|
1
|
|
|
# |
|
2
|
|
|
# Copyright 2015 Quantopian, Inc. |
|
3
|
|
|
# |
|
4
|
|
|
# Licensed under the Apache License, Version 2.0 (the "License"); |
|
5
|
|
|
# you may not use this file except in compliance with the License. |
|
6
|
|
|
# You may obtain a copy of the License at |
|
7
|
|
|
# |
|
8
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0 |
|
9
|
|
|
# |
|
10
|
|
|
# Unless required by applicable law or agreed to in writing, software |
|
11
|
|
|
# distributed under the License is distributed on an "AS IS" BASIS, |
|
12
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
|
13
|
|
|
# See the License for the specific language governing permissions and |
|
14
|
|
|
# limitations under the License. |
|
15
|
|
|
|
|
16
|
|
|
import bcolz |
|
17
|
|
|
from logbook import Logger |
|
18
|
|
|
|
|
19
|
|
|
import numpy as np |
|
20
|
|
|
import pandas as pd |
|
21
|
|
|
from pandas.tslib import normalize_date |
|
22
|
|
|
from six import iteritems |
|
23
|
|
|
|
|
24
|
|
|
from zipline.assets import Asset, Future, Equity |
|
25
|
|
|
from zipline.data.us_equity_pricing import NoDataOnDate |
|
26
|
|
|
|
|
27
|
|
|
from zipline.utils import tradingcalendar |
|
28
|
|
|
from zipline.errors import ( |
|
29
|
|
|
NoTradeDataAvailableTooEarly, |
|
30
|
|
|
NoTradeDataAvailableTooLate |
|
31
|
|
|
) |
|
32
|
|
|
|
|
33
|
|
|
log = Logger('DataPortal') |
|
34
|
|
|
|
|
35
|
|
|
BASE_FIELDS = { |
|
36
|
|
|
'open': 'open', |
|
37
|
|
|
'open_price': 'open', |
|
38
|
|
|
'high': 'high', |
|
39
|
|
|
'low': 'low', |
|
40
|
|
|
'close': 'close', |
|
41
|
|
|
'close_price': 'close', |
|
42
|
|
|
'volume': 'volume', |
|
43
|
|
|
'price': 'close' |
|
44
|
|
|
} |
|
45
|
|
|
|
|
46
|
|
|
|
|
47
|
|
|
class DataPortal(object): |
|
48
|
|
|
def __init__(self, |
|
49
|
|
|
env, |
|
50
|
|
|
equity_daily_reader=None, |
|
51
|
|
|
equity_minute_reader=None, |
|
52
|
|
|
future_daily_reader=None, |
|
53
|
|
|
future_minute_reader=None, |
|
54
|
|
|
adjustment_reader=None): |
|
55
|
|
|
self.env = env |
|
56
|
|
|
|
|
57
|
|
|
# This is a bit ugly, but is here for performance reasons. In minute |
|
58
|
|
|
# simulations, we need to very quickly go from dt -> (# of minutes |
|
59
|
|
|
# since Jan 1 2002 9:30 Eastern). |
|
60
|
|
|
# |
|
61
|
|
|
# The clock that heartbeats the simulation has all the necessary |
|
62
|
|
|
# information to do this calculation very quickly. This value is |
|
63
|
|
|
# calculated there, and then set here |
|
64
|
|
|
self.cur_data_offset = 0 |
|
65
|
|
|
|
|
66
|
|
|
self.views = {} |
|
67
|
|
|
|
|
68
|
|
|
self._asset_finder = env.asset_finder |
|
69
|
|
|
|
|
70
|
|
|
self._carrays = { |
|
71
|
|
|
'open': {}, |
|
72
|
|
|
'high': {}, |
|
73
|
|
|
'low': {}, |
|
74
|
|
|
'close': {}, |
|
75
|
|
|
'volume': {}, |
|
76
|
|
|
'sid': {}, |
|
77
|
|
|
'dt': {}, |
|
78
|
|
|
} |
|
79
|
|
|
|
|
80
|
|
|
self._adjustment_reader = adjustment_reader |
|
81
|
|
|
|
|
82
|
|
|
# caches of sid -> adjustment list |
|
83
|
|
|
self._splits_dict = {} |
|
84
|
|
|
self._mergers_dict = {} |
|
85
|
|
|
self._dividends_dict = {} |
|
86
|
|
|
|
|
87
|
|
|
# Cache of sid -> the first trading day of an asset, even if that day |
|
88
|
|
|
# is before 1/2/2002. |
|
89
|
|
|
self._asset_start_dates = {} |
|
90
|
|
|
self._asset_end_dates = {} |
|
91
|
|
|
|
|
92
|
|
|
# Handle extra sources, like Fetcher. |
|
93
|
|
|
self._augmented_sources_map = {} |
|
94
|
|
|
self._extra_source_df = None |
|
95
|
|
|
|
|
96
|
|
|
self.MINUTE_PRICE_ADJUSTMENT_FACTOR = 0.001 |
|
97
|
|
|
|
|
98
|
|
|
self._equity_daily_reader = equity_daily_reader |
|
99
|
|
|
self._equity_minute_reader = equity_minute_reader |
|
100
|
|
|
self._future_daily_reader = future_daily_reader |
|
101
|
|
|
self._future_minute_reader = future_minute_reader |
|
102
|
|
|
|
|
103
|
|
|
# The following values are used by _minute_offset to calculate the |
|
104
|
|
|
# index into the minute bcolz date. |
|
105
|
|
|
|
|
106
|
|
|
# A lookup of table every minute to the corresponding day, to avoid |
|
107
|
|
|
# calling `.date()` on every lookup. |
|
108
|
|
|
self._minutes_to_day = {} |
|
109
|
|
|
# A map of days (keyed by midnight) to a DatetimeIndex of market |
|
110
|
|
|
# minutes for that day. |
|
111
|
|
|
self._minutes_by_day = {} |
|
112
|
|
|
# A dict of day to the offset into the minute bcolz on which that |
|
113
|
|
|
# days data starts. |
|
114
|
|
|
self._day_offsets = None |
|
115
|
|
|
|
|
116
|
|
|
def handle_extra_source(self, source_df, sim_params): |
|
117
|
|
|
""" |
|
118
|
|
|
Extra sources always have a sid column. |
|
119
|
|
|
|
|
120
|
|
|
We expand the given data (by forward filling) to the full range of |
|
121
|
|
|
the simulation dates, so that lookup is fast during simulation. |
|
122
|
|
|
""" |
|
123
|
|
|
if source_df is None: |
|
124
|
|
|
return |
|
125
|
|
|
|
|
126
|
|
|
self._extra_source_df = source_df |
|
127
|
|
|
|
|
128
|
|
|
# source_df's sid column can either consist of assets we know about |
|
129
|
|
|
# (such as sid(24)) or of assets we don't know about (such as |
|
130
|
|
|
# palladium). |
|
131
|
|
|
# |
|
132
|
|
|
# In both cases, we break up the dataframe into individual dfs |
|
133
|
|
|
# that only contain a single asset's information. ie, if source_df |
|
134
|
|
|
# has data for PALLADIUM and GOLD, we split source_df into two |
|
135
|
|
|
# dataframes, one for each. (same applies if source_df has data for |
|
136
|
|
|
# AAPL and IBM). |
|
137
|
|
|
# |
|
138
|
|
|
# We then take each child df and reindex it to the simulation's date |
|
139
|
|
|
# range by forward-filling missing values. this makes reads simpler. |
|
140
|
|
|
# |
|
141
|
|
|
# Finally, we store the data. For each column, we store a mapping in |
|
142
|
|
|
# self.augmented_sources_map from the column to a dictionary of |
|
143
|
|
|
# asset -> df. In other words, |
|
144
|
|
|
# self.augmented_sources_map['days_to_cover']['AAPL'] gives us the df |
|
145
|
|
|
# holding that data. |
|
146
|
|
|
|
|
147
|
|
|
if sim_params.emission_rate == "daily": |
|
148
|
|
|
source_date_index = self.env.days_in_range( |
|
149
|
|
|
start=sim_params.period_start, |
|
150
|
|
|
end=sim_params.period_end |
|
151
|
|
|
) |
|
152
|
|
|
else: |
|
153
|
