|
1
|
|
|
import blaze as bz |
|
2
|
|
|
from datashape import istabular |
|
3
|
|
|
from odo import odo |
|
4
|
|
|
import pandas as pd |
|
5
|
|
|
from six import iteritems |
|
6
|
|
|
from toolz import valmap |
|
7
|
|
|
|
|
8
|
|
|
from .core import TS_FIELD_NAME, SID_FIELD_NAME |
|
9
|
|
|
from zipline.pipeline.loaders.base import PipelineLoader |
|
10
|
|
|
from zipline.pipeline.loaders.earnings import EarningsCalendarLoader |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
ANCMT_FIELD_NAME = 'announcement_date' |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
class BlazeEarningsCalendarLoader(PipelineLoader): |
|
17
|
|
|
"""A pipeline loader for the ``EarningsCalendar`` dataset that loads |
|
18
|
|
|
data from a blaze expression. |
|
19
|
|
|
|
|
20
|
|
|
Parameters |
|
21
|
|
|
---------- |
|
22
|
|
|
expr : Expr |
|
23
|
|
|
The expression representing the data to load. |
|
24
|
|
|
resources : any, optional |
|
25
|
|
|
The resources to use when computing ``expr``. If expr is already |
|
26
|
|
|
bound to resources this can be omitted. |
|
27
|
|
|
odo_kwargs : dict, optional |
|
28
|
|
|
Extra keyword arguments to pass to odo when executing the expression. |
|
29
|
|
|
|
|
30
|
|
|
Notes |
|
31
|
|
|
----- |
|
32
|
|
|
The expression should have a tabular dshape of:: |
|
33
|
|
|
|
|
34
|
|
|
Dim * {{ |
|
35
|
|
|
{SID_FIELD_NAME}: int64, |
|
36
|
|
|
{TS_FIELD_NAME}: datetime64, |
|
37
|
|
|
{ANCMT_FIELD_NAME}: datetime64, |
|
38
|
|
|
}} |
|
39
|
|
|
|
|
40
|
|
|
Where each row of the table is a record including the sid to identify the |
|
41
|
|
|
company, the timestamp where we learned about the announcement, and the |
|
42
|
|
|
date when the earnings will be announced. |
|
43
|
|
|
|
|
44
|
|
|
If the '{TS_FIELD_NAME}' field is not included it is assumed that we |
|
45
|
|
|
start the backtest with knowledge of all announcements. |
|
46
|
|
|
""" |
|
47
|
|
|
__doc__ = __doc__.format( |
|
48
|
|
|
TS_FIELD_NAME=TS_FIELD_NAME, |
|
49
|
|
|
SID_FIELD_NAME=SID_FIELD_NAME, |
|
50
|
|
|
ANCMT_FIELD_NAME=ANCMT_FIELD_NAME, |
|
51
|
|
|
) |
|
52
|
|
|
|
|
53
|
|
|
_expected_fields = frozenset({ |
|
54
|
|
|
TS_FIELD_NAME, |
|
55
|
|
|
SID_FIELD_NAME, |
|
56
|
|
|
ANCMT_FIELD_NAME, |
|
57
|
|
|
}) |
|
58
|
|
|
|
|
59
|
|
|
def __init__(self, |
|
60
|
|
|
expr, |
|
61
|
|
|
resources=None, |
|
62
|
|
|
compute_kwargs=None, |
|
63
|
|
|
odo_kwargs=None): |
|
64
|
|
|
dshape = expr.dshape |
|
65
|
|
|
|
|
66
|
|
|
if not istabular(dshape): |
|
67
|
|
|
raise ValueError( |
|
68
|
|
|
'expression dshape must be tabular, got: %s' % dshape, |
|
69
|
|
|
) |
|
70
|
|
|
|
|
