forked from hoseinzadeehsan/CNNpred-Keras
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path3D_CNNpred_modification.py
384 lines (322 loc) · 16.4 KB
/
3D_CNNpred_modification.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
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
import numpy as np
import pandas as pd
from sklearn.preprocessing import scale
from os.path import join
from sklearn.metrics import accuracy_score as accuracy, f1_score, mean_absolute_error as mae
import os
from tensorflow import keras
from keras.models import Sequential, load_model
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPool2D, GlobalAveragePooling2D
from pathlib2 import Path
from keras import backend as K, callbacks
def f1(y_true, y_pred):
def recall(y_true, y_pred):
"""Recall metric.
Only computes a batch-wise average of recall.
Computes the recall, a metric for multi-label classification of
how many relevant items are selected.
"""
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))
recall = true_positives / (possible_positives + K.epsilon())
return recall
def precision(y_true, y_pred):
"""Precision metric.
Only computes a batch-wise average of precision.
Computes the precision, a metric for multi-label classification of
how many selected items are relevant.
"""
true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))
precision = true_positives / (predicted_positives + K.epsilon())
return precision
precision_pos = precision(y_true, y_pred)
recall_pos = recall(y_true, y_pred)
precision_neg = precision((K.ones_like(y_true)-y_true), (K.ones_like(y_pred)-K.clip(y_pred, 0, 1)))
recall_neg = recall((K.ones_like(y_true)-y_true), (K.ones_like(y_pred)-K.clip(y_pred, 0, 1)))
f_posit = 2*((precision_pos*recall_pos)/(precision_pos+recall_pos+K.epsilon()))
f_neg = 2 * ((precision_neg * recall_neg) / (precision_neg + recall_neg + K.epsilon()))
return f_posit / 2
def load_data(file_fir):
try:
df_raw = pd.read_csv(file_fir, parse_dates=['Date'])
df_raw.index = df_raw['Date']
except IOError:
print("IO ERROR")
return df_raw
def construct_data_warehouse(ROOT_PATH, file_names):
global number_of_stocks
global samples_in_each_stock
global number_feature
global predict_index
global order_stocks
tottal_train_data = np.empty((0,82))
tottal_train_target = np.empty((0))
tottal_test_data = np.empty((0,82))
tottal_test_target = np.empty((0))
for stock_file_name in file_names:
file_dir = os.path.join(ROOT_PATH, stock_file_name)
## Loading Data
try:
df_raw = load_data(file_dir)
except ValueError:
print("Couldn't Read {} file".format(file_dir))
number_of_stocks += 1
data = df_raw
df_name = data['Name'][0]
order_stocks.append(df_name)
del data['Name']
target = (data['Close'][predict_day:] / data['Close'][:-predict_day].values).astype(int)
data = data[:-predict_day]
target.index = data.index
# Becasue of using 200 days Moving Average as one of the features
data = data[200:]
data = data.fillna(0)
data['target'] = target
target = data['target']
del data['target']
del data['Date']
# data['Date'] = data['Date'].apply(lambda x: x.weekday())
number_feature = data.shape[1]
samples_in_each_stock = data.shape[0]
train_data = data[data.index < '2016-04-21']
train_data = scale(train_data)
if df_name == predict_index:
tottal_train_target = target[target.index < '2016-04-21']
tottal_test_target = target[target.index >= '2016-04-21']
data = pd.DataFrame(scale(data.values), columns=data.columns)
data.index = target.index
test_data = data[data.index >= '2016-04-21']
tottal_train_data = np.concatenate((tottal_train_data, train_data))
tottal_test_data = np.concatenate((tottal_test_data, test_data))
train_size = int(tottal_train_data.shape[0]/number_of_stocks)
test_size = int(tottal_test_data.shape[0] / number_of_stocks)
tottal_train_data = tottal_train_data.reshape(number_of_stocks, train_size, number_feature)
tottal_test_data = tottal_test_data.reshape(number_of_stocks, test_size, number_feature)
return tottal_train_data, tottal_test_data, tottal_train_target, tottal_test_target
def cnn_data_sequence(data, target, seque_len):
print ('sequencing data ...')