|
|
source_date_index = self.env.minutes_for_days_in_range( |
|
154
|
|
|
start=sim_params.period_start, |
|
155
|
|
|
end=sim_params.period_end |
|
156
|
|
|
) |
|
157
|
|
|
|
|
158
|
|
|
# Break the source_df up into one dataframe per sid. This lets |
|
159
|
|
|
# us (more easily) calculate accurate start/end dates for each sid, |
|
160
|
|
|
# de-dup data, and expand the data to fit the backtest start/end date. |
|
161
|
|
|
grouped_by_sid = source_df.groupby(["sid"]) |
|
162
|
|
|
group_names = grouped_by_sid.groups.keys() |
|
163
|
|
|
group_dict = {} |
|
164
|
|
|
for group_name in group_names: |
|
165
|
|
|
group_dict[group_name] = grouped_by_sid.get_group(group_name) |
|
166
|
|
|
|
|
167
|
|
|
for identifier, df in iteritems(group_dict): |
|
168
|
|
|
# Before reindexing, save the earliest and latest dates |
|
169
|
|
|
earliest_date = df.index[0] |
|
170
|
|
|
latest_date = df.index[-1] |
|
171
|
|
|
|
|
172
|
|
|
# Since we know this df only contains a single sid, we can safely |
|
173
|
|
|
# de-dupe by the index (dt) |
|
174
|
|
|
df = df.groupby(level=0).last() |
|
175
|
|
|
|
|
176
|
|
|
# Reindex the dataframe based on the backtest start/end date. |
|
177
|
|
|
# This makes reads easier during the backtest. |
|
178
|
|
|
df = df.reindex(index=source_date_index, method='ffill') |
|
179
|
|
|
|
|
180
|
|
|
if not isinstance(identifier, Asset): |
|
181
|
|
|
# for fake assets we need to store a start/end date |
|
182
|
|
|
self._asset_start_dates[identifier] = earliest_date |
|
183
|
|
|
self._asset_end_dates[identifier] = latest_date |
|
184
|
|
|
|
|
185
|
|
|
for col_name in df.columns.difference(['sid']): |
|
186
|
|
|
if col_name not in self._augmented_sources_map: |
|
187
|
|
|
self._augmented_sources_map[col_name] = {} |
|
188
|
|
|
|
|
189
|
|
|
self._augmented_sources_map[col_name][identifier] = df |
|
190
|
|
|
|
|
191
|
|
|
def _open_minute_file(self, field, asset): |
|
192
|
|
|
sid_str = str(int(asset)) |
|
193
|
|
|
|
|
194
|
|
|
try: |
|
195
|
|
|
carray = self._carrays[field][sid_str] |
|
196
|
|
|
except KeyError: |
|
197
|
|
|
carray = self._carrays[field][sid_str] = \ |
|
198
|
|
|
self._get_ctable(asset)[field] |
|
199
|
|
|
|
|
200
|
|
|
return carray |
|
201
|
|
|
|
|
202
|
|
|
def _get_ctable(self, asset): |
|
203
|
|
|
sid = int(asset) |
|
204
|
|
|
|
|
205
|
|
|
if isinstance(asset, Future): |
|
206
|
|
|
if self._future_minute_reader.sid_path_func is not None: |
|
207
|
|
|
path = self._future_minute_reader.sid_path_func( |
|
208
|
|
|
self._future_minute_reader.rootdir, sid |
|
209
|
|
|
) |
|
210
|
|
|
else: |
|
211
|
|
|
path = "{0}/{1}.bcolz".format( |
|
212
|
|
|
self._future_minute_reader.rootdir, sid) |
|
213
|
|
|
elif isinstance(asset, Equity): |
|
214
|
|
|
if self._equity_minute_reader.sid_path_func is not None: |
|
215
|
|
|
path = self._equity_minute_reader.sid_path_func( |
|
216
|
|
|
self._equity_minute_reader.rootdir, sid |
|
217
|
|
|
) |
|
218
|
|
|
else: |
|
219
|
|
|
path = "{0}/{1}.bcolz".format( |
|
220
|
|
|
self._equity_minute_reader.rootdir, sid) |
|
221
|
|
|
|
|
222
|
|
|
else: |
|
223
|
|
|
# TODO: Figure out if assets should be allowed if neither, and |
|
224
|
|
|
# why this code path is being hit. |
|
225
|
|
|
if self._equity_minute_reader.sid_path_func is not None: |
|
226
|
|
|
path = self._equity_minute_reader.sid_path_func( |
|
227
|
|
|
self._equity_minute_reader.rootdir, sid |
|
228
|
|
|
) |
|
229
|
|
|
else: |
|
230
|
|
|
path = "{0}/{1}.bcolz".format( |
|
231
|
|
|
self._equity_minute_reader.rootdir, sid) |
|
232
|
|
|
|
|
233
|
|
|
return bcolz.open(path, mode='r') |
|
234
|
|
|
|
|
235
|
|
|
def get_previous_value(self, asset, field, dt, data_frequency): |
|
236
|
|
|
""" |
|
237
|
|
|
Given an asset and a column and a dt, returns the previous value for |
|
238
|
|
|
the same asset/column pair. If this data portal is in minute mode, |
|
239
|
|
|
it's the previous minute value, otherwise it's the previous day's |
|
240
|
|
|
value. |
|
241
|
|
|
|
|
242
|
|
|
Parameters |
|
243
|
|
|
--------- |
|
244
|
|
|
asset : Asset |
|
245
|
|
|
The asset whose data is desired. |
|
246
|
|
|
|
|
247
|
|
|
field: string |
|
248
|
|
|
The desired field of the asset. Valid values are "open", |
|
249
|
|
|
"open_price", "high", "low", "close", "close_price", "volume", and |
|
250
|
|
|
"price". |
|
251
|
|
|
|
|
252
|
|
|
dt: pd.Timestamp |
|
253
|
|
|
The timestamp from which to go back in time one slot. |
|
254
|
|
|
|
|
255
|
|
|
data_frequency: string |
|
256
|
|
|
The frequency of the data to query; i.e. whether the data is |
|
257
|
|
|
'daily' or 'minute' bars |
|
258
|
|
|
|
|
259
|
|
|
Returns |
|
260
|
|
|
------- |
|
261
|
|
|
The value of the desired field at the desired time. |
|
262
|
|
|
""" |
|
263
|
|
|
if data_frequency == 'daily': |
|
264
|
|
|
prev_dt = self.env.previous_trading_day(dt) |
|
265
|
|
|
elif data_frequency == 'minute': |
|
266
|
|
|
prev_dt = self.env.previous_market_minute(dt) |
|
267
|
|
|
|
|
268
|
|
|
return self.get_spot_value(asset, field, prev_dt, data_frequency) |
|
269
|
|
|
|
|
270
|
|
|
def _check_extra_sources(self, asset, column, day): |
|
271
|
|
|
# If we have an extra source with a column called "price", only look |
|
272
|
|
|
# at it if it's on something like palladium and not AAPL (since our |
|
273
|
|
|
# own price data always wins when dealing with assets). |
|
274
|
|
|
look_in_augmented_sources = column in self._augmented_sources_map and \ |
|
275
|
|
|
not (column in BASE_FIELDS and isinstance(asset, Asset)) |
|
276
|
|
|
|
|
277
|
|
|
if look_in_augmented_sources: |
|
278
|
|
|
# we're being asked for a field in an extra source |
|
279
|
|
|
try: |
|
280
|
|
|
return self._augmented_sources_map[column][asset].\ |
|
281
|
|
|
loc[day, column] |
|
282
|
|
|
except: |
|
283
|
|
|
log.error( |
|
284
|
|
|
"Could not find value for asset={0}, day={1}," |
|
285
|
|
|
"column={2}".format( |
|
286
|
|
|
str(asset), |
|
287
|
|
|
str(day), |
|
288
|
|
|
str(column))) |
|
289
|
|
|
|
|
290
|
|
|
raise KeyError |
|
291
|
|
|
|
|
292
|
|
|
def get_spot_value(self, asset, field, dt, data_frequency): |
|
293
|
|
|
""" |
|
294
|
|
|
Public API method that returns a scalar value representing the value |
|
295
|
