-
Notifications
You must be signed in to change notification settings - Fork 0
/
models.py
406 lines (339 loc) · 16.3 KB
/
models.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
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
from sktime.forecasting.compose import NetworkPipelineForecaster
import pandas as pd
from sklearn import linear_model
from sklearn.ensemble import RandomForestRegressor, RandomForestClassifier
from sklearn import svm
from sklearn.neighbors import KNeighborsRegressor, KNeighborsClassifier
from sktime.forecasting.compose import make_reduction
from sktime.forecasting.compose._reduce import _sliding_window_transform
import numpy as np
from sktime.forecasting.base import ForecastingHorizon
from hmmlearn import hmm
from utils import converter
from abc import ABC, abstractmethod
from sktime.classification.base import BaseClassifier
from sktime.base import BaseEstimator
from sktime.forecasting.model_selection import SlidingWindowSplitter
class BasePipe(ABC):
def get_model_name(self):
return self.model_name
def get_model(self):
return self.model
@abstractmethod
def get_fh(self):
pass
@abstractmethod
def build(self):
pass
# Directional change reduced to forecasting
class RegressorPipe(BasePipe):
"""
Parameters
----------
model_name : str
name of model
regressor : object
sktime compatible regressor
window_length : int
parameter to sktime.forecasting.compose.make_reduction
"""
def __init__(self, model_name, regressor, window_length):
self.model_name = model_name
self.regressor = regressor
self.window_length = window_length
def get_fh(self, **kwargs):
return ForecastingHorizon(np.arange(1,len(kwargs['y_test'])+1))
def build(self, **kwargs):
forecaster = make_reduction(self.regressor, window_length=self.window_length, strategy="recursive")
forecaster_pipe = NetworkPipelineForecaster([
('forecaster',forecaster, {'fit':{'y': 'original_y', 'fh':'original_fh'},
'predict':{'fh':'original_fh'}
}),
('converter', converter, {'fit':None,
'predict':{'y1':kwargs['y_train'],'y2':'forecaster'}})
])
self.model = forecaster_pipe
return forecaster_pipe
########################################### REGRESSORS #######################################################
lasso_regression = RegressorPipe(model_name='lasso_regressor', regressor=linear_model.Lasso(), window_length=15)
random_forest_regressor_pipeline = RegressorPipe(model_name="random_forest_regressor",
regressor=RandomForestRegressor(),
window_length=15)
svm_regressor_pipeline = RegressorPipe(model_name='svm_regressor', regressor=svm.SVR(), window_length=15)
k_neighbours_regressor = RegressorPipe(model_name ='k_neighbours_regressor',regressor=KNeighborsRegressor(), window_length=15)
########################################### SUPERVISED CLASSIFICATION ##########################################
class SupervisedClassificationPipe(BasePipe):
"""
Parameters
----------
model_name : str
name of model
model : object
sklearn classifier
window_lenght : int
For converting time series to tabular format.
"""
def __init__(self, model_name, classifier, window_length):
self.model_name = model_name
self.classifier = classifier
self.window_length = window_length
def time_series_to_tabular(self,y,window_length, fh, return_value):
"""
Converts a pd.Series to tabular format. Lables target variable 1 for up 0 for down
Parameters
----------
y : pd.Series
input time series
window_length : int
number of features for each raw of X
fh : ForecastingHorizon
ForecastingHorizon object
stage : str
must be `fit` or `predict`
return_value : string
acceptable values `X` and `y` only
Returns
-------
tuple of (numpy.ndarray, numpy.ndarray) where the first element are the target (y) variable and the second element are the features (X)
"""
fh=fh
y_tmp,x_dc = _sliding_window_transform(y,window_length=window_length,fh=fh)
y_dc = np.zeros(len(y_tmp))
y_tmp = y_tmp.reshape(x_dc[:,-1].shape)
y_mask = (x_dc[:,-1] > y_tmp) #up observations
y_dc[y_mask]=1
if return_value == 'X':
return x_dc
if return_value == 'y':
return y_dc
def concatenator(self,y_train, y_test, window_length):
return pd.concat([y_train[-window_length-1:-1],y_test])
def build(self, **kwargs):
classifier_pipe = NetworkPipelineForecaster([
('concatenator',self.concatenator, {'fit':None,
'predict':{'y_train': kwargs['y_train'], 'y_test': kwargs['y_test'], 'window_length':self.window_length}}),
('time_series_to_tabular_x', self.time_series_to_tabular,{