new_train = []
new_target = []
for index in range(data.shape[1] - seque_len + 1):
new_train.append(data[:, index: index + seque_len])
new_target.append(target[index + seque_len - 1])
new_train = np.array(new_train)
new_target = np.array(new_target)
return new_train, new_target
def sklearn_acc(model, test_data, test_target):
overall_results = model.predict(test_data)
print("overall_results "+ str(overall_results))
print("overall_results shape"+ str(overall_results.shape))
print("test_target "+ str(test_target))
test_pred = (overall_results > 0.5).astype(int)
acc_results = [mae(overall_results, test_target), accuracy(test_pred, test_target),
f1_score(test_pred, test_target, average='macro')]
return acc_results
def CNN(train_data, test_data, train_target, test_target):
# hisory of data in each sample
seq_len = 252
epoc = 100
drop = 0.1
# creating sample each containing #seq_len history
cnn_train_data, cnn_train_target = cnn_data_sequence(train_data, train_target, seq_len)
print("cnn_train_data "+str(cnn_train_data))
print("cnn_train_data shape "+str(cnn_train_data.shape))
print("cnn_train_target "+str(cnn_train_target))
print("cnn_train_target shape "+str(cnn_train_target.shape))
cnn_test_data, cnn_test_target = cnn_data_sequence(test_data, test_target, seq_len)
print("cnn_test_data " + str(cnn_test_data))
print("cnn_test_data shape " + str(cnn_test_data.shape))
print("cnn_test_target " + str(cnn_test_target))
print("cnn_test_target shape " + str(cnn_test_target.shape))
result = []
# Running CNNpred several times
for i in range(1,2):
K.clear_session()
print ('i: ', i)
my_file = Path( join(Base_dir, '3D-models/{}/model/{}-{}-{}-{}-{}.h5'.format(predict_index, epoc, seq_len, number_filter, drop, i)))
filepath = join(Base_dir, '3D-models/{}/model/{}-{}-{}-{}-{}.h5'.format(predict_index, epoc, seq_len, number_filter, drop, i))
# If the trained model doesn't exit, it is trained
if my_file.is_file():
print('loading model')
else:
print('fitting model')
model = Sequential()
# n_feature_maps = 32
# nb_classes=2
# input_layer = keras.layers.Input((seq_len, number_feature) )
# print("Input shape 1" + str((number_of_stocks,seq_len, number_feature)))
# conv_x = keras.layers.Conv1D(filters=n_feature_maps, kernel_size=8, padding='same')(input_layer)
# print("conv_x "+str(conv_x.shape))
# conv_x = keras.layers.BatchNormalization()(conv_x)
# print("conv_x " + str(conv_x.shape))
# conv_x = keras.layers.Activation('relu')(conv_x)
# print("conv_x " + str(conv_x.shape))
#
# conv_y = keras.layers.Conv1D(filters=n_feature_maps, kernel_size=5, padding='same')(conv_x)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.BatchNormalization()(conv_y)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.Activation('relu')(conv_y)
# print("conv_y " + str(conv_y.shape))
#
# conv_z = keras.layers.Conv1D(filters=n_feature_maps, kernel_size=3, padding='same')(conv_y)
# print("conv_z " + str(conv_z.shape))
# conv_z = keras.layers.BatchNormalization()(conv_z)
# print("conv_z " + str(conv_z.shape))
#
# # expand channels for the sum
# shortcut_y = keras.layers.Conv1D(filters=n_feature_maps, kernel_size=1, padding='same')(input_layer)
# print("shortcut_y " + str(shortcut_y.shape))
# shortcut_y = keras.layers.BatchNormalization()(shortcut_y)
# print("shortcut_y " + str(shortcut_y.shape))
#
# output_block_1 = keras.layers.add([shortcut_y, conv_z])
# print("output_block_1 " + str(output_block_1.shape))
# output_block_1 = keras.layers.Activation('relu')(output_block_1)
# print("output_block_1 " + str(output_block_1.shape))
#
# # BLOCK 2
#
# print("Input shape 1" + str((number_of_stocks,seq_len, number_feature)))
# conv_x = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=8, padding='same')(output_block_1)
# print("conv_x " + str(conv_x.shape))
# conv_x = keras.layers.BatchNormalization()(conv_x)
# print("conv_x "+str(conv_x.shape))
# conv_x = keras.layers.Activation('relu')(conv_x)
# print("conv_x " + str(conv_x.shape))
#
# conv_y = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=5, padding='same')(conv_x)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.BatchNormalization()(conv_y)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.Activation('relu')(conv_y)
# print("conv_y " + str(conv_y.shape))
#
#
# conv_z = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=3, padding='same')(conv_y)
# print("conv_z " + str(conv_z.shape))
# conv_z = keras.layers.BatchNormalization()(conv_z)
# print("conv_z " + str(conv_z.shape))
#
# # expand channels for the sum
# shortcut_y = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=1, padding='same')(output_block_1)
# print("shortcut_y " + str(shortcut_y.shape))
# shortcut_y = keras.layers.BatchNormalization()(shortcut_y)
# print("shortcut_y " + str(shortcut_y.shape))
#
# output_block_2 = keras.layers.add([shortcut_y, conv_z])
# print("output_block_2 " + str(output_block_2.shape))
# output_block_2 = keras.layers.Activation('relu')(output_block_2)