|
|
of the desired asset's field at either the given dt. |
|
296
|
|
|
|
|
297
|
|
|
Parameters |
|
298
|
|
|
--------- |
|
299
|
|
|
asset : Asset |
|
300
|
|
|
The asset whose data is desired.gith |
|
301
|
|
|
|
|
302
|
|
|
field: string |
|
303
|
|
|
The desired field of the asset. Valid values are "open", |
|
304
|
|
|
"open_price", "high", "low", "close", "close_price", "volume", and |
|
305
|
|
|
"price". |
|
306
|
|
|
|
|
307
|
|
|
dt: pd.Timestamp |
|
308
|
|
|
The timestamp for the desired value. |
|
309
|
|
|
|
|
310
|
|
|
data_frequency: string |
|
311
|
|
|
The frequency of the data to query; i.e. whether the data is |
|
312
|
|
|
'daily' or 'minute' bars |
|
313
|
|
|
|
|
314
|
|
|
Returns |
|
315
|
|
|
------- |
|
316
|
|
|
The value of the desired field at the desired time. |
|
317
|
|
|
""" |
|
318
|
|
|
extra_source_val = self._check_extra_sources( |
|
319
|
|
|
asset, |
|
320
|
|
|
field, |
|
321
|
|
|
dt, |
|
322
|
|
|
) |
|
323
|
|
|
|
|
324
|
|
|
if extra_source_val is not None: |
|
325
|
|
|
return extra_source_val |
|
326
|
|
|
|
|
327
|
|
|
if field not in BASE_FIELDS: |
|
328
|
|
|
raise KeyError("Invalid column: " + str(field)) |
|
329
|
|
|
|
|
330
|
|
|
column_to_use = BASE_FIELDS[field] |
|
331
|
|
|
|
|
332
|
|
|
if isinstance(asset, int): |
|
333
|
|
|
asset = self._asset_finder.retrieve_asset(asset) |
|
334
|
|
|
|
|
335
|
|
|
self._check_is_currently_alive(asset, dt) |
|
336
|
|
|
|
|
337
|
|
|
if data_frequency == "daily": |
|
338
|
|
|
day_to_use = dt |
|
339
|
|
|
day_to_use = normalize_date(day_to_use) |
|
340
|
|
|
return self._get_daily_data(asset, column_to_use, day_to_use) |
|
341
|
|
|
else: |
|
342
|
|
|
if isinstance(asset, Future): |
|
343
|
|
|
return self._get_minute_spot_value_future( |
|
344
|
|
|
asset, column_to_use, dt) |
|
345
|
|
|
else: |
|
346
|
|
|
return self._get_minute_spot_value( |
|
347
|
|
|
asset, column_to_use, dt) |
|
348
|
|
|
|
|
349
|
|
|
def _get_minute_spot_value_future(self, asset, column, dt): |
|
350
|
|
|
# Futures bcolz files have 1440 bars per day (24 hours), 7 days a week. |
|
351
|
|
|
# The file attributes contain the "start_dt" and "last_dt" fields, |
|
352
|
|
|
# which represent the time period for this bcolz file. |
|
353
|
|
|
|
|
354
|
|
|
# The start_dt is midnight of the first day that this future started |
|
355
|
|
|
# trading. |
|
356
|
|
|
|
|
357
|
|
|
# figure out the # of minutes between dt and this asset's start_dt |
|
358
|
|
|
start_date = self._get_asset_start_date(asset) |
|
359
|
|
|
minute_offset = int((dt - start_date).total_seconds() / 60) |
|
360
|
|
|
|
|
361
|
|
|
if minute_offset < 0: |
|
362
|
|
|
# asking for a date that is before the asset's start date, no dice |
|
363
|
|
|
return 0.0 |
|
364
|
|
|
|
|
365
|
|
|
# then just index into the bcolz carray at that offset |
|
366
|
|
|
carray = self._open_minute_file(column, asset) |
|
367
|
|
|
result = carray[minute_offset] |
|
368
|
|
|
|
|
369
|
|
|
# if there's missing data, go backwards until we run out of file |
|
370
|
|
|
while result == 0 and minute_offset > 0: |
|
371
|
|
|
minute_offset -= 1 |
|
372
|
|
|
result = carray[minute_offset] |
|
373
|
|
|
|
|
374
|
|
|
if column != 'volume': |
|
375
|
|
|
return result * self.MINUTE_PRICE_ADJUSTMENT_FACTOR |
|
376
|
|
|
else: |
|
377
|
|
|
return result |
|
378
|
|
|
|
|
379
|
|
|
def setup_offset_cache(self, minutes_by_day, minutes_to_day, trading_days): |
|
380
|
|
|
# TODO: This case should not be hit, but is when tests are setup |
|
381
|
|
|
# with data_frequency of daily, but run with minutely. |
|
382
|
|
|
if self._equity_minute_reader is None: |
|
383
|
|
|
return |
|
384
|
|
|
|
|
385
|
|
|
self._minutes_to_day = minutes_to_day |
|
386
|
|
|
self._minutes_by_day = minutes_by_day |
|
387
|
|
|
start = trading_days[0] |
|
388
|
|
|
first_trading_day_idx = self._equity_minute_reader.trading_days.\ |
|
389
|
|
|
searchsorted(start) |
|
390
|
|
|
self._day_offsets = { |
|
391
|
|
|
day: (i + first_trading_day_idx) * 390 |
|
392
|
|
|
for i, day in enumerate(trading_days)} |
|
393
|
|
|
|
|
394
|
|
|
def _minute_offset(self, dt): |
|
395
|
|
|
if self._day_offsets is not None: |
|
396
|
|
|
try: |
|
397
|
|
|
day = self._minutes_to_day[dt] |
|
398
|
|
|
minutes = self._minutes_by_day[day] |
|
399
|
|
|
return self._day_offsets[day] + minutes.get_loc(dt) |
|
400
|
|
|
except KeyError: |
|
401
|
|
|
return None |
|
402
|
|
|
|
|
403
|
|
|
def _get_minute_spot_value(self, asset, column, dt): |
|
404
|
|
|
# if dt is before the first market minute, minute_index |
|
405
|
|
|
# will be 0. if it's after the last market minute, it'll |
|
406
|
|
|
# be len(minutes_for_day) |
|
407
|
|
|
minute_offset_to_use = self._minute_offset(dt) |
|
408
|
|
|
|
|
409
|
|
|
if minute_offset_to_use is None: |
|
410
|
|
|
given_day = pd.Timestamp(dt.date(), tz='utc') |
|
411
|
|
|
day_index = self._equity_minute_reader.trading_days.searchsorted( |
|
412
|
|
|
given_day) |
|
413
|
|
|
|
|
414
|
|
|
# if dt is before the first market minute, minute_index |
|
415
|
|
|
# will be 0. if it's after the last market minute, it'll |
|
416
|
|
|
# be len(minutes_for_day) |
|
417
|
|
|
minute_index = self.env.market_minutes_for_day(given_day).\ |
|
418
|
|
|
searchsorted(dt) |
|
419
|
|
|
|
|
420
|
|
|
minute_offset_to_use = (day_index * 390) + minute_index |
|
421
|
|
|
|
|
422
|
|
|
carray = self._equity_minute_reader._open_minute_file(column, asset) |
|
423
|
|
|
result = carray[minute_offset_to_use] |
|
424
|
|
|
|
|
425
|
|
|
if result == 0: |
|
426
|
|
|
# if the given minute doesn't have data, we need to seek |
|
427
|
|
|
# backwards until we find data. This makes the data |
|
428
|
|
|
# forward-filled. |
|
429
|
|
|
|
|
430
|
|
|
# get this asset's start date, so that we don't look before it. |
|
431
|
|
|
start_date = self._get_asset_start_date(asset) |
|
432
|
|
|
start_date_idx = self._equity_minute_reader.trading_days.\ |
|
433
|
|
|
searchsorted(start_date) |
|
434
|
|
|
start_day_offset = start_date_idx * 390 |
|
435
|
|
|
|
|
436
|
|
|
original_start = minute_offset_to_use |
|
437
|
|
|
|
|
438
|
|
|
while result == 0 and minute_offset_to_use > start_day_offset: |
|
439
|
|
|
minute_offset_to_use -= 1 |
|
440
|
|
|
