def select_sampling_indices(dates, frequency):
"""
Choose entries from ``dates`` to use for downsampling at ``frequency``.
Parameters
----------
dates : pd.DatetimeIndex
Dates from which to select sample choices.
{frequency}
Returns
-------
indices : np.array[int64]
An array condtaining indices of dates on which samples should be taken.
The resulting index will always include 0 as a sample index, and it
will include the first date of each subsequent year/quarter/month/week,
as determined by ``frequency``.
Notes
-----
This function assumes that ``dates`` does not have large gaps.
In particular, it assumes that the maximum distance between any two entries
in ``dates`` is never greater than a year, which we rely on because we use
``np.diff(dates.<frequency>)`` to find dates where the sampling
period has changed.
"""
if frequency == "week_start":
# dates.week is deprecated, uses dates.isocalendar().week
dates = dates.isocalendar()
return changed_locations(_dt_to_period[frequency](dates),
include_first=True)
def select_sampling_indices(dates, frequency):
"""
Choose entries from ``dates`` to use for downsampling at ``frequency``.
Parameters
----------
dates : pd.DatetimeIndex
Dates from which to select sample choices.
{frequency}
Returns
-------
indices : np.array[int64]
An array condtaining indices of dates on which samples should be taken.
The resulting index will always include 0 as a sample index, and it
will include the first date of each subsequent year/quarter/month/week,
as determined by ``frequency``.
Notes
-----
This function assumes that ``dates`` does not have large gaps.
In particular, it assumes that the maximum distance between any two entries
in ``dates`` is never greater than a year, which we rely on because we use
``np.diff(dates.<frequency>)`` to find dates where the sampling
period has changed.
"""
return changed_locations(
_dt_to_period[frequency](dates),
include_first=True
)
def compute(self, today, assets, out, sales, asof_date):
# 原始数据已经reindex全部季度，所以只需要单个资产的asof_date
idx = changed_locations(asof_date.T[0], include_first=True)[1:]
adj = quarterly_multiplier(asof_date[idx])
# 将各季度调整为年销售额，然后再平均
res = np.multiply(sales[idx], adj).mean()
# 收入单位为万元
out[:] = res * 10000