-
Notifications
You must be signed in to change notification settings - Fork 2
/
dataset.py
279 lines (239 loc) · 11.1 KB
/
dataset.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
276
277
278
279
import csv
import time
from IPython import embed
import datetime
import os
import torch
import random
from tqdm import tqdm
from torch.utils.data import DataLoader
from torch.utils.data.sampler import Sampler, BatchSampler
import numpy as np
from collections import defaultdict
import itertools
import math
# c : closing price
# h: highest price
# l: lowest price
# o: open price
# v: volume
# data: pre 3 days, history: 20 days now
def get_loader_daily(dataset, batch_size, train_mode):
instances = []
for d, history_set in tqdm(dataset):
history_list = list(history_set)
history_list.sort() # sort according to date （day)
num_days = len(history_list)
for i in range(20, num_days):
if train_mode:
if (history_list[i].year < 2017):
continue
else:
if (history_list[i].year == 2018) and (history_list[i].month > 6):
continue
# tensor
tensor = torch.FloatTensor(20, 5) # last 20 days c, h, l, o, v
tensor[:, 0] = -1 # c
tensor[:, 1] = -1e10 # h
tensor[:, 2] = 1e10 # l
tensor[:, 3] = -1 # o
tensor[:, 4] = 0 # v
for t in range(20): # last 20 days c h l o v
index = t
for hour in range(9, 15):
for minute in range(0, 65, 5):
z = history_list[i-20+t] + datetime.timedelta(hours=hour, minutes=minute)
if z in d:
tensor[index, 2] = min(tensor[index, 2].item(), d[z][2]) # l
tensor[index, 1] = max(tensor[index, 1].item(), d[z][1]) # h
tensor[index, 4] += d[z][4] # v
open_time = history_list[i-20+t] + datetime.timedelta(hours=9) # 9:00 as opening time
close_time = history_list[i-20+t] + datetime.timedelta(hours=15) # 15:00 as close time
if open_time in d :
tensor[index, 3] = d[open_time][3] # c
if close_time in d:
tensor[index, 0] = d[close_time][0] # o
# v
v = torch.FloatTensor([0])
# for i-th day
for hour in range(9, 15):
for minute in range(0, 60, 5):
time_stamp = history_list[i] + datetime.timedelta(hours=hour, minutes=minute)
if time_stamp in d:
v += d[time_stamp][4] # v
ticker_id = d[time_stamp][5] # ticker id
closing_time = history_list[i] + datetime.timedelta(hours=15) # 15:00 as close time
if closing_time in d:
v += d[closing_time][4] # v
else:
continue # discard data
if 1e9 > tensor.min().item() > -0.5: # data validation
instances.append((ticker_id, tensor, v))
if train_mode:
random.shuffle(instances)
l = len(instances)
trainloader = DataLoader(instances[: int(l * 0.75)], batch_size=batch_size)
devloader = DataLoader(instances[int(l * 0.75):], batch_size=batch_size)
return trainloader, devloader
else: # Test loader
random.shuffle(instances)
l = len(instances)
testloader = DataLoader(instances, batch_size=batch_size)
return testloader
def get_loader_hourly(dataset, batch_size, train_mode):
instances = []
for d, history_set in tqdm(dataset):
history_list = list(history_set)
history_list.sort()
num_days = len(history_list)
for i in range(20, num_days):
if train_mode:
if (history_list[i].year < 2017):
continue
else:
if (history_list[i].year == 2018) and (history_list[i].month > 6):
continue
# tensor
tensor = torch.FloatTensor(12, 5)
tensor[:, 0] = -1 # c
tensor[:, 1] = -1e10 # h
tensor[:, 2] = 1e10 # l
tensor[:, 3] = -1 # o
tensor[:, 4] = 0 # v
# last two days hourly data
for t in range(2):
for hour in range(9, 15):
index = t * 6 + hour - 9
y = history_list[i-2+t] + datetime.timedelta(hours=hour)
if y.hour == 12: #
if y + datetime.timedelta(minutes=30) in d: # 12.30 下午开盘
tensor[index, 3] = d[y + datetime.timedelta(minutes=30)][3] # o
else:
if y in d:
tensor[index, 3] = d[y][3] # o 开盘价
if y.hour == 11:
if y + datetime.timedelta(minutes=30) in d: # 11:30 上午收盘
tensor[index, 0] = d[y + datetime.timedelta(minutes=30)][0] # c
else:
if y + datetime.timedelta(minutes=55) in d:
tensor[index, 0] = d[y + datetime.timedelta(minutes=55)][0] # c 每个小时的收盘价
for d_min in range(0, 60, 5):
z = y + datetime.timedelta(minutes=d_min)
if z in d:
tensor[index, 2] = min(tensor[index, 2].item(), d[z][2]) # l
tensor[index, 1] = max(tensor[index, 1].item(), d[z][1]) # h
tensor[index, 4] = tensor[index, 4] + d[z][4] # v
# history: last 20 days c, h, l, o, v of 9:00 - 9:55
history = torch.FloatTensor(20, 5)