result = carray[minute_offset_to_use] |
|
441
|
|
|
|
|
442
|
|
|
# once we've found data, we need to check whether it needs |
|
443
|
|
|
# to be adjusted. |
|
444
|
|
|
if result != 0: |
|
445
|
|
|
minutes = self.env.market_minute_window( |
|
446
|
|
|
start=dt, |
|
447
|
|
|
count=(original_start - minute_offset_to_use + 1), |
|
448
|
|
|
step=-1 |
|
449
|
|
|
).order() |
|
450
|
|
|
|
|
451
|
|
|
# only need to check for adjustments if we've gone back |
|
452
|
|
|
# far enough to cross the day boundary. |
|
453
|
|
|
if minutes[0].date() != minutes[-1].date(): |
|
454
|
|
|
# create a np array of size minutes, fill it all with |
|
455
|
|
|
# the same value. and adjust the array. |
|
456
|
|
|
arr = np.array([result] * len(minutes), |
|
457
|
|
|
dtype=np.float64) |
|
458
|
|
|
self._apply_all_adjustments( |
|
459
|
|
|
data=arr, |
|
460
|
|
|
asset=asset, |
|
461
|
|
|
dts=minutes, |
|
462
|
|
|
field=column |
|
463
|
|
|
) |
|
464
|
|
|
|
|
465
|
|
|
# The first value of the adjusted array is the value |
|
466
|
|
|
# we want. |
|
467
|
|
|
result = arr[0] |
|
468
|
|
|
|
|
469
|
|
|
if column != 'volume': |
|
470
|
|
|
return result * self.MINUTE_PRICE_ADJUSTMENT_FACTOR |
|
471
|
|
|
else: |
|
472
|
|
|
return result |
|
473
|
|
|
|
|
474
|
|
|
def _get_daily_data(self, asset, column, dt): |
|
475
|
|
|
while True: |
|
476
|
|
|
try: |
|
477
|
|
|
value = self._equity_daily_reader.spot_price(asset, dt, column) |
|
478
|
|
|
if value != -1: |
|
479
|
|
|
return value |
|
480
|
|
|
else: |
|
481
|
|
|
dt -= tradingcalendar.trading_day |
|
482
|
|
|
except NoDataOnDate: |
|
483
|
|
|
return 0 |
|
484
|
|
|
|
|
485
|
|
|
def _get_history_daily_window(self, assets, end_dt, bar_count, |
|
486
|
|
|
field_to_use): |
|
487
|
|
|
""" |
|
488
|
|
|
Internal method that returns a dataframe containing history bars |
|
489
|
|
|
of daily frequency for the given sids. |
|
490
|
|
|
""" |
|
491
|
|
|
day_idx = tradingcalendar.trading_days.searchsorted(end_dt.date()) |
|
492
|
|
|
days_for_window = tradingcalendar.trading_days[ |
|
493
|
|
|
(day_idx - bar_count + 1):(day_idx + 1)] |
|
494
|
|
|
|
|
495
|
|
|
if len(assets) == 0: |
|
496
|
|
|
return pd.DataFrame(None, |
|
497
|
|
|
index=days_for_window, |
|
498
|
|
|
columns=None) |
|
499
|
|
|
|
|
500
|
|
|
data = [] |
|
501
|
|
|
|
|
502
|
|
|
for asset in assets: |
|
503
|
|
|
if isinstance(asset, Future): |
|
504
|
|
|
data.append(self._get_history_daily_window_future( |
|
505
|
|
|
asset, days_for_window, end_dt, field_to_use |
|
506
|
|
|
)) |
|
507
|
|
|
else: |
|
508
|
|
|
data.append(self._get_history_daily_window_equity( |
|
509
|
|
|
asset, days_for_window, end_dt, field_to_use |
|
510
|
|
|
)) |
|
511
|
|
|
|
|
512
|
|
|
return pd.DataFrame( |
|
513
|
|
|
np.array(data).T, |
|
514
|
|
|
index=days_for_window, |
|
515
|
|
|
columns=assets |
|
516
|
|
|
) |
|
517
|
|
|
|
|
518
|
|
|
def _get_history_daily_window_future(self, asset, days_for_window, |
|
519
|
|
|
end_dt, column): |
|
520
|
|
|
# Since we don't have daily bcolz files for futures (yet), use minute |
|
521
|
|
|
# bars to calculate the daily values. |
|
522
|
|
|
data = [] |
|
523
|
|
|
data_groups = [] |
|
524
|
|
|
|
|
525
|
|
|
# get all the minutes for the days NOT including today |
|
526
|
|
|
for day in days_for_window[:-1]: |
|
527
|
|
|
minutes = self.env.market_minutes_for_day(day) |
|
528
|
|
|
|
|
529
|
|
|
values_for_day = np.zeros(len(minutes), dtype=np.float64) |
|
530
|
|
|
|
|
531
|
|
|
for idx, minute in enumerate(minutes): |
|
532
|
|
|
minute_val = self._get_minute_spot_value_future( |
|
533
|
|
|
asset, column, minute |
|
534
|
|
|
) |
|
535
|
|
|
|
|
536
|
|
|
values_for_day[idx] = minute_val |
|
537
|
|
|
|
|
538
|
|
|
data_groups.append(values_for_day) |
|
539
|
|
|
|
|
540
|
|
|
# get the minutes for today |
|
541
|
|
|
last_day_minutes = pd.date_range( |
|
542
|
|
|
start=self.env.get_open_and_close(end_dt)[0], |
|
543
|
|
|
end=end_dt, |
|
544
|
|
|
freq="T" |
|
545
|
|
|
) |
|
546
|
|
|
|
|
547
|
|
|
values_for_last_day = np.zeros(len(last_day_minutes), dtype=np.float64) |
|
548
|
|
|
|
|
549
|
|
|
for idx, minute in enumerate(last_day_minutes): |
|
550
|
|
|
minute_val = self._get_minute_spot_value_future( |
|
551
|
|
|
asset, column, minute |
|
552
|
|
|
) |
|
553
|
|
|
|
|
554
|
|
|
values_for_last_day[idx] = minute_val |
|
555
|
|
|
|
|
556
|
|
|
data_groups.append(values_for_last_day) |
|
557
|
|
|
|
|
558
|
|
|
for group in data_groups: |
|
559
|
|
|
if len(group) == 0: |
|
560
|
|
|
continue |
|
561
|
|
|
|
|
562
|
|
|
if column == 'volume': |
|
563
|
|
|
data.append(np.sum(group)) |
|
564
|
|
|
elif column == 'open': |
|
565
|
|
|
data.append(group[0]) |
|
566
|
|
|
elif column == 'close': |
|
567
|
|
|
data.append(group[-1]) |
|
568
|
|
|
elif column == 'high': |
|
569
|
|
|
data.append(np.amax(group)) |
|
570
|
|
|
elif column == 'low': |
|
571
|
|
|
data.append(np.amin(group)) |
|
572
|
|
|
|
|
573
|
|
|
return data |
|
574
|
|
|
|
|
575
|
|
|
def _get_history_daily_window_equity(self, asset, days_for_window, |
|
576
|
|
|
end_dt, field_to_use): |
|
577
|
|
|
sid = int(asset) |
|
578
|
|
|
ends_at_midnight = end_dt.hour == 0 and end_dt.minute == 0 |
|
579
|
|
|
|
|
580
|
|
|
# get the start and end dates for this sid |
|
581
|
|
|
end_date = self._get_asset_end_date(asset) |
|
582
|
|
|
|
|
583
|
|
|
if ends_at_midnight or (days_for_window[-1] > end_date): |
|
584
|
|
|
# two cases where we use daily data for the whole range: |
|
585
|
|
|
# 1) the history window ends at midnight utc. |
|
586
|
|
|
# 2) the last desired day of the window is after the |
|
587
|
|
|
# last trading day, use daily data for the whole range. |
|
588
|
|
|
return self._get_daily_window_for_sid( |
|
589
|
|
|
asset, |
|
590
|
|
|
field_to_use, |
|
591
|
|
|
days_for_window, |
|
592
|
|
|
extra_slot=False |
|
593
|
|
|
) |
|
594
|
|
|
else: |
|
595
|
|
|
# for the last day of the desired window, use minute |
|
596
|
|
|
# data and aggregate it. |
|
597
|
|
|
all_minutes_for_day = self.env.market_minutes_for_day( |
|
598
|
|
|
pd.Timestamp(end_dt.date())) |
|
599
|
|
|
|
|
600
|
|
|
last_minute_idx = all_minutes_for_day.searchsorted(end_dt) |
|
601
|
|
|
|
|
602
|
|
|
# these are the minutes for the partial day |
|