# print("output_block_2 " + str(output_block_2.shape))
#
# # BLOCK 3
#
# conv_x = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=8, padding='same')(output_block_2)
# print("conv_x " + str(conv_x.shape))
# conv_x = keras.layers.BatchNormalization()(conv_x)
# print("conv_x " + str(conv_x.shape))
# conv_x = keras.layers.Activation('relu')(conv_x)
# print("conv_x " + str(conv_x.shape))
#
# conv_y = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=5, padding='same')(conv_x)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.BatchNormalization()(conv_y)
# print("conv_y " + str(conv_y.shape))
# conv_y = keras.layers.Activation('relu')(conv_y)
# print("conv_y " + str(conv_y.shape))
#
# conv_z = keras.layers.Conv1D(filters=n_feature_maps * 2, kernel_size=3, padding='same')(conv_y)
# print("conv_z " + str(conv_z.shape))
# conv_z = keras.layers.BatchNormalization()(conv_z)
# print("conv_z " + str(conv_z.shape))
#
# # no need to expand channels because they are equal
# shortcut_y = keras.layers.BatchNormalization()(output_block_2)
# print("shortcut_y " + str(shortcut_y.shape))
#
# output_block_3 = keras.layers.add([shortcut_y, conv_z])
# print("output_block_3 " + str(output_block_3.shape))
# output_block_3 = keras.layers.Activation('relu')(output_block_3)
# print("output_block_3 " + str(output_block_3.shape))
#
# # FINAL
#
# gap_layer = keras.layers.GlobalAveragePooling1D()(output_block_3) #[0,:,:,:]
# print("gap_layer " + str(gap_layer.shape))
#
# output_layer = keras.layers.Dense(1, activation='sigmoid')(gap_layer)
# # output_layer = keras.layers.Dense(1, activation='sigmoid')(gap_layer)
# print("output_layer " + str(output_layer.shape))
#
# model = keras.models.Model(inputs=input_layer, outputs=output_layer)
#
# model.compile(loss='mae', optimizer=keras.optimizers.Adam(),
# metrics=['accuracy'])
#
# reduce_lr = keras.callbacks.ReduceLROnPlateau(monitor='loss', factor=0.5, patience=50, min_lr=0.0001)
#
# model_checkpoint = keras.callbacks.ModelCheckpoint(filepath=filepath, monitor='loss',
# save_best_only=True)
#
#layer 1
model.add(Conv2D(number_filter[0], (1, 1), activation='relu', input_shape=(number_of_stocks,seq_len, number_feature),
data_format='channels_last', padding="same"))
print("Conv2D shape 1" + str(model.output_shape))
# # Added layer 2
# model.add(Conv2D(number_filter[1], (1, 21), activation='relu'))
# print("Conv2D Added shape 1b" + str(model.output_shape))
# model.add(MaxPool2D(pool_size=(1, 4)))
# print("MaxPool2D Added shape 1c" + str(model.output_shape))
#layer 2
model.add(Conv2D(16, (number_of_stocks, 3), activation='relu'))
print("Conv2D shape 2" + str(model.output_shape))
# model.add(MaxPool2D(pool_size=(1, 2)))
# print("MaxPool2D shape 3" + str(model.output_shape))
#layer 3
model.add(Conv2D(number_filter[2], (1, 3), activation='relu', padding="same"))
print("Conv2D shape 4" + str(model.output_shape))
# model.add(MaxPool2D(pool_size=(1, 2)))
# print("MaxPool2D shape 5" + str(model.output_shape))
model.add(GlobalAveragePooling2D())
print("GlobalAveragePooling2D shape 5" + str(model.output_shape))
# model.add(Flatten())
# print("Flatten shape 6" + str(model.output_shape))
# model.add(Dropout(drop))
# print("Dropout shape 7" + str(model.output_shape))
model.add(Dense(1, activation='sigmoid'))
print("Dense shape 8" + str(model.output_shape))
model.compile(optimizer='Adam', loss='mae', metrics=['acc',f1])
best_model = callbacks.ModelCheckpoint(filepath, monitor='val_f1', verbose=0, save_best_only=True,
save_weights_only=False, mode='max', period=1)
model.fit(cnn_train_data, cnn_train_target, epochs=epoc, batch_size=128, verbose=0,callbacks=[best_model], validation_split=0.25)
model = load_model(filepath, custom_objects={'f1': f1})
test_pred = sklearn_acc(model,cnn_test_data, cnn_test_target)
print (test_pred)
result.append(test_pred)
print('saving results')
print("result "+str(result))
results = pd.DataFrame(result , columns=['MAE', 'Accuracy', 'F-score'])
print("results.shape " + str(results.shape))
results = results.append([results.mean(), results.max(), results.std()], ignore_index=True)
print("results.shape " + str(results.shape))
results.to_csv(join(Base_dir, '3D-models/{}/new results.csv'.format(predict_index)), index=False)
Base_dir = ''
TRAIN_ROOT_PATH = join(Base_dir, 'Dataset')
train_file_names = os.listdir(join(Base_dir, 'Dataset'))
# if moving average = 0 then we have no moving average
moving_average_day = 0
number_of_stocks = 0
number_feature = 0
samples_in_each_stock = 0
number_filter = [8,8,8]
predict_day = 1
order_stocks = []
# Name of the index that is going to be predicted
predict_index = 'DJI' # RUT, S&P, NYA, NASDAQ, DJI
print ('Loading train data ...')
train_data, test_data, train_target, test_target = construct_data_warehouse(TRAIN_ROOT_PATH, train_file_names)
print ('number of stocks = ', number_of_stocks)
print ('fitting model')
CNN(train_data, test_data, train_target, test_target)