history[ :, 0] = -1 # c
history[ :, 1] = -1e10 # h
history[ :, 2] = 1e10 # l
history[ :, 3] = -1 # o
history[ :, 4] = 0 # v
x = history_list[i] + datetime.timedelta(hours=9) # 9:00
for t in range(20):
y = history_list[i-20+t] + datetime.timedelta(hours=9) # last 50 days
z = y + datetime.timedelta(minutes=55) # 9:55
if y in d:
history[t, 3] = d[y][3] # o
if z in d:
history[t, 0] = d[z][0] # c
for d_min in range(0, 60, 5): # 9:00 - 9:55 : v
z = y + datetime.timedelta(minutes=d_min)
if z in d:
history[t, 2] = min(history[t, 2].item(), d[z][2]) # l
history[t, 1] = max(history[t, 1].item(), d[z][1]) # h
history[t, 4] = history[t, 4] + d[z][4] # v
# v
v = torch.FloatTensor([0])
x = history_list[i] + datetime.timedelta(hours=9)
for minute in range(0, 65, 5):
y = x + datetime.timedelta(minutes=minute)
if y in d:
v += d[y][4] # v
ticker_id = d[y][5] # ticker id
if (1e9 > history.min().item() > -0.5) and (1e9 > tensor.min().item() > -0.5):
instances.append((ticker_id, tensor, history, v))
if train_mode:
random.shuffle(instances)
l = len(instances)
trainloader = DataLoader(instances[: int(l * 0.75)], batch_size=batch_size)
devloader = DataLoader(instances[int(l * 0.75):], batch_size=batch_size)
return trainloader, devloader
else: # Test loader
random.shuffle(instances)
l = len(instances)
testloader = DataLoader(instances, batch_size=batch_size)
return testloader
def read_csv():
if not os.path.exists('raw_data.pt'):
train_dataset, test_dataset = [], []
for path in tqdm(os.listdir('topix500/')):
train_data, train_history, test_data, test_history = {}, set(), {}, set()
with open('topix500/'+path, 'r') as f:
reader = csv.reader(f)
for row in reader:
if row[0] == '':
continue
time_stamp = datetime.datetime.strptime(row[2] + ' ' + row[6], "%Y-%m-%d %H:%M")
chlov = [float(row[1]), float(row[3]), float(row[4]), float(row[5]), float(row[7])]
if time_stamp.year == 2018:
test_data[time_stamp] = chlov
test_history.add(datetime.datetime.strptime(row[2], "%Y-%m-%d"))
else:
train_data[time_stamp] = chlov
train_history.add(datetime.datetime.strptime(row[2], "%Y-%m-%d"))
train_dataset.append((train_data, train_history))
test_dataset.append((test_data, test_history))
torch.save((train_dataset, test_dataset), 'raw_data.pt')
else:
train_dataset, test_dataset = torch.load('raw_data.pt')
return train_dataset, test_dataset
def prepare_dataset(batch_size, data_name):
if not os.path.exists("%s-cor.pt" % (data_name)):
print('loading raw data')
t0 = time.time()
train_dataset, test_dataset = read_csv()
print("loading raw data cost: ", time.time() - t0, "s")
print('raw data loaded. processing raw data...')
if data_name == 'hourly':
trainloader, devloader = get_loader_hourly(train_dataset, batch_size, train_mode=True)
testloader = get_loader_hourly(test_dataset, batch_size, train_mode=False)
elif data_name == 'daily':
trainloader, devloader = get_loader_daily(train_dataset, batch_size, train_mode=True)
testloader = get_loader_daily(test_dataset, batch_size, train_mode=False)
else:
assert(False)
torch.save((trainloader, devloader, testloader), '%s-cor.pt' % data_name)
else:
print('loading from saved data pt file')
trainloader, devloader, testloader = torch.load("%s-cor.pt" % (data_name))
return trainloader, devloader, testloader
def test_EMA_daily(device, testloader):
MSE, MAE, correct, cnt = 0, 0, 0, 0
k = 0.04
with torch.no_grad():
for _, chlov, v in testloader:
history = chlov
chlov, history, v = chlov.to(device), history.to(device), v.to(device)
chlov, history, v = torch.log(chlov+1), torch.log(history+1), torch.log(v+1)
output = history[:, 0, -1].exp().clone()
for i in range(1, 20):
output = history[:, i, -1].exp() * k + output * (1-k)
output = output.log().view(-1, 1)
MSE += ((output - v) ** 2).mean().item()
MAE += ((output - v).abs()).mean().item()
correct += ((output - chlov[:, -1, -1:]) * (v - chlov[:, -1, -1:])).ge(0).float().mean().item()
cnt += 1
MSE /= cnt
MAE /= cnt
correct /= cnt
RMSE = math.sqrt(MSE)
print('Test EMA: MSE: {:.6f}, RMSE: {:.6f}, MAE: {:.6f}, ACC: {:.6f} '.format(MSE, RMSE, MAE, correct))
if __name__ == "__main__":
train, dev, test = prepare_dataset(32, "daily")
print(len(train.dataset))
print(len(dev.dataset))
print(len(train))
print(len(test.dataset))
print(len(train))