603
|
|
|
minutes_for_partial_day =\ |
|
604
|
|
|
all_minutes_for_day[0:(last_minute_idx + 1)] |
|
605
|
|
|
|
|
606
|
|
|
daily_data = self._get_daily_window_for_sid( |
|
607
|
|
|
sid, |
|
608
|
|
|
field_to_use, |
|
609
|
|
|
days_for_window[0:-1] |
|
610
|
|
|
) |
|
611
|
|
|
|
|
612
|
|
|
minute_data = self._get_minute_window_for_equity( |
|
613
|
|
|
sid, |
|
614
|
|
|
field_to_use, |
|
615
|
|
|
minutes_for_partial_day |
|
616
|
|
|
) |
|
617
|
|
|
|
|
618
|
|
|
if field_to_use == 'volume': |
|
619
|
|
|
minute_value = np.sum(minute_data) |
|
620
|
|
|
elif field_to_use == 'open': |
|
621
|
|
|
minute_value = minute_data[0] |
|
622
|
|
|
elif field_to_use == 'close': |
|
623
|
|
|
minute_value = minute_data[-1] |
|
624
|
|
|
elif field_to_use == 'high': |
|
625
|
|
|
minute_value = np.amax(minute_data) |
|
626
|
|
|
elif field_to_use == 'low': |
|
627
|
|
|
minute_value = np.amin(minute_data) |
|
628
|
|
|
|
|
629
|
|
|
# append the partial day. |
|
630
|
|
|
daily_data[-1] = minute_value |
|
631
|
|
|
|
|
632
|
|
|
return daily_data |
|
633
|
|
|
|
|
634
|
|
|
def _get_history_minute_window(self, assets, end_dt, bar_count, |
|
635
|
|
|
field_to_use): |
|
636
|
|
|
""" |
|
637
|
|
|
Internal method that returns a dataframe containing history bars |
|
638
|
|
|
of minute frequency for the given sids. |
|
639
|
|
|
""" |
|
640
|
|
|
# get all the minutes for this window |
|
641
|
|
|
minutes_for_window = self.env.market_minute_window( |
|
642
|
|
|
end_dt, bar_count, step=-1)[::-1] |
|
643
|
|
|
|
|
644
|
|
|
first_trading_day = self._equity_minute_reader.first_trading_day |
|
645
|
|
|
|
|
646
|
|
|
# but then cut it down to only the minutes after |
|
647
|
|
|
# the first trading day. |
|
648
|
|
|
modified_minutes_for_window = minutes_for_window[ |
|
649
|
|
|
minutes_for_window.slice_indexer(first_trading_day)] |
|
650
|
|
|
|
|
651
|
|
|
modified_minutes_length = len(modified_minutes_for_window) |
|
652
|
|
|
|
|
653
|
|
|
if modified_minutes_length == 0: |
|
654
|
|
|
raise ValueError("Cannot calculate history window that ends" |
|
655
|
|
|
"before 2002-01-02 14:31 UTC!") |
|
656
|
|
|
|
|
657
|
|
|
data = [] |
|
658
|
|
|
bars_to_prepend = 0 |
|
659
|
|
|
nans_to_prepend = None |
|
660
|
|
|
|
|
661
|
|
|
if modified_minutes_length < bar_count: |
|
662
|
|
|
first_trading_date = first_trading_day.date() |
|
663
|
|
|
if modified_minutes_for_window[0].date() == first_trading_date: |
|
664
|
|
|
# the beginning of the window goes before our global trading |
|
665
|
|
|
# start date |
|
666
|
|
|
bars_to_prepend = bar_count - modified_minutes_length |
|
667
|
|
|
nans_to_prepend = np.repeat(np.nan, bars_to_prepend) |
|
668
|
|
|
|
|
669
|
|
|
if len(assets) == 0: |
|
670
|
|
|
return pd.DataFrame( |
|
671
|
|
|
None, |
|
672
|
|
|
index=modified_minutes_for_window, |
|
673
|
|
|
columns=None |
|
674
|
|
|
) |
|
675
|
|
|
|
|
676
|
|
|
for asset in assets: |
|
677
|
|
|
asset_minute_data = self._get_minute_window_for_asset( |
|
678
|
|
|
asset, |
|
679
|
|
|
field_to_use, |
|
680
|
|
|
modified_minutes_for_window |
|
681
|
|
|
) |
|
682
|
|
|
|
|
683
|
|
|
if bars_to_prepend != 0: |
|
684
|
|
|
asset_minute_data = np.insert(asset_minute_data, 0, |
|
685
|
|
|
nans_to_prepend) |
|
686
|
|
|
|
|
687
|
|
|
data.append(asset_minute_data) |
|
688
|
|
|
|
|
689
|
|
|
return pd.DataFrame( |
|
690
|
|
|
np.array(data).T, |
|
691
|
|
|
index=minutes_for_window, |
|
692
|
|
|
columns=assets |
|
693
|
|
|
) |
|
694
|
|
|
|
|
695
|
|
|
def get_history_window(self, assets, end_dt, bar_count, frequency, field, |
|
696
|
|
|
ffill=True): |
|
697
|
|
|
""" |
|
698
|
|
|
Public API method that returns a dataframe containing the requested |
|
699
|
|
|
history window. Data is fully adjusted. |
|
700
|
|
|
|
|
701
|
|
|
Parameters |
|
702
|
|
|
--------- |
|
703
|
|
|
assets : list of zipline.data.Asset objects |
|
704
|
|
|
The assets whose data is desired. |
|
705
|
|
|
|
|
706
|
|
|
bar_count: int |
|
707
|
|
|
The number of bars desired. |
|
708
|
|
|
|
|
709
|
|
|
frequency: string |
|
710
|
|
|
"1d" or "1m" |
|
711
|
|
|
|
|
712
|
|
|
field: string |
|
713
|
|
|
The desired field of the asset. |
|
714
|
|
|
|
|
715
|
|
|
ffill: boolean |
|
716
|
|
|
Forward-fill missing values. Only has effect if field |
|
717
|
|
|
is 'price'. |
|
718
|
|
|
|
|
719
|
|
|
Returns |
|
720
|
|
|
------- |
|
721
|
|
|
A dataframe containing the requested data. |
|
722
|
|
|
""" |
|
723
|
|
|
try: |
|
724
|
|
|
field_to_use = BASE_FIELDS[field] |
|
725
|
|
|
except KeyError: |
|
726
|
|
|
raise ValueError("Invalid history field: " + str(field)) |
|
727
|
|
|
|
|
728
|
|
|
# sanity check in case sids were passed in |
|
729
|
|
|
assets = [(self.env.asset_finder.retrieve_asset(asset) if |
|
730
|
|
|
isinstance(asset, int) else asset) for asset in assets] |
|
731
|
|
|
|
|
732
|
|
|
if frequency == "1d": |
|
733
|
|
|
df = self._get_history_daily_window(assets, end_dt, bar_count, |
|
734
|
|
|
field_to_use) |
|
735
|
|
|
elif frequency == "1m": |
|
736
|
|
|
df = self._get_history_minute_window(assets, end_dt, bar_count, |
|
737
|
|
|
field_to_use) |
|
738
|
|
|
else: |
|
739
|
|
|
raise ValueError("Invalid frequency: {0}".format(frequency)) |
|
740
|
|
|
|
|
741
|
|
|
# forward-fill if needed |
|
742
|
|
|
if field == "price" and ffill: |
|
743
|
|
|
df.fillna(method='ffill', inplace=True) |
|
744
|
|
|
|
|
745
|
|
|
return df |
|
746
|
|
|
|
|
747
|
|
|
def _get_minute_window_for_asset(self, asset, field, minutes_for_window): |
|
748
|
|
|
""" |
|
749
|
|
|
Internal method that gets a window of adjusted minute data for an asset |
|
750
|
|
|
and specified date range. Used to support the history API method for |
|
751
|
|
|
minute bars. |
|
752
|
|
|
|
|
753
|
|
|
Missing bars are filled with NaN. |
|
754
|
|
|
|
|
755
|
|
|
Parameters |
|
756
|
|
|
---------- |
|
757
|
|
|
asset : Asset |
|
758
|
|
|
The asset whose data is desired. |
|
759
|
|
|
|
|
760
|
|
|
field: string |
|
761
|
|
|
The specific field to return. "open", "high", "close_price", etc. |
|
762
|
|
|
|
|
763
|
|
|
minutes_for_window: pd.DateTimeIndex |
|
764
|
|
|
The list of minutes representing the desired window. Each minute |
|
765
|
|
|
is a pd.Timestamp. |