'fit':{'y':'original_y', 'window_length':self.window_length, 'fh': 'original_fh', 'return_value':'X'},
'predict': {'y':'concatenator', 'window_length':self.window_length, 'fh': 'original_fh', 'return_value':'X'} }),
('time_series_to_tabular_y', self.time_series_to_tabular,{
'fit':{'y':'original_y', 'window_length':self.window_length, 'fh': 'original_fh', 'return_value':'y'},
'predict': None }),
('classifier', self.classifier, {'fit':{'X': 'time_series_to_tabular_x', 'y': 'time_series_to_tabular_y' },
'predict': {'X': 'time_series_to_tabular_x'} })
])
self.model = classifier_pipe
return classifier_pipe
def get_fh(self, **kwargs):
return ForecastingHorizon([1])
logistic_regression = SupervisedClassificationPipe(model_name='LogisticRegression', classifier=linear_model.LogisticRegression(), window_length=15)
rf_classifier = SupervisedClassificationPipe(model_name='RandomForestClassifier', classifier=RandomForestClassifier(), window_length=15)
svm_classifier = SupervisedClassificationPipe(model_name='SVMClassifier', classifier=svm.SVC(), window_length=15)
k_neighbours_classifier = SupervisedClassificationPipe(model_name='K_NeighboursClassifier', classifier=KNeighborsClassifier(), window_length=15)
########################################### EXOGENOUS ##########################################
class HHMExogenousPipeRegressor(BasePipe):
"""
Parameters
----------
model_name : str
name of model
regressor_model : object
primary model
exogenous_model : object
secondary model
"""
def __init__(self, model_name, regressor_model, exogenous_model, window_length):
self.model_name = model_name
self.exogenous_model = exogenous_model
self.regressor_model = regressor_model
self.window_length = window_length
def reshape(self,y):
return y.values.reshape(-1,1)
def decoder(self,est,y):
X = est.decode(y.values.reshape(-1,1))[1]
X = pd.Series(X)
X.index = y.index
return X
def exogenous(self, method, y_train, y_test=None):
"""
Simulates sequential arrival of the y_test data.
Exogenous model makes 1 steap ahead predictions.
Used to avoid leakage of the test data
Parameters
----------
method : str (fit or predict)
y_train : pd.Series
y_test : pd.Series
Retrurns
--------
list
"""
if method == 'predict':
comb_y = pd.concat([y_train, y_test])
predictions = []
splitter = SlidingWindowSplitter(fh=[1], window_length=y_train.shape[0])
for split in splitter.split(comb_y):
y = comb_y.iloc[split[0]].values.reshape(-1,1)
self.exogenous_model.fit(y)
predictions.append(self.exogenous_model.decode(y)[1][-1])
return pd.Series(index=y_test.index,data=predictions)
if method == 'fit':
y = y_train.values.reshape(-1,1)
self.exogenous_model.fit(y)
decoded_y = self.exogenous_model.decode(y)[1]
return pd.Series(index=y_train.index, data=decoded_y)
def build(self, **kwargs):
exogenous_pipe = NetworkPipelineForecaster([
('exogenous_model', self.exogenous, {'fit': {'method':'fit', 'y_train':'original_y'},
'predict':{'method':'predict', 'y_train':kwargs['y_train'], 'y_test':kwargs['y_test']}}),
('regressor', make_reduction(self.regressor_model, window_length=5, strategy='recursive'),
{'fit':{'y': 'original_y', 'X': 'exogenous_model', 'fh':'original_fh'},
'predict':{'X': 'exogenous_model'} } ),
('converter', converter, {'fit':None,
'predict':{'y1':kwargs['y_train'],'y2':'regressor'}})
])
self.model = exogenous_pipe
return exogenous_pipe
def get_fh(self, **kwargs):
return ForecastingHorizon(np.arange(1,len(kwargs['y_test'])+1))
rf_hmm_exogenous = HHMExogenousPipeRegressor(model_name='RF_HMM_exogenous', regressor_model=RandomForestRegressor(),
exogenous_model=hmm.GaussianHMM(n_components=3, covariance_type="diag", n_iter=1000),
window_length=15)
from mlfinlab.filters import cusum_filter
class CusumFilter:
def __init__(self, filter=cusum_filter, search_range = np.arange(5,0.01, step=-0.01), pct_positive_annotations=0.1, threshold=None):
self.filter = filter
self.search_range = search_range
self.pct_positive_annotations = pct_positive_annotations
self.threshold=threshold
def _find_threshold(self, X):
thresholds = self.search_range
sufficient_num_annotations = int(len(X) * self.pct_positive_annotations)
for threshold in thresholds:
num_annotations = len(self.filter(X, threshold))
if num_annotations > sufficient_num_annotations:
self.threshold = threshold
return threshold
self.threshold = thresholds[-1]
return thresholds[-1]
def _generate_predictions(self, X):
filter_annotations = self.filter(X, threshold=self.threshold)
