-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathgridSearch2.py
252 lines (210 loc) · 10.8 KB
/
gridSearch2.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
# coding: utf-8
import cPickle, json, pdb, pickle, theano, sys, datetime
import copy
import numpy as np
import os.path
import theano.tensor as T
from dataset.Nikkei import Nikkei
from experiment.CompressSparseVector.SparseAutoencoder import SparseAutoencoder, train_sae
from experiment.CompressSparseVector.RBM import RBM, train_rbm
# from experiment.PredictPrices.SdA_theano import SdA, train_SdA, pretrain_SdA, finetune_SdA
from experiment.PredictPrices import SdA
from experiment.PredictPrices import SdA_RNN
from experiment.PredictPrices import DBN
from run import *
from sklearn.svm import SVR, SVC
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
from sklearn.decomposition import PCA
from sklearn.grid_search import GridSearchCV
from nikkei225 import getNikkei225
# theano.config.floatX = 'float64'
# from run import *
default_model_dir = '/home/fujikawa/StockPredict/src/deeplearning/experiment/Model'
params = {
# 'experiment_type' : 'chi2_selected',
'label_type' : 1,
'experiment_type' : 'average',
'dataset_type' : 'article',
# 'experiment_type' : 'max',
'STEP1' : {
'reg_weight' : 0.0,
'model' : 'rbm',
'n_hidden' : 1000,
'learning_rate' : 0.05,
'batch_size' : 20
},
'STEP3' : {
'brandcode' : '0101'
},
'STEP4' : {
'dropout' : False,
'recurrent' : False,
'model' : 'sda_4_dropout_recurrent',
'hidden_recurrent' : 500,
'corruption_levels' : [.3, .3, .4],
'k' : 1,
'hidden_layers_sizes' : [1000, 500, 500],
'pretrain' : {
'batch_size' : 20,
'learning_rate' : 1e-2,
'epochs' : 100
},
'finetune' : {
'batch_size' : 20,
'learning_rate' : 1e-2,
'epochs' : 100
}
}
}
### grid searchのパラメータ格納
params['STEP3']['brandcode'] = ['0101', '7203', '6758', '6502', '7201', '6501', '6702', '6753', '8058', '8031', '7751']
# params['STEP3']['brandcode'] = ['0101']
# params['STEP3']['brandcode'] = getNikkei225()
ALL_brandcodes = getNikkei225()
NG_brandcodes = ['2768', '3382', '3893', '4188', '4324', '4568', '4689', '4704', '5411', '6674', '8303', '8306', '8308', '8309', '8316', '8411', '8766', '8795', '9983', '6796', '9984', '6366', '2282', '7004', '7013', '9020', '9432']
brandcodes = list(set(ALL_brandcodes) - set(NG_brandcodes))
params['STEP3']['brandcode'] = brandcodes
params['STEP4']['hidden_layers_sizes'] = [
[1000]
# [500],
# [1000, 500]
# [params['STEP1']['n_hidden'] , params['STEP1']['n_hidden'], params['STEP1']['n_hidden'] / 2]
]
params['STEP4']['pretrain'] = {
'batch_size' : [100, 50],
'epochs' : [300]
}
params['STEP4']['finetune'] = {
'batch_size' : [30],
'learning_rate' : [1e-1, 1e-2, 5e-2, 1e-3],
'epochs' : [100]
}
if params['experiment_type'] == 'chi2_selected':
model_dirs['STEP4_logs'] = default_model_dir + '/STEP4_logs/baseline_chi2_selected'
if __name__ == '__main__':
### 銘柄数種について実験
if len(sys.argv) < 6:
print len(sys.argv)
print 'args invalid'
sys.exit()
else:
params['STEP4']['model'] = '%s_%s_%s_%s' % (sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4])
params['label_type'] = int(sys.argv[2])
if sys.argv[3] == 'dropout':
params['STEP4']['dropout'] = True
if sys.argv[4] == 'recurrent':
params['STEP4']['recurrent'] = True
params['experiment_type'] = sys.argv[5]
model_dirs = {
'STEP1' : '%s/%s/h%d_lr%s_s%s_b%s_%s.%s' % (default_model_dir, 'STEP1', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['dataset_type'], params['STEP1']['model']),
'STEP2' : '%s/%s/h%d_lr%s_s%s_b%s_%s.%s' % (default_model_dir, 'STEP2', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['dataset_type'], params['STEP1']['model']),
'STEP3' : '%s/%s/h%d_lr%s_s%s_b%s_%s.%s' % (default_model_dir, 'STEP3', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['dataset_type'], params['STEP1']['model']),
'STEP3_logs' : '%s/%s/%s_h%d_lr%s_s%s_b%s_%s_%s_%s.csv' % (default_model_dir, 'STEP3_logs', params['STEP4']['model'], params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['dataset_type'], params['STEP1']['model'], params['experiment_type']),
}
# model_dirs = {
# 'STEP1' : '%s/%s/h%d_lr%s_s%s_b%s.%s' % (default_model_dir, 'STEP1', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['STEP1']['model']),
# 'STEP2' : '%s/%s/h%d_lr%s_s%s_b%s.%s' % (default_model_dir, 'STEP2', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['STEP1']['model']),
# 'STEP3' : '%s/%s/h%d_lr%s_s%s_b%s.%s' % (default_model_dir, 'STEP3', params['STEP1']['n_hidden'], str(params['STEP1']['learning_rate']), str(params['STEP1']['reg_weight']), str(params['STEP1']['batch_size']), params['STEP1']['model']),
# 'STEP3_logs' : '%s/%s/%s_%s_idf.csv' % (default_model_dir, 'STEP3_logs', params['STEP4']['model'], params['experiment_type']),
# }
print model_dirs['STEP3_logs']
if os.path.exists(model_dirs['STEP3_logs']):
print 'ファイルが存在します．'
# sys.exit()
else:
out = open(model_dirs['STEP3_logs'], 'w')
out.write('brandcode,train_acc,test_acc\n')
out.close()
# model_dirs = reload_model_dirs()
if params['STEP4']['model'].split('_')[0] == 'sda':
model = SdA
params['STEP4']['pretrain']['learning_rate'] = [1e-5]
elif params['STEP4']['model'].split('_')[0] == 'dbn':
model = DBN
params['STEP4']['pretrain']['learning_rate'] = [1e-5]
elif params['STEP4']['model'].split('_')[0] == 'svr':
model = SVR
elif params['STEP4']['model'].split('_')[0] == 'svc':
model = SVC
else:
sys.exit()
i = 0
label_type = params['label_type']
if params['experiment_type'] == 'baseline':
print 'start to load baseline dataset...'