|
766
|
|
|
|
|
767
|
|
|
Returns |
|
768
|
|
|
------- |
|
769
|
|
|
A numpy array with requested values. |
|
770
|
|
|
""" |
|
771
|
|
|
if isinstance(asset, int): |
|
772
|
|
|
asset = self.env.asset_finder.retrieve_asset(asset) |
|
773
|
|
|
|
|
774
|
|
|
if isinstance(asset, Future): |
|
775
|
|
|
return self._get_minute_window_for_future(asset, field, |
|
776
|
|
|
minutes_for_window) |
|
777
|
|
|
else: |
|
778
|
|
|
return self._get_minute_window_for_equity(asset, field, |
|
779
|
|
|
minutes_for_window) |
|
780
|
|
|
|
|
781
|
|
|
def _get_minute_window_for_future(self, asset, field, minutes_for_window): |
|
782
|
|
|
# THIS IS TEMPORARY. For now, we are only exposing futures within |
|
783
|
|
|
# equity trading hours (9:30 am to 4pm, Eastern). The easiest way to |
|
784
|
|
|
# do this is to simply do a spot lookup for each desired minute. |
|
785
|
|
|
return_data = np.zeros(len(minutes_for_window), dtype=np.float64) |
|
786
|
|
|
for idx, minute in enumerate(minutes_for_window): |
|
787
|
|
|
return_data[idx] = \ |
|
788
|
|
|
self._get_minute_spot_value_future(asset, field, minute) |
|
789
|
|
|
|
|
790
|
|
|
# Note: an improvement could be to find the consecutive runs within |
|
791
|
|
|
# minutes_for_window, and use them to read the underlying ctable |
|
792
|
|
|
# more efficiently. |
|
793
|
|
|
|
|
794
|
|
|
# Once futures are on 24-hour clock, then we can just grab all the |
|
795
|
|
|
# requested minutes in one shot from the ctable. |
|
796
|
|
|
|
|
797
|
|
|
# no adjustments for futures, yay. |
|
798
|
|
|
return return_data |
|
799
|
|
|
|
|
800
|
|
|
def _get_minute_window_for_equity(self, asset, field, minutes_for_window): |
|
801
|
|
|
# each sid's minutes are stored in a bcolz file |
|
802
|
|
|
# the bcolz file has 390 bars per day, starting at 1/2/2002, regardless |
|
803
|
|
|
# of when the asset started trading and regardless of half days. |
|
804
|
|
|
# for a half day, the second half is filled with zeroes. |
|
805
|
|
|
|
|
806
|
|
|
# find the position of start_dt in the entire timeline, go back |
|
807
|
|
|
# bar_count bars, and that's the unadjusted data |
|
808
|
|
|
raw_data = self._equity_minute_reader._open_minute_file(field, asset) |
|
809
|
|
|
|
|
810
|
|
|
start_idx = max( |
|
811
|
|
|
self._equity_minute_reader._find_position_of_minute( |
|
812
|
|
|
minutes_for_window[0]), |
|
813
|
|
|
0) |
|
814
|
|
|
end_idx = self._equity_minute_reader._find_position_of_minute( |
|
815
|
|
|
minutes_for_window[-1]) + 1 |
|
816
|
|
|
|
|
817
|
|
|
if end_idx == 0: |
|
818
|
|
|
# No data to return for minute window. |
|
819
|
|
|
return np.full(len(minutes_for_window), np.nan) |
|
820
|
|
|
|
|
821
|
|
|
return_data = np.zeros(len(minutes_for_window), dtype=np.float64) |
|
822
|
|
|
|
|
823
|
|
|
data_to_copy = raw_data[start_idx:end_idx] |
|
824
|
|
|
|
|
825
|
|
|
num_minutes = len(minutes_for_window) |
|
826
|
|
|
|
|
827
|
|
|
# data_to_copy contains all the zeros (from 1pm to 4pm of an early |
|
828
|
|
|
# close). num_minutes is the number of actual trading minutes. if |
|
829
|
|
|
# these two have different lengths, that means that we need to trim |
|
830
|
|
|
# away data due to early closes. |
|
831
|
|
|
if len(data_to_copy) != num_minutes: |
|
832
|
|
|
# get a copy of the minutes in Eastern time, since we depend on |
|
833
|
|
|
# an early close being at 1pm Eastern. |
|
834
|
|
|
eastern_minutes = minutes_for_window.tz_convert("US/Eastern") |
|
835
|
|
|
|
|
836
|
|
|
# accumulate a list of indices of the last minute of an early |
|
837
|
|
|
# close day. For example, if data_to_copy starts at 12:55 pm, and |
|
838
|
|
|
# there are five minutes of real data before 180 zeroes, we would |
|
839
|
|
|
# put 5 into last_minute_idx_of_early_close_day, because the fifth |
|
840
|
|
|
# minute is the last "real" minute of the day. |
|
841
|
|
|
last_minute_idx_of_early_close_day = [] |
|
842
|
|
|
for minute_idx, minute_dt in enumerate(eastern_minutes): |
|
843
|
|
|
if minute_idx == (num_minutes - 1): |
|
844
|
|
|
break |
|
845
|
|
|
|
|
846
|
|
|
if minute_dt.hour == 13 and minute_dt.minute == 0: |
|
847
|
|
|
next_minute = eastern_minutes[minute_idx + 1] |
|
848
|
|
|
if next_minute.hour != 13: |
|
849
|
|
|
# minute_dt is the last minute of an early close day |
|
850
|
|
|
last_minute_idx_of_early_close_day.append(minute_idx) |
|
851
|
|
|
|
|
852
|
|
|
# spin through the list of early close markers, and use them to |
|
853
|
|
|
# chop off 180 minutes at a time from data_to_copy. |
|
854
|
|
|
for idx, early_close_minute_idx in \ |
|
855
|
|
|
enumerate(last_minute_idx_of_early_close_day): |
|
856
|
|
|
early_close_minute_idx -= (180 * idx) |
|
857
|
|
|
data_to_copy = np.delete( |
|
858
|
|
|
data_to_copy, |
|
859
|
|
|
range( |
|
860
|
|
|
early_close_minute_idx + 1, |
|
861
|
|
|
early_close_minute_idx + 181 |
|
862
|
|
|
) |
|
863
|
|
|
) |
|
864
|
|
|
|
|
865
|
|
|
return_data[0:len(data_to_copy)] = data_to_copy |
|
866
|
|
|
|
|
867
|
|
|
self._apply_all_adjustments( |
|
868
|
|
|
return_data, |
|
869
|
|
|
asset, |
|
870
|
|
|
minutes_for_window, |
|
871
|
|
|
field, |
|
872
|
|
|
self.MINUTE_PRICE_ADJUSTMENT_FACTOR |
|
873
|
|
|
) |
|
874
|
|
|
|
|
875
|
|
|
return return_data |
|
876
|
|
|
|
|
877
|
|
|
def _apply_all_adjustments(self, data, asset, dts, field, |
|
878
|
|
|
price_adj_factor=1.0): |
|
879
|
|
|
""" |
|
880
|
|
|
Internal method that applies all the necessary adjustments on the |
|
881
|
|
|
given data array. |
|
882
|
|
|
|
|
883
|
|
|
The adjustments are: |
|
884
|
|
|
- splits |
|
885
|
|
|
- if field != "volume": |
|
886
|
|
|
- mergers |
|
887
|
|
|
- dividends |
|
888
|
|
|
- * 0.001 |
|
889
|
|
|
- any zero fields replaced with NaN |
|
890
|
|
|
- all values rounded to 3 digits after the decimal point. |
|
891
|
|
|
|
|
892
|
|
|
Parameters |
|
893
|
|
|
---------- |
|
894
|
|
|
data : np.array |
|
895
|
|
|
The data to be adjusted. |
|
896
|
|
|
|
|
897
|
|
|
asset: Asset |
|
898
|
|
|
The asset whose data is being adjusted. |
|
899
|
|
|
|
|
900
|
|
|
dts: pd.DateTimeIndex |
|
901
|
|
|
The list of minutes or days representing the desired window. |
|
902
|
|
|
|
|
903
|
|
|
field: string |
|
904
|
|
|
The field whose values are in the data array. |
