-
Notifications
You must be signed in to change notification settings - Fork 0
/
time_series_original.py
416 lines (350 loc) · 13.3 KB
/
time_series_original.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
407
408
409
410
411
412
413
414
415
416
""" Time Series Modelling module """
import os
import pickle
import logging
import numpy as np
import pandas as pd
from fbprophet import Prophet # TODO: Add to requirements.txt
from datetime import datetime
# import country_converter # TODO: Add to requirements.txt
from sklearn.model_selection import ParameterGrid
import boto3 # TODO: add to requirements.txt
from botocore.exceptions import ClientError
from pathlib import Path
logger = logging.getLogger("corona")
DEFAULT_PROPHET_PARAMS_GRID = {
"growth": ["linear"],
"seasonality_prior_scale": [1, 3, 10, 30],
"changepoint_range": [0.8, 0.9],
"changepoint_prior_scale": [0.01, 0.03, 0.1, 0.3, 1, 3],
"n_changepoints": [5, 10, 30, 50],
"seasonality_mode": ["additive"],
}
black_friday = pd.DataFrame(
{
"holiday": "black_friday",
"ds": pd.to_datetime(["2018-11-23", "2019-11-29", "2020-11-11"]),
#'lower_window': -6,
#'upper_window': 6,
}
)
class _suppress_stdout_stderr(object):
"""
A context manager for doing a "deep suppression" of stdout and stderr in
Python, i.e. will suppress all print, even if the print originates in a
compiled C/Fortran sub-function.
This will not suppress raised exceptions, since exceptions are printed
to stderr just before a script exits, and after the context manager has
exited (at least, I think that is why it lets exceptions through).
"""
def __init__(self):
# Open a pair of null files
self.null_fds = [os.open(os.devnull, os.O_RDWR) for x in range(2)]
# Save the actual stdout (1) and stderr (2) file descriptors.
self.save_fds = [os.dup(1), os.dup(2)]
def __enter__(self):
# Assign the null pointers to stdout and stderr.
os.dup2(self.null_fds[0], 1)
os.dup2(self.null_fds[1], 2)
def __exit__(self, *_):
# Re-assign the real stdout/stderr back to (1) and (2)
os.dup2(self.save_fds[0], 1)
os.dup2(self.save_fds[1], 2)
# Close the null files
for fd in self.null_fds + self.save_fds:
os.close(fd)
class TimeSeries(object):
"""
This class generates time series forecasts using Facebook Prophet model for different set of options.
The class can be used to make predictions from a pre-trained model, for training a model specifying
the parameters and for finding the optimal parameters for the input series (tuning and training).
Parameters
----------
input_series: pd.Series
input_freq: str
validation_split_date: str
train_until_date: str
horizon_date: str
pretrained_model: str
optimized_params: dict
prophet_params_grid: dict
add_country_holidays: bool
add_black_friday: bool
add_montly_seasonality: bool
country_name: str
transform: str
save: bool
file_name: str
file_path: str
use_pretrained_model: bool
load_path_to_file: str
Methods
----------
forecast
"""
def __init__(
self,
input_series: pd.Series,
input_freq: str = "D",
validation_split_date: str = None,
train_until_date: str = None,
horizon_date: str = None,
optimized_params: dict = None,
prophet_params_grid: dict = None,
add_country_holidays: bool = False, # TODO
add_black_friday: bool = False,
add_montly_seasonality: bool = False,
country_name: str = None, # TODO
transform: str = None, # Allow logaritmic and Boxcox transformations
save: bool = False,
use_pretrained_model: bool = None,
local_path: str = None,
model_name: str = None,
s3_load: bool = False,
s3_save: bool = False,
):
assert isinstance(
input_series.index, pd.DatetimeIndex
), "Index of the input must be 'DatetimeIndex'"
if not input_series.index.is_monotonic:
input_series.sort_index(inplace=True)
self.input_series = input_series
self.input_freq = input_freq
if validation_split_date is not None:
self.validation_split_date = pd.Timestamp(validation_split_date)
if prophet_params_grid is not None:
self.prophet_params_grid = prophet_params_grid
else:
self.prophet_params_grid = DEFAULT_PROPHET_PARAMS_GRID
if horizon_date is not None:
self.horizon_date = pd.Timestamp(horizon_date)
if train_until_date is not None:
self.train_until_date = pd.Timestamp(train_until_date)
# TODO: review logic for using pre-trained model, tuning+training, training
if use_pretrained_model is True:
self.ts_model = self._load_model(local_path, model_name, s3_load=s3_load)
else:
if optimized_params is None:
optimized_params = self._tune_parameters()
self.ts_model = self._train_model(optimized_params, validation=False)
if save is True:
self._save_model(local_path, model_name, s3_save=s3_save)
def _load_model(self, local_path, model_name, s3_load):
"""
Load pre-trained model. The model can be load either from a local location or from the S3 bucket.
Parameters
----------
local_path : str
Path to pre-trained model.
model_name: str
Name of the pickle file.
s3_load: bool
Whether or not you want to load the model from the S3 bucket
"""
if s3_load:
try:
client = boto3.client("s3")
response = client.get_object(
Bucket="ngap--marketplace-analytics--prod--eu-west-1",
Key="qa/analytics/prophet-models/{name}".format(name=model_name),
)
serialized_object = response["Body"].read()
return pickle.loads(serialized_object)
logger.info("{} Prophet model loaded from S3 bucket".format(model_name))
except ClientError as error:
response = error.response.get("Error", dict()).get("Code", "")
if response == "ExpiredToken":
logger.error(
"AWS Token has expired: run gimme-aws-creds to get the credentials"
)
else:
logger.error("Unexpected AWS error: %s" % error)
raise error
else:
file = Path(local_path, model_name)
if file.exists():
with open(file, "rb") as f:
return pickle.load(f)
raise OSError("Can't find model in local directory nor in AWS")
def _eval_model(self, model):
"""
Evaluate prophet model performance.
