/
tiny_pipeline.py
267 lines (189 loc) · 9.17 KB
/
tiny_pipeline.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
import yfinance
from ta_indicators import get_ta
import pandas as pd
from hmmlearn.hmm import GaussianHMM
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from mlxtend.feature_extraction import PrincipalComponentAnalysis
from sklearn.ensemble import ExtraTreesRegressor
from sklearn.pipeline import make_pipeline
from sklearn.svm import SVR
from mlxtend.evaluate import feature_importance_permutation
import matplotlib.pyplot as plt
from itertools import combinations
from random import shuffle
import numpy as np
from itertools import product
from multiprocessing import Pool, cpu_count, Queue, Process
from time import sleep
import sqlite3
from joblib import dump, load
import namegenerator
import matplotlib
from simple_trader_hmm import trader
conn = sqlite3.connect('tiny_pipeline.db')
class pipeline():
def __init__(self, train, test, features, model_name = None):
self.train = train
self.test = test
self.features = list(features)
if model_name is None:
self.get_model()
if self.rename_passed == False:
return
self.predict_new()
if not self.results[self.results['test_count']<25].empty:
return
if len(self.results.dropna())<3:
return
corr_1 = self.results[['train_mean','test_mean']].corr().values[0][1]
corr_2 = self.results[['train_mean','test_next_mean']].corr().values[0][1]
corr_3 = self.results[['test_mean','test_next_mean']].corr().values[0][1]
# ensure at least one mean and var fits our needs
best_state = self.results[self.results['test_mean']==self.results['test_mean'].max()]
if float(best_state['test_mean'])<.25 or float(best_state['test_var'])>4.0:
return
if corr_1<0.6 or corr_2<0.6 or corr_3<0.6:
return
self.name = namegenerator.gen()
self.results['corr_1'] = corr_1
self.results['corr_2'] = corr_2
self.results['corr_3'] = corr_3
self.results['features'] = str(self.features)
self.results['name'] = self.name
self.results['safe_return'] = trader('QQQ', 'QLD', data = self.test).return_percentage
self.results['moderate_return'] = trader('QQQ', 'TQQQ', data = self.test).return_percentage
self.results['extreme_return'] = trader('QLD', 'TQQQ', data = self.test).return_percentage
self.results.to_sql('models', conn, if_exists='append')
#self.test[['date','close','state','state_name']].to_csv('%s.csv' % self.name)
dump(self.pipe_pca, './models/%s.joblib' % self.name)
self.plot()
print(self.features)
print(self.results)
elif model_name is not None:
self.name = model_name
self.pipe_pca = load('./models/%s.joblib' % self.name)
self.get_model()
self.predict_new()
self.results['extreme_return'] = trader('QLD', 'TQQQ', data = self.test).return_percentage
print(self.test)
self.plot(show=True)
def apply_rename_states(self):
for index, group in self.results.iterrows():
self.test.loc[self.test['state']==index, 'state_name'] = group['state_name']
self.test.loc[self.test['state']==index, 'state_num'] = group['state_num']
print(self.results)