|
905
|
|
|
|
|
906
|
|
|
price_adj_factor: float |
|
907
|
|
|
Factor with which to adjust OHLC values. |
|
908
|
|
|
Returns |
|
909
|
|
|
------- |
|
910
|
|
|
None. The data array is modified in place. |
|
911
|
|
|
""" |
|
912
|
|
|
self._apply_adjustments_to_window( |
|
913
|
|
|
self._get_adjustment_list( |
|
914
|
|
|
asset, self._splits_dict, "SPLITS" |
|
915
|
|
|
), |
|
916
|
|
|
data, |
|
917
|
|
|
dts, |
|
918
|
|
|
field != 'volume' |
|
919
|
|
|
) |
|
920
|
|
|
|
|
921
|
|
|
if field != 'volume': |
|
922
|
|
|
self._apply_adjustments_to_window( |
|
923
|
|
|
self._get_adjustment_list( |
|
924
|
|
|
asset, self._mergers_dict, "MERGERS" |
|
925
|
|
|
), |
|
926
|
|
|
data, |
|
927
|
|
|
dts, |
|
928
|
|
|
True |
|
929
|
|
|
) |
|
930
|
|
|
|
|
931
|
|
|
self._apply_adjustments_to_window( |
|
932
|
|
|
self._get_adjustment_list( |
|
933
|
|
|
asset, self._dividends_dict, "DIVIDENDS" |
|
934
|
|
|
), |
|
935
|
|
|
data, |
|
936
|
|
|
dts, |
|
937
|
|
|
True |
|
938
|
|
|
) |
|
939
|
|
|
|
|
940
|
|
|
data *= price_adj_factor |
|
941
|
|
|
|
|
942
|
|
|
# if anything is zero, it's a missing bar, so replace it with NaN. |
|
943
|
|
|
# we only want to do this for non-volume fields, because a missing |
|
944
|
|
|
# volume should be 0. |
|
945
|
|
|
data[data == 0] = np.NaN |
|
946
|
|
|
|
|
947
|
|
|
np.around(data, 3, out=data) |
|
948
|
|
|
|
|
949
|
|
|
def _get_daily_window_for_sid(self, asset, field, days_in_window, |
|
950
|
|
|
extra_slot=True): |
|
951
|
|
|
""" |
|
952
|
|
|
Internal method that gets a window of adjusted daily data for a sid |
|
953
|
|
|
and specified date range. Used to support the history API method for |
|
954
|
|
|
daily bars. |
|
955
|
|
|
|
|
956
|
|
|
Parameters |
|
957
|
|
|
---------- |
|
958
|
|
|
asset : Asset |
|
959
|
|
|
The asset whose data is desired. |
|
960
|
|
|
|
|
961
|
|
|
start_dt: pandas.Timestamp |
|
962
|
|
|
The start of the desired window of data. |
|
963
|
|
|
|
|
964
|
|
|
bar_count: int |
|
965
|
|
|
The number of days of data to return. |
|
966
|
|
|
|
|
967
|
|
|
field: string |
|
968
|
|
|
The specific field to return. "open", "high", "close_price", etc. |
|
969
|
|
|
|
|
970
|
|
|
extra_slot: boolean |
|
971
|
|
|
Whether to allocate an extra slot in the returned numpy array. |
|
972
|
|
|
This extra slot will hold the data for the last partial day. It's |
|
973
|
|
|
much better to create it here than to create a copy of the array |
|
974
|
|
|
later just to add a slot. |
|
975
|
|
|
|
|
976
|
|
|
Returns |
|
977
|
|
|
------- |
|
978
|
|
|
A numpy array with requested values. Any missing slots filled with |
|
979
|
|
|
nan. |
|
980
|
|
|
|
|
981
|
|
|
""" |
|
982
|
|
|
bar_count = len(days_in_window) |
|
983
|
|
|
# create an np.array of size bar_count |
|
984
|
|
|
if extra_slot: |
|
985
|
|
|
return_array = np.zeros((bar_count + 1,)) |
|
986
|
|
|
else: |
|
987
|
|
|
return_array = np.zeros((bar_count,)) |
|
988
|
|
|
|
|
989
|
|
|
return_array[:] = np.NAN |
|
990
|
|
|
|
|
991
|
|
|
start_date = self._get_asset_start_date(asset) |
|
992
|
|
|
end_date = self._get_asset_end_date(asset) |
|
993
|
|
|
day_slice = days_in_window.slice_indexer(start_date, end_date) |
|
994
|
|
|
active_days = days_in_window[day_slice] |
|
995
|
|
|
|
|
996
|
|
|
if active_days.shape[0]: |
|
997
|
|
|
data = self._equity_daily_reader.history_window(field, |
|
998
|
|
|
active_days[0], |
|
999
|
|
|
active_days[-1], |
|
1000
|
|
|
asset) |
|
1001
|
|
|
return_array[day_slice] = data |
|
1002
|
|
|
self._apply_all_adjustments( |
|
1003
|
|
|
return_array, |
|
1004
|
|
|
asset, |
|
1005
|
|
|
active_days, |
|
1006
|
|
|
field, |
|
1007
|
|
|
) |
|
1008
|
|
|
|
|
1009
|
|
|
return return_array |
|
1010
|
|
|
|
|
1011
|
|
|
@staticmethod |
|
1012
|
|
|
def _apply_adjustments_to_window(adjustments_list, window_data, |
|
1013
|
|
|
dts_in_window, multiply): |
|
1014
|
|
|
if len(adjustments_list) == 0: |
|
1015
|
|
|
return |
|
1016
|
|
|
|
|
1017
|
|
|
# advance idx to the correct spot in the adjustments list, based on |
|
1018
|
|
|
# when the window starts |
|
1019
|
|
|
idx = 0 |
|
1020
|
|
|
|
|
1021
|
|
|
while idx < len(adjustments_list) and dts_in_window[0] >\ |
|
1022
|
|
|
adjustments_list[idx][0]: |
|
1023
|
|
|
idx += 1 |
|
1024
|
|
|
|
|
1025
|
|
|
# if we've advanced through all the adjustments, then there's nothing |
|
1026
|
|
|
# to do. |
|
1027
|
|
|
if idx == len(adjustments_list): |
|
1028
|
|
|
return |
|
1029
|
|
|
|
|
1030
|
|
|
while idx < len(adjustments_list): |
|
1031
|
|
|
adjustment_to_apply = adjustments_list[idx] |
|
1032
|
|
|
|
|
1033
|
|
|
if adjustment_to_apply[0] > dts_in_window[-1]: |
|
1034
|
|
|
break |
|
1035
|
|
|
|
|
1036
|
|
|
range_end = dts_in_window.searchsorted(adjustment_to_apply[0]) |
|
1037
|
|
|
if multiply: |
|
1038
|
|
|
window_data[0:range_end] *= adjustment_to_apply[1] |
|
1039
|
|
|
else: |
|
1040
|
|
|
window_data[0:range_end] /= adjustment_to_apply[1] |
|
1041
|
|
|
|
|
1042
|
|
|
idx += 1 |
|
1043
|
|
|
|
|
1044
|
|
|
def _get_adjustment_list(self, asset, adjustments_dict, table_name): |
|
1045
|
|
|
""" |
|
1046
|
|
|
Internal method that returns a list of adjustments for the given sid. |
|
1047
|
|
|
|
|
1048
|
|
|
Parameters |
|
1049
|
|
|
---------- |
|
1050
|
|
|
asset : Asset |
|
1051
|
|
|
The asset for which to return adjustments. |
|
1052
|
|
|
|
|
1053
|
|
|
adjustments_dict: dict |
|
1054
|
|
|
A dictionary of sid -> list that is used as a cache. |
|
1055
|
|
|
|
|
1056
|
|
|
table_name: string |
|
1057
|
|
|
The table that contains this data in the adjustments db. |
|
1058
|
|
|
|
|
1059
|
|
|
Returns |
|
1060
|
|
|
------- |
|
1061
|
|
|
adjustments: list |
|
1062
|
|
|
A list of [multiplier, pd.Timestamp], earliest first |
|
1063
|
|
|
|
|
1064
|
|
|
""" |
|
1065
|
|
|
if self._adjustment_reader is None: |
|
1066
|
|
|
return [] |
|
1067
|
|
|
|
|
1068
|
|
|
sid = int(asset) |
|
1069
|
|
|
|
|
1070
|
|
|
try: |
|
1071
|
|
|
adjustments = adjustments_dict[sid] |
|
1072
|
|
|
except KeyError: |
|
1073
|
|
|
adjustments = adjustments_dict[sid] = self._adjustment_reader.\ |
|
1074
|
|
|
get_adjustments_for_sid(table_name, sid) |
|
1075
|
|
|
|
|
1076
|
|
|
return adjustments |
|
1077
|
|
|
|
|
1078