predictions = pd.Series(data=np.zeros(len(X)), index=X.index)
predictions.loc[filter_annotations] = 1
return predictions
def get_params(self, **kwargs):
return {'threshold':self.threshold}
def fit(self, X):
self._find_threshold(X)
return self
def decode(self,X):
return self._generate_predictions(X)
def predict(self, X):
return self._generate_predictions(X)
class CUSUMExogenousPipeRegressor(BasePipe):
"""
Parameters
----------
model_name : str
name of model
regressor_model : object
primary model
exogenous_model : object
secondary model
"""
def __init__(self, model_name, regressor_model, exogenous_model, window_length):
self.model_name = model_name
self.exogenous_model = exogenous_model
self.regressor_model = regressor_model
self.window_length = window_length
def reshape(self,y):
return y.values.reshape(-1,1)
def decoder(self, est, y):
return est._generate_predictions(y)
def exogenous(self, method, y_train, y_test=None):
if method == 'fit':
self.exogenous_model._find_threshold(y_train)
return self.exogenous_model._generate_predictions(y_train)
if method == 'predict':
comb_y = pd.concat([y_train, y_test])
predictions = []
splitter = SlidingWindowSplitter(fh=[1], window_length=y_train.shape[0])
for split in splitter.split(comb_y):
y = comb_y.iloc[split[0]]
predictions.append(self.exogenous_model.predict(y)[-1])
return pd.Series(index=y_test.index,data=predictions)
def build(self, **kwargs):
exogenous_pipe = NetworkPipelineForecaster([
('exogenous_model', self.exogenous, {'fit': {'method':'fit', 'y_train':'original_y'},
'predict':{'method':'predict', 'y_train':kwargs['y_train'], 'y_test':kwargs['y_test']}}),
('regressor', make_reduction(self.regressor_model, window_length=5, strategy='recursive'),
{'fit':{'y': 'original_y', 'X': 'exogenous_model', 'fh':'original_fh'},
'predict':{'X': 'exogenous_model'} } ),
('converter', converter, {'fit':None,
'predict':{'y1':kwargs['y_train'],'y2':'regressor'}})
])
self.model = exogenous_pipe
return exogenous_pipe
def get_fh(self, **kwargs):
return ForecastingHorizon(np.arange(1,len(kwargs['y_test'])+1))
rf_cusum_exogenous = CUSUMExogenousPipeRegressor(model_name='RF_CUSUM_exogenous', regressor_model=RandomForestRegressor(),
exogenous_model=CusumFilter(),
window_length=15)
########################################### DC PIPELINES ##########################################
class DCPipeRegressor():
"""
Parameters
----------
model_name : str
name of model
regressor_model : object
primary model
dc_model : object
secondary model
"""
def __init__(self, model_name, regressor_model, dc_model, window_length):
self.model_name = model_name
self.dc_model = dc_model
self.regressor_model = regressor_model
self.window_length = window_length
def reshape(self,y):
return y.values.reshape(-1,1)
def _concatenate_train_test_sets(self, y_train, y_test):
return pd.concat([y_train, y_test], ignore_index=False)
def generate_dc(self, y_train, y_test):
y = self._concatenate_train_test_sets(y_train=y_train, y_test=y_test)
self.dc_model._find_threshold(y)
dc_events = self.dc_model._generate_predictions(y)
print (dc_events)
def get_model_name(self):
return self.model_name
def decoder(self, est, y):
return est._generate_predictions(y)
def dc(self, method, y_train, y_test=None):
if method == 'fit':
self.dc_model._find_threshold(y_train)
return self.dc_model._generate_predictions(y_train)
if method == 'predict':
comb_y = pd.concat([y_train, y_test])
predictions = []
splitter = SlidingWindowSplitter(fh=[1], window_length=y_train.shape[0])
for split in splitter.split(comb_y):
y = comb_y.iloc[split[0]]
predictions.append(self.dc_model.predict(y)[-1])
return pd.Series(index=y_test.index,data=predictions)
def build(self, **kwargs):
dc_pipe = NetworkPipelineForecaster([
('dc_model', self.dc, {'fit': {'method':'fit', 'y_train':'original_y'},
'predict':{'method':'predict', 'y_train':kwargs['y_train'], 'y_test':kwargs['y_test']}}),
('regressor', make_reduction(self.regressor_model, window_length=5, strategy='recursive'),
{'fit':{'y': 'original_y', 'X': 'dc_model', 'fh':'original_fh'},
'predict':{'X': 'dc_model'} } ),
('converter', converter, {'fit':None,
'predict':{'y1':kwargs['y_train'],'y2':'regressor'}})
])
self.model = dc_pipe
return dc_pipe
def get_fh(self, **kwargs):
return ForecastingHorizon(np.arange(1,len(kwargs['y_test'])+1))
rf_cusum_dc = DCPipeRegressor(model_name='RF_CUSUM_dc', regressor_model=RandomForestRegressor(),
dc_model=CusumFilter(),
window_length=15)