True, name='lstm1', activation='tanh', use_bias=True))
# model.add(layers.Dropout(0.2, name='drop1'))
model.add(layers.LSTM(NEURONS_2,return_sequences=False, name='lstm2', activation='tanh', use_bias=True))
# model.add(layers.Dropout(0.2, name='drop2'))
model.add(layers.Dense(1, name='dense'))
model.add(layers.Activation('linear'))
rmsprop = optimizers.RMSprop(decay=0.0001)
model.compile(optimizer=rmsprop, loss='mse', metrics=['mae'])
return model
def train(model, x_train_data, y_train_data, x_test_data, y_test_data):
verbose = 0
if DEBUG >=2:
model.summary()
verbose = 1
history = model.fit(x_train_data, y_train_data, epochs=EPOCHS, batch_size=BATCH_SIZE, validation_split=0.05, verbose=verbose, shuffle=True)
import utils
if DEBUG >=3:
if args['save_config']:
utils.plot(history.history, 'train.png')
else:
utils.plot(history.history)
model.reset_states()
prediction = model.predict(x_test_data)
log=False
if DEBUG >=1:
log=True
under, over, avg_err = utils.predict_result(prediction, y_test_data, log)
# print('{0}>{1}, {2}<{3}, {4:.3f}<0.02'.format(under, len(prediction)/2, over, len(prediction)/5, abs(avg_err)))
err = float(np.max(y_test_data))*0.006
total = len(y_test_data)
result={
'under': under,
'under_r': under*100/total,
'over': over,
'over_r': over*100/total,
'avg_err': avg_err
}
if result['under_r'] > 60 and result['over_r'] < 20 and abs(avg_err) < err:
model.save('{0}_lstm_b{1}p{2}_{3:.1f}_{4:.1f}_{5:.3f}.h5'.format(name, BLOCK_SIZE, PREDICT_PERIOD, result['under_r'], result['over_r'], avg_err))
if DEBUG >=3 :
if(args['save_config']):
utils.plot_predict(prediction, y_test_data, 'predict.png')
else:
utils.plot_predict(prediction, y_test_data)
return result
(x_train_data, y_train_data), (x_test_data, y_test_data) = loadData()
# tune hyperparameter manually
model = build_lstm((x_train_data[0].shape))
repeat = int(args['repeat'])
from colorama import Fore
import time
for i in range(repeat):
t1 = time.time()
result = train(model, x_train_data, y_train_data, x_test_data, y_test_data)
t2 = time.time()
if result['avg_err'] < 0.5:
color = Fore.GREEN
else:
color = Fore.WHITE
print('{0}train {1}: under={2:.2f}%({3}), over={4:.2f}%({5}), avg_err={6:.3f}, time={7:.2f}'.format(color, i, result['under_r'], result['under'], result['over_r'], result['over'], result['avg_err'], t2-t1))
print(Fore.WHITE)
#tune by scikit-learn
# from sklearn.model_selection import GridSearchCV
# from keras.wrappers.scikit_learn import KerasRegressor
# model = KerasRegressor(build_fn=build_lstm, verbose=0, input_shape=(x_train_data.shape[1],))
# batch_size = [4, 16, 32, 64, 128, 256]
# epochs = [50, 100, 200]
# param_grid = dict(batch_size=batch_size, epochs=epochs)
# grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs = -1)
# grid_result = grid.fit(x_train_data, y_train_data)
# # summarize results
# print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
# means = grid_result.cv_results_['mean_test_score']
# stds = grid_result.cv_results_['std_test_score']
# params = grid_result.cv_results_['params']
# for mean, stdev, param in zip(means, stds, params):
# print("%f (%f) with: %r" % (mean, stdev, param))