Parameters
----------
model : Prophet object
Model to evaluate
"""
y_true = self.input_series[
self.validation_split_date : self.train_until_date
].rename("true")
y_forecast = self._eval_forecast(model).rename("forecast")
df_eval = pd.concat([y_true, y_forecast], axis=1).dropna()
wmape = sum(np.abs(df_eval.true - df_eval.forecast)) / sum(df_eval.true) * 100
return wmape
def _train_eval_model(self, prophet_parameters):
"""
Train and evaluate prophet model performance.
Parameters
----------
prophet_parameters : dict
Parameters compatible with Prophet model.
"""
model = self._train_model(prophet_parameters)
acc = self._eval_model(model)
return acc
def _tune_parameters(self):
"""
Finds the best parameters for prophet model on a validation set.
"""
grid = ParameterGrid(self.prophet_params_grid)
# best_parameters = min(
# grid,
# key=lambda parameters: self._train_eval_model(prophet_parameters=parameters),
# )
acc_lst = []
for i, p in enumerate(grid):
print("Tuning parameters: {:2.1%}".format((i + 1) / len(grid)), end="\r")
p["acc"] = self._train_eval_model(prophet_parameters=p)
acc_lst.append(p)
best_parameters = min(acc_lst, key=lambda x: x["acc"])
del best_parameters["acc"]
return best_parameters
def _train_model(self, prophet_parameters, validation=True):
"""
Train prophet model.
Parameters
----------
prophet_parameters : dict
Parameters compatible with Prophet model.
validation : bool, default=True
Whether to train with or without validation set.
"""
date_col = self.input_series.index.name
val_col = self.input_series.name
if validation:
train_df = (
self.input_series[: self.validation_split_date]
.reset_index()
.rename(columns={date_col: "ds", val_col: "y"})
)
else:
train_df = (
self.input_series[: self.train_until_date]
.reset_index()
.rename(columns={date_col: "ds", val_col: "y"})
)
# Disable prophet logging, see https://github.com/facebook/prophet/issues/223 for info
with _suppress_stdout_stderr():
if self.input_freq == "D":
model = Prophet(
daily_seasonality=False, # TODO: add option of monthly seasonality
yearly_seasonality=True,
weekly_seasonality=True, # TODO: True if freq == 'D', False if freq == 'W'
**prophet_parameters,
)
elif self.input_freq[0] == "W":
model = Prophet(
daily_seasonality=False, # TODO: add option of monthly seasonality
yearly_seasonality=True,
weekly_seasonality=False,
**prophet_parameters,
)
model.fit(train_df)
return model
def _eval_forecast(self, model):
"""
Train and evaluate prophet model performance.
Parameters
----------
model : Prophet
Model to use for forecasting
"""
date_col = self.input_series.index.name
val_col = self.input_series.name
if self.input_freq == "D":
predict_range = (
self.train_until_date - self.validation_split_date
).days + 1
elif self.input_freq[0] == "W":
predict_range = int(
((self.train_until_date - self.validation_split_date).days + 1) / 7
)
future = model.make_future_dataframe(
periods=predict_range, freq=self.input_freq, include_history=False
)
forecast = model.predict(future)
forecast_series = (
forecast.assign(**{date_col: lambda x: pd.to_datetime(x.ds)})
.rename(columns={"yhat": val_col})
.set_index(date_col)[val_col]
)
return forecast_series
def forecast(self):
"""
Make predictions with trained model.
"""
date_col = self.input_series.index.name
val_col = self.input_series.name
if self.input_freq == "D":
predict_range = (self.horizon_date - self.train_until_date).days + 1
elif self.input_freq[0] == "W":
predict_range = int(
((self.horizon_date - self.train_until_date).days + 1) / 7
)
future = self.ts_model.make_future_dataframe(
periods=predict_range, freq=self.input_freq, include_history=False
)
forecast = self.ts_model.predict(future)
# TODO: output also upper and lower forecasts
forecast_series = (
forecast.assign(**{date_col: lambda x: pd.to_datetime(x.ds)})
.rename(columns={"yhat": val_col})
.set_index(date_col)[val_col]
.clip(lower=0) # replacing negative predictions with 0
)
return forecast_series
# TODO:
def _add_country_holidays(self):
pass
def _save_model(self, local_path, model_name, s3_save):
"""
Saves the model in a pickle file to a local directory or to the S3 bucket
Parameters
----------
local_path : str
Path to location where to save the model
model_name : str
Name of the file
s3_save: bool
Whether or not you want to save the model to the S3 bucket
"""
if s3_save:
try:
client = boto3.client("s3")
serialized_object = pickle.dumps(self.ts_model)
client.put_object(
Bucket="ngap--marketplace-analytics--prod--eu-west-1",
Key="qa/analytics/prophet-models/{}".format(model_name),
Body=serialized_object,
)
logger.info(
"Model saved in S3 bucket with filename: {}".format(model_name)
)
except ClientError as error:
response = error.response.get("Error", dict()).get("Code", "")
if response == "ExpiredToken":
logger.error(
"AWS Token has expired: run gimme-aws-creds to get the credentials"
)
else:
logger.error("Unexpected AWS error: %s" % error)
raise error
else:
path = Path(local_path)
if path.is_dir() and model_name != None:
file = Path(local_path, model_name)
with open(file, "wb") as f:
pickle.dump(self.ts_model, f)
logger.info("Model saved to local machine in %s" % file)
else:
logger.info("Path doesn't exist or model_name is not especified")