-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtime_based_features.py
203 lines (132 loc) · 8.21 KB
/
time_based_features.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
import numpy as np
import pandas as pd
def auto_add_time_based_features_periods(time_based_features_periods, train_timespan, freq):
if (
train_timespan >= pd.to_timedelta('5 days') and
freq < pd.to_timedelta('1 day') and
'time-in-day' not in time_based_features_periods
):
time_based_features_periods.append('time-in-day')
if train_timespan >= pd.to_timedelta('14 days') and 'day-in-week' not in time_based_features_periods:
time_based_features_periods.append('day-in-week')
if train_timespan >= pd.to_timedelta('730 days') and 'week-in-year' not in time_based_features_periods:
time_based_features_periods.append('week-in-year')
if train_timespan >= pd.to_timedelta('730 days') and 'month-in-year' not in time_based_features_periods:
time_based_features_periods.append('month-in-year')
def get_integer_encodings_for_values(values):
unique_values = set(values)
value_to_int = {value: i for i, value in enumerate(sorted(unique_values))}
integer_encodings = np.array([value_to_int[value] for value in values])
return integer_encodings
def get_time_in_hour_integer_feature_encodings(timestamps):
values = list(zip(timestamps.minute, timestamps.second))
integer_feature_encodings = get_integer_encodings_for_values(values)
return integer_feature_encodings
def get_time_in_day_integer_feature_encodings(timestamps):
values = list(zip(timestamps.hour, timestamps.minute, timestamps.second))
integer_feature_encodings = get_integer_encodings_for_values(values)
return integer_feature_encodings
def get_hour_in_day_integer_feature_encodings(timestamps):
return timestamps.hour.values
def get_day_in_week_integer_feature_encodings(timestamps):
return timestamps.day_of_week.values
def get_day_in_month_integer_feature_encodings(timestamps):
return timestamps.day.values - 1
def get_day_in_year_integer_feature_encodings(timestamps):
return timestamps.day_of_year.values - 1
def get_week_in_year_integer_feature_encodings(timestamps):
return (timestamps.day_of_year.values - 1) // 7
def get_month_in_year_integer_feature_encodings(timestamps):
return timestamps.month.values - 1
GET_INTEGER_FEATURE_ENCODINGS_FUNC_DICT = {
'time-in-hour': get_time_in_hour_integer_feature_encodings,
'time-in-day': get_time_in_day_integer_feature_encodings,
'hour-in-day': get_hour_in_day_integer_feature_encodings,
'day-in-week': get_day_in_week_integer_feature_encodings,
'day-in-month': get_day_in_month_integer_feature_encodings,
'day-in-year': get_day_in_year_integer_feature_encodings,
'week-in-year': get_week_in_year_integer_feature_encodings,
'month-in-year': get_month_in_year_integer_feature_encodings
}
def get_integer_feature_encodings(timestamps, period):
return GET_INTEGER_FEATURE_ENCODINGS_FUNC_DICT[period](timestamps)
def get_time_in_hour_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=integer_feature_encodings.max() + 1)
def get_time_in_day_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=integer_feature_encodings.max() + 1)
def get_hour_in_day_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=24)
def get_day_in_week_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=7)
def get_day_in_month_period_lengths(timestamps, integer_feature_encodings):
return timestamps.days_in_month.values
def get_day_in_year_period_lengths(timestamps, integer_feature_encodings):
return 365 + timestamps.is_leap_year.values
def get_week_in_year_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=52)
def get_month_in_year_period_lengths(timestamps, integer_feature_encodings):
return np.full(len(timestamps), fill_value=12)
GET_PERIOD_LENGTHS_FUNC_DICT = {
'time-in-hour': get_time_in_hour_period_lengths,
'time-in-day': get_time_in_day_period_lengths,
'hour-in-day': get_hour_in_day_period_lengths,
'day-in-week': get_day_in_week_period_lengths,
'day-in-month': get_day_in_month_period_lengths,
'day-in-year': get_day_in_year_period_lengths,
'week-in-year': get_week_in_year_period_lengths,
'month-in-year': get_month_in_year_period_lengths
}
def get_period_lengths(timestamps, integer_feature_encodings, period):
return GET_PERIOD_LENGTHS_FUNC_DICT[period](timestamps, integer_feature_encodings)
def get_one_hot_time_based_features(integer_feature_encodings, period):
num_categories = integer_feature_encodings.max() + 1
one_hot_time_based_features = np.zeros((len(integer_feature_encodings), num_categories), dtype=np.float32)
one_hot_time_based_features[np.arange(len(integer_feature_encodings)), integer_feature_encodings] = 1
one_hot_time_based_feature_names = [period.replace('-', '_') + f'_{i}' for i in range(num_categories)]
return one_hot_time_based_features, one_hot_time_based_feature_names
def get_sin_cos_time_based_features(integer_feature_encodings, period_lengths, period):
x = integer_feature_encodings / period_lengths * (np.pi * 2)
sin_cos_time_based_features = np.stack([np.sin(x), np.cos(x)], axis=1, dtype=np.float32)
sin_cos_time_based_feature_names = [period.replace('-', '_') + '_sin', period.replace('-', '_') + '_cos']
return sin_cos_time_based_features, sin_cos_time_based_feature_names
def create_time_based_features(unix_timestamps, time_based_features_types, time_based_features_periods,
last_train_timestamp_idx):
if not time_based_features_types or not time_based_features_periods:
return (
np.empty((len(unix_timestamps), 1, 0), dtype=np.float32), [],
np.empty((len(unix_timestamps), 1, 0), dtype=np.float32), []
)
timestamps = pd.to_datetime(unix_timestamps, origin='unix', unit='s')
timestamps = pd.DatetimeIndex(timestamps, freq='infer')
if 'auto' in time_based_features_periods:
train_timespan = timestamps[last_train_timestamp_idx] - timestamps[0] + timestamps.freq
time_based_features_periods = list(time_based_features_periods)
time_based_features_periods.remove('auto')
auto_add_time_based_features_periods(time_based_features_periods=time_based_features_periods,
train_timespan=train_timespan,
freq=timestamps.freq)
time_based_features, time_based_feature_names = [], []
for period in time_based_features_periods:
integer_feature_encodings = get_integer_feature_encodings(timestamps=timestamps, period=period)
if 'one-hot' in time_based_features_types:
cur_period_one_hot_time_based_features, cur_period_one_hot_time_based_feature_names = (
get_one_hot_time_based_features(integer_feature_encodings=integer_feature_encodings,
period=period)
)
time_based_features.append(cur_period_one_hot_time_based_features)
time_based_feature_names += cur_period_one_hot_time_based_feature_names
if 'sin-cos' in time_based_features_types:
period_lengths = get_period_lengths(
timestamps=timestamps, integer_feature_encodings=integer_feature_encodings, period=period
)
cur_period_sin_cos_time_based_features, cur_period_sin_cos_time_based_feature_names = (
get_sin_cos_time_based_features(integer_feature_encodings=integer_feature_encodings,
period_lengths=period_lengths,
period=period)
)
time_based_features.append(cur_period_sin_cos_time_based_features)
time_based_feature_names += cur_period_sin_cos_time_based_feature_names
time_based_features = np.concatenate(time_based_features, axis=1)
time_based_features = np.expand_dims(time_based_features, axis=1)
numerical_time_based_features_mask = np.zeros(time_based_features.shape[2], dtype=bool)
return time_based_features, time_based_feature_names, numerical_time_based_features_mask