71
|
|
|
expected_fields = self._expected_fields |
|
72
|
|
|
self._has_ts = has_ts = TS_FIELD_NAME in dshape.measure.dict |
|
73
|
|
|
if not has_ts: |
|
74
|
|
|
# This field is optional. |
|
75
|
|
|
expected_fields - {TS_FIELD_NAME} |
|
76
|
|
|
|
|
77
|
|
|
# bind the resources into the expression |
|
78
|
|
|
if resources is None: |
|
79
|
|
|
resources = {} |
|
80
|
|
|
elif not isinstance(resources, dict): |
|
81
|
|
|
leaves = expr._leaves() |
|
82
|
|
|
if len(leaves) != 1: |
|
83
|
|
|
raise ValueError('no data resources found') |
|
84
|
|
|
|
|
85
|
|
|
resources = {leaves[0]: resources} |
|
86
|
|
|
|
|
87
|
|
|
self._expr = expr[list(expected_fields)]._subs({ |
|
88
|
|
|
k: bz.Data(v, dshape=k.dshape) for k, v in iteritems(resources) |
|
89
|
|
|
}) |
|
90
|
|
|
self._odo_kwargs = odo_kwargs if odo_kwargs is not None else {} |
|
91
|
|
|
|
|
92
|
|
|
def load_adjusted_array(self, columns, dates, assets, mask): |
|
93
|
|
|
expr = self._expr |
|
94
|
|
|
filtered = expr[expr[TS_FIELD_NAME] <= dates[0]] |
|
95
|
|
|
lower = odo( |
|
96
|
|
|
bz.by( |
|
97
|
|
|
filtered[SID_FIELD_NAME], |
|
98
|
|
|
timestamp=filtered[TS_FIELD_NAME].max(), |
|
99
|
|
|
).timestamp.min(), |
|
100
|
|
|
pd.Timestamp, |
|
101
|
|
|
**self._odo_kwargs or {} |
|
102
|
|
|
) |
|
103
|
|
|
if lower is pd.NaT: |
|
104
|
|
|
# If there is no lower date, just query for data in the date |
|
105
|
|
|
# range. It must all be null anyways. |
|
106
|
|
|
lower = dates[0] |
|
107
|
|
|
|
|
108
|
|
|
raw = odo( |
|
109
|
|
|
expr[ |
|
110
|
|
|
(expr[TS_FIELD_NAME] >= lower) & |
|
111
|
|
|
(expr[TS_FIELD_NAME] <= dates[-1]) |
|
112
|
|
|
], |
|
113
|
|
|
pd.DataFrame, |
|
114
|
|
|
**self._odo_kwargs or {} |
|
115
|
|
|
) |
|
116
|
|
|
|
|
117
|
|
|
sids = raw.loc[:, SID_FIELD_NAME] |
|
118
|
|
|
raw.drop( |
|
119
|
|
|
sids[~(sids.isin(assets) | sids.notnull())].index, |
|
120
|
|
|
inplace=True |
|
121
|
|
|
) |
|
122
|
|
|
|
|
123
|
|
|
gb = raw.groupby(SID_FIELD_NAME) |
|
124
|
|
|
if self._has_ts: |
|
125
|
|
|
def mkseries(idx, raw_loc=raw.loc): |
|
126
|
|
|
vs = raw_loc[idx, [TS_FIELD_NAME, ANCMT_FIELD_NAME]].values |
|
127
|
|
|
return pd.Series( |
|
128
|
|
|
index=pd.DatetimeIndex(vs[:, 0]), |
|
129
|
|
|
data=vs[:, 1], |
|
130
|
|
|
) |
|
131
|
|
|
else: |
|
132
|
|
|
def mkseries(idx, raw_loc=raw.loc): |
|
133
|
|
|
return pd.DatetimeIndex(raw_loc[idx, ANCMT_FIELD_NAME]) |
|
134
|
|
|
|
|
135
|
|
|
return EarningsCalendarLoader( |
|
136
|
|
|
dates, |
|
137
|
|
|
valmap(mkseries, gb.groups), |
|
138
|
|
|
).load_adjusted_array(columns, dates, assets, mask) |
|
139
|
|
|
|
quantopian /
zipline