|
|
|
def _check_is_currently_alive(self, asset, dt): |
|
1079
|
|
|
sid = int(asset) |
|
1080
|
|
|
|
|
1081
|
|
|
if sid not in self._asset_start_dates: |
|
1082
|
|
|
self._get_asset_start_date(asset) |
|
1083
|
|
|
|
|
1084
|
|
|
start_date = self._asset_start_dates[sid] |
|
1085
|
|
|
if self._asset_start_dates[sid] > dt: |
|
1086
|
|
|
raise NoTradeDataAvailableTooEarly( |
|
1087
|
|
|
sid=sid, |
|
1088
|
|
|
dt=normalize_date(dt), |
|
1089
|
|
|
start_dt=start_date |
|
1090
|
|
|
) |
|
1091
|
|
|
|
|
1092
|
|
|
end_date = self._asset_end_dates[sid] |
|
1093
|
|
|
if self._asset_end_dates[sid] < dt: |
|
1094
|
|
|
raise NoTradeDataAvailableTooLate( |
|
1095
|
|
|
sid=sid, |
|
1096
|
|
|
dt=normalize_date(dt), |
|
1097
|
|
|
end_dt=end_date |
|
1098
|
|
|
) |
|
1099
|
|
|
|
|
1100
|
|
|
def _get_asset_start_date(self, asset): |
|
1101
|
|
|
self._ensure_asset_dates(asset) |
|
1102
|
|
|
return self._asset_start_dates[asset] |
|
1103
|
|
|
|
|
1104
|
|
|
def _get_asset_end_date(self, asset): |
|
1105
|
|
|
self._ensure_asset_dates(asset) |
|
1106
|
|
|
return self._asset_end_dates[asset] |
|
1107
|
|
|
|
|
1108
|
|
|
def _ensure_asset_dates(self, asset): |
|
1109
|
|
|
sid = int(asset) |
|
1110
|
|
|
|
|
1111
|
|
|
if sid not in self._asset_start_dates: |
|
1112
|
|
|
self._asset_start_dates[sid] = asset.start_date |
|
1113
|
|
|
self._asset_end_dates[sid] = asset.end_date |
|
1114
|
|
|
|
|
1115
|
|
|
def get_splits(self, sids, dt): |
|
1116
|
|
|
""" |
|
1117
|
|
|
Returns any splits for the given sids and the given dt. |
|
1118
|
|
|
|
|
1119
|
|
|
Parameters |
|
1120
|
|
|
---------- |
|
1121
|
|
|
sids : list |
|
1122
|
|
|
Sids for which we want splits. |
|
1123
|
|
|
|
|
1124
|
|
|
dt: pd.Timestamp |
|
1125
|
|
|
The date for which we are checking for splits. Note: this is |
|
1126
|
|
|
expected to be midnight UTC. |
|
1127
|
|
|
|
|
1128
|
|
|
Returns |
|
1129
|
|
|
------- |
|
1130
|
|
|
list: List of splits, where each split is a (sid, ratio) tuple. |
|
1131
|
|
|
""" |
|
1132
|
|
|
if self._adjustment_reader is None or len(sids) == 0: |
|
1133
|
|
|
return {} |
|
1134
|
|
|
|
|
1135
|
|
|
# convert dt to # of seconds since epoch, because that's what we use |
|
1136
|
|
|
# in the adjustments db |
|
1137
|
|
|
seconds = int(dt.value / 1e9) |
|
1138
|
|
|
|
|
1139
|
|
|
splits = self._adjustment_reader.conn.execute( |
|
1140
|
|
|
"SELECT sid, ratio FROM SPLITS WHERE effective_date = ?", |
|
1141
|
|
|
(seconds,)).fetchall() |
|
1142
|
|
|
|
|
1143
|
|
|
sids_set = set(sids) |
|
1144
|
|
|
splits = [split for split in splits if split[0] in sids_set] |
|
1145
|
|
|
|
|
1146
|
|
|
return splits |
|
1147
|
|
|
|
|
1148
|
|
|
def get_stock_dividends(self, sid, trading_days): |
|
1149
|
|
|
""" |
|
1150
|
|
|
Returns all the stock dividends for a specific sid that occur |
|
1151
|
|
|
in the given trading range. |
|
1152
|
|
|
|
|
1153
|
|
|
Parameters |
|
1154
|
|
|
---------- |
|
1155
|
|
|
sid: int |
|
1156
|
|
|
The asset whose stock dividends should be returned. |
|
1157
|
|
|
|
|
1158
|
|
|
trading_days: pd.DatetimeIndex |
|
1159
|
|
|
The trading range. |
|
1160
|
|
|
|
|
1161
|
|
|
Returns |
|
1162
|
|
|
------- |
|
1163
|
|
|
list: A list of objects with all relevant attributes populated. |
|
1164
|
|
|
All timestamp fields are converted to pd.Timestamps. |
|
1165
|
|
|
""" |
|
1166
|
|
|
|
|
1167
|
|
|
if self._adjustment_reader is None: |
|
1168
|
|
|
return [] |
|
1169
|
|
|
|
|
1170
|
|
|
if len(trading_days) == 0: |
|
1171
|
|
|
return [] |
|
1172
|
|
|
|
|
1173
|
|
|
start_dt = trading_days[0].value / 1e9 |
|
1174
|
|
|
end_dt = trading_days[-1].value / 1e9 |
|
1175
|
|
|
|
|
1176
|
|
|
dividends = self._adjustment_reader.conn.execute( |
|
1177
|
|
|
"SELECT * FROM stock_dividend_payouts WHERE sid = ? AND " |
|
1178
|
|
|
"ex_date > ? AND pay_date < ?", (int(sid), start_dt, end_dt,)).\ |
|
1179
|
|
|
fetchall() |
|
1180
|
|
|
|
|
1181
|
|
|
dividend_info = [] |
|
1182
|
|
|
for dividend_tuple in dividends: |
|
1183
|
|
|
dividend_info.append({ |
|
1184
|
|
|
"declared_date": dividend_tuple[1], |
|
1185
|
|
|
"ex_date": pd.Timestamp(dividend_tuple[2], unit="s"), |
|
1186
|
|
|
"pay_date": pd.Timestamp(dividend_tuple[3], unit="s"), |
|
1187
|
|
|
"payment_sid": dividend_tuple[4], |
|
1188
|
|
|
"ratio": dividend_tuple[5], |
|
1189
|
|
|
"record_date": pd.Timestamp(dividend_tuple[6], unit="s"), |
|
1190
|
|
|
"sid": dividend_tuple[7] |
|
1191
|
|
|
}) |
|
1192
|
|
|
|
|
1193
|
|
|
return dividend_info |
|
1194
|
|
|
|
|
1195
|
|
|
def contains(self, asset, field): |
|
1196
|
|
|
return field in BASE_FIELDS or \ |
|
1197
|
|
|
(field in self._augmented_sources_map and |
|
1198
|
|
|
asset in self._augmented_sources_map[field]) |
|
1199
|
|
|
|
|
1200
|
|
|
def get_fetcher_assets(self, day): |
|
1201
|
|
|
""" |
|
1202
|
|
|
Returns a list of assets for the current date, as defined by the |
|
1203
|
|
|
fetcher data. |
|
1204
|
|
|
|
|
1205
|
|
|
Notes |
|
1206
|
|
|
----- |
|
1207
|
|
|
Data is forward-filled. If there is no fetcher data defined for day |
|
1208
|
|
|
N, we use day N-1's data (if available, otherwise we keep going back). |
|
1209
|
|
|
|
|
1210
|
|
|
Returns |
|
1211
|
|
|
------- |
|
1212
|
|
|
list: a list of Asset objects. |
|
1213
|
|
|
""" |
|
1214
|
|
|
# return a list of assets for the current date, as defined by the |
|
1215
|
|
|
# fetcher source |
|
1216
|
|
|
if self._extra_source_df is None: |
|
1217
|
|
|
return [] |
|
1218
|
|
|
|
|
1219
|
|
|
if day in self._extra_source_df.index: |
|
1220
|
|
|
date_to_use = day |
|
1221
|
|
|
else: |
|
1222
|
|
|
# current day isn't in the fetcher df, go back the last |
|
1223
|
|
|
# available day |
|
1224
|
|
|
idx = self._extra_source_df.index.searchsorted(day) |
|
1225
|
|
|
if idx == 0: |
|
1226
|
|
|
return [] |
|
1227
|
|
|
|
|
1228
|
|
|
date_to_use = self._extra_source_df.index[idx - 1] |
|
1229
|
|
|
|
|
1230
|
|
|
asset_list = self._extra_source_df.loc[date_to_use]["sid"] |
|
1231
|
|
|
|
|
1232
|
|
|
# make sure they're actually assets |
|
1233
|
|
|
asset_list = [asset for asset in asset_list |
|
1234
|
|
|
if isinstance(asset, Asset)] |
|
1235
|
|
|
|
|
1236
|
|
|
return asset_list |
|
1237
|
|
|
|
quantopian /
zipline