# todo: renaming states is not working!!!
print(self.test[ ['date','state'] ])
self.test['state'] = self.test['state_num']
print(self.test[ ['date','state','state_num','state_name','close'] ])
def plot(self, show=False):
self.test.plot.scatter(x='date',
y='close',
c='state',
colormap='viridis')
fig = matplotlib.pyplot.gcf()
fig.set_size_inches(18.5, 10.5, forward=True)
file_name = self.name
print('plotting', file_name)
if show==False:
plt.savefig('./plots_tiny_pipeline/%s.png'% file_name )
else:
plt.show()
def get_model(self):
self.pipe_pca = make_pipeline(StandardScaler(),
PrincipalComponentAnalysis(n_components=3),
GaussianHMM(n_components=3, covariance_type='full', random_state=7))
self.pipe_pca.fit(self.train[ ['return'] + self.features ])
model = self.pipe_pca.steps[2][1]
results = []
for i in range(3):
result = [i, model.means_[i][0], np.diag(model.covars_[i])[0]]
results.append(result)
results = pd.DataFrame(results)
results.columns = ['state', 'train_mean', 'train_var']
self.results = results.set_index('state')
self.get_renamed_states()
def get_renamed_states(self):
self.rename_passed = True
self.results = self.results.sort_values(by=['train_var'])
self.results['state_name'] = None
self.results['state_num'] = None
if len(self.results.loc[self.results['train_mean']<0])>1:
self.rename_passed = False
return
self.results.loc[self.results['train_mean']<0, 'state_name'] = 'sell'
self.results.loc[self.results['train_mean']<0, 'state_num'] = 0
# select the remaining groups
groups = self.results[pd.isnull(self.results).any(axis=1)].sort_values(by=['train_var'])
# assing the one with the lowest variation
self.results.loc[self.results.index == int(groups.head(1).index.values[0]), 'state_name'] = 'strong_buy'
self.results.loc[self.results.index == int(groups.head(1).index.values[0]), 'state_num'] = 2
self.results.loc[pd.isnull(self.results).any(axis=1), 'state_name'] = 'buy'
self.results.loc[pd.isnull(self.results).any(axis=1), 'state_num'] = 1
#print(self.results)
self.results = self.results.dropna()
#print(self.results)
if len(self.results)<3:
self.rename_passed = False
def predict_new(self):
self.test['state'] = self.pipe_pca.predict(self.test[ ['return'] + self.features ])
self.apply_rename_states()
for state, group in self.test.groupby(by='state'):
self.results.loc[state, 'test_mean'] = group['return'].mean()
self.results.loc[state, 'test_var'] = group['return'].std()
self.results.loc[state, 'test_count'] = group['return'].count()
self.results.loc[state, 'test_next_mean'] = group['next_return'].mean()
self.results.loc[state, 'test_next_var'] = group['next_return'].std()
def run_decision_tree(train, test_cols):
# get features
clf = ExtraTreesRegressor(n_estimators=150)
clf = clf.fit(train[test_cols], train['next_return'])
df = pd.DataFrame([test_cols, clf.feature_importances_]).T
df.columns = ['feature', 'importances']
df = df.sort_values(by='importances').tail(50)
starting_features = list(df['feature'].values)
return starting_features
def get_data(symbol):
history = yfinance.Ticker(symbol).history(period='7y').reset_index()
history = get_ta(history, volume=True, pattern=False)
history.columns = map(str.lower, history.columns)
history['return'] = history['close'].pct_change() * 100
history = history.dropna()
history['next_return'] = history['return'].shift(-1)
num_rows = len(history)
train = history.head( int(num_rows * .75) )
test = history.tail( int(num_rows *.25) )
test_cols = train.columns.drop(['date','return', 'next_return'])
return train, test, test_cols
def pipeline_runner(input_queue):
while input_queue.qsize():
train, test, features = input_queue.get()
pipeline(train, test, features)
def queue_monitor(input_queue):
while input_queue.qsize():
start_size = input_queue.qsize()
sleep(60)
end_size = input_queue.qsize()
diff = start_size - end_size
print('====================')
print('queue monitor')
print(diff, round(end_size / diff, 2))
print('====================')
if __name__ == '__main__':
input_queue = Queue()
train, test, test_cols = get_data('QLD')
starting_features = run_decision_tree(train, test_cols)
#features_from_previous()
feature_combos = list(combinations(starting_features, 5))
shuffle(feature_combos)
feature_input_list = []
for features in feature_combos:
input_queue.put( [train, test, features] )
for i in range(cpu_count()-1):
#for i in range(1):
p = Process(target=pipeline_runner, args=(input_queue,) )
p.start()
p = Process(target=queue_monitor, args=(input_queue, ))
p.start()
p.join()