dataset = cPickle.load(open(default_model_dir + '/STEP2/baseline_original'))
print 'start to unify stockprice...'
dataset.unify_stockprices(dataset=dataset.baseline, brandcodes=params['STEP3']['brandcode'], dataset_type=params['experiment_type'], label_type=label_type, y_type=get_y_type(label_type), y_force_list=params['STEP4']['recurrent'])
else:
print 'start to load proposed dataset...'
dataset = cPickle.load(open(model_dirs['STEP2']))
if params['experiment_type'] == 'average':
print 'start to unify stockprice (average pooling)...'
usedata = dataset.unified_mean
else:
print 'start to unify stockprice (max pooling)...'
usedata = dataset.unified_max
dataset.unify_stockprices(dataset=usedata, brandcodes=params['STEP3']['brandcode'], dataset_type=params['experiment_type'], label_type=label_type, y_type=get_y_type(label_type), y_force_list=params['STEP4']['recurrent'])
if params['experiment_type'] != 'baseline':
reguralize_data(dataset, params['STEP3']['brandcode'])
optimizeGPU(dataset, params['STEP3']['brandcode'])
# if params['experiment_type'] == 'chi2_selected':
# dataset = Nikkei(dataset_type=params['experiment_type'], brandcode=brandcode)
# dataset.unify_stockprices(dataset=dataset.raw_data[brandcode], brandcode=brandcode, dataset_type=params['experiment_type'])
# else:
# dataset = ""
# dataset = cPickle.load(open(model_dirs['STEP2']))
# dataset.unify_stockprices(dataset=dataset.unified, brandcodes=params['STEP3']['brandcode'], dataset_type=params['experiment_type'], label_type=params['label_type'])
# reguralize_data(dataset, params['STEP3']['brandcode'])
def transformY(data_y):
y = []
for data in data_y:
y.append(data[0])
return np.array(y)
train_x = dataset.phase2['train']['x']
train_x = np.append(train_x, dataset.phase2['valid']['x'], 0)
test_x = dataset.phase2['test']['x']
train_x_original = train_x
test_x_original = test_x
startbrand = '9007'
flag = True
for i, brandcode in enumerate(params['STEP3']['brandcode']):
## y に 各銘柄の正解データを格納
change_brand(dataset, brandcode)
print brandcode
if flag:
if get_y_type(label_type) == 0:
## 回帰問題の場合
train_y = transformY(dataset.phase2['train']['y'])
train_y = np.append(train_y, transformY(dataset.phase2['valid']['y']), 0)
test_y = transformY(dataset.phase2['test']['y'])
else:
## 分類問題の場合
train_y = dataset.phase2['train']['y']
train_y = np.append(train_y, dataset.phase2['valid']['y'], 0)
test_y = dataset.phase2['test']['y']
#### 各種分類アルゴリズムの詳細設定 ####
if model == SVR:
tuned_parameters = [{'kernel': ['rbf', 'linear'], 'gamma': [10**i for i in range(-4,0)], 'C': [10**i for i in range(0,4)]}]
gscv = GridSearchCV(model(), tuned_parameters, cv=5, scoring="mean_squared_error", n_jobs=5)
gscv.fit(train_x, train_y)
best_model = gscv.best_estimator_
elif model == SVC:
tuned_parameters = [{'kernel': ['rbf', 'linear'], 'gamma': [10**i for i in range(-4,0)], 'C': [10**i for i in range(0,4)]}]
gscv = GridSearchCV(model(), tuned_parameters, cv=5, n_jobs=5)
gscv.fit(train_x, train_y)
best_model = gscv.best_estimator_
else:
best_model = model()
best_model.fit(train_x, train_y)
predict_y = best_model.predict(test_x)
result_train = (best_model.predict(train_x) == train_y).sum()
result_test = (best_model.predict(test_x) == test_y).sum()
out = open(model_dirs['STEP3_logs'], 'a')
train_acc = float(result_train) / len(train_y)
test_acc = float(result_test) / len(test_y)
out.write('%s,%f,%f\n' % (brandcode, train_acc, test_acc))
# pdb.set_trace()
out.close()
if brandcode == startbrand:
flag = True