alexander/urban-meal-delivery
1
1
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity Actions

Add OrderHistory.make_real_time_time_series()

Browse source 
- the method slices out a real-time time series from the data within
  an `OrderHistory` object
This commit is contained in:
Alexander Hess 2021-01-09 17:30:00 +01:00
parent 5330ceb771
commit 100fac659a
Signed by: alexander
GPG key ID: 344EA5AB10D868E0
3 changed files with 257 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

90
src/urban_meal_delivery/forecasts/timify.py
View file

@ -294,3 +294,93 @@ class OrderHistory:
]
return training_df, frequency, actuals_df
def make_real_time_time_series( # noqa:WPS210
self, pixel_id: int, predict_at: dt.datetime, train_horizon: int,
) -> Tuple[pd.DataFrame, int, int]:
"""Slice a vertical real-time time series out of the `.totals`.
Create a time series covering `train_horizon` weeks that can be used
for training a forecasting model to predict the demand at `predict_at`.
For explanation of the terms "horizontal", "vertical", and "real-time"
in the context of time series, see section 3.2 in the following paper:
https://github.com/webartifex/urban-meal-delivery-demand-forecasting/blob/main/paper.pdf
Args:
pixel_id: pixel in which the time series is aggregated
predict_at: time step (i.e., "start_at") for which a prediction is made
train_horizon: weeks of historic data used to predict `predict_at`
Returns:
training time series, frequency, actual order count at `predict_at`
Raises:
LookupError: `pixel_id` is not in the `grid`
RuntimeError: desired time series slice is not entirely in `.totals`
"""
try:
intra_pixel = self.totals.loc[pixel_id]
except KeyError:
raise LookupError('The `pixel_id` is not in the `grid`') from None
if predict_at >= config.CUTOFF_DAY: # pragma: no cover
raise RuntimeError('Internal error: cannot predict beyond the given data')
# The first and last training day are just before the `predict_at` day
# and span exactly `train_horizon` weeks covering all times of the day,
# including times on the `predict_at` day that are earlier than `predict_at`.
first_train_day = predict_at.date() - dt.timedelta(weeks=train_horizon)
first_start_at = dt.datetime(
first_train_day.year,
first_train_day.month,
first_train_day.day,
config.SERVICE_START,
0,
)
# Predicting the first time step on the `predict_at` day is a corner case.
# Then, the previous day is indeed the `last_train_day`. Predicting any
# other time step implies that the `predict_at` day is the `last_train_day`.
# `last_train_time` is the last "start_at" before the one being predicted.
if predict_at.hour == config.SERVICE_START:
last_train_day = predict_at.date() - dt.timedelta(days=1)
last_train_time = dt.time(config.SERVICE_END, 0)
else:
last_train_day = predict_at.date()
last_train_time = predict_at.time()
last_start_at = dt.datetime(
last_train_day.year,
last_train_day.month,
last_train_day.day,
last_train_time.hour,
last_train_time.minute,
) - dt.timedelta(minutes=self._time_step)
# The frequency is the number of weekdays times the number of daily time steps.
frequency = 7 * self._n_daily_time_steps
# Take all the counts between `first_train_day` and `last_train_day`,
# including the ones on the `predict_at` day prior to `predict_at`.
training_df = intra_pixel.loc[
first_start_at:last_start_at # type: ignore
]
n_time_steps_on_predict_day = (
(
predict_at
- dt.datetime(
predict_at.year,
predict_at.month,
predict_at.day,
config.SERVICE_START,
0,
)
).seconds
// 60 # -> minutes
// self._time_step
)
if len(training_df) != frequency * train_horizon + n_time_steps_on_predict_day:
raise RuntimeError('Not enough historic data for `predict_day`')
actual_df = intra_pixel.loc[[predict_at]]
return training_df, frequency, actual_df

3
tests/config.py
View file

@ -8,6 +8,9 @@ from urban_meal_delivery import config
# The day on which most test cases take place.
YEAR, MONTH, DAY = 2016, 7, 1
# The hour when most test cases take place.
NOON = 12
# `START` and `END` constitute a 15-day time span.
# That implies a maximum `train_horizon` of `2` as that needs full 7-day weeks.
START = datetime.datetime(YEAR, MONTH, DAY, config.SERVICE_START, 0)

167
tests/forecasts/timify/test_make_time_series.py
View file

@ -1,4 +1,8 @@
"""Test the time series related code."""
"""Test the code generating time series with the order totals.
Unless otherwise noted, each `time_step` is 60 minutes long implying
12 time steps per day (i.e., we use `LONG_TIME_STEP` by default).
"""
# pylint:disable=no-self-use,unused-argument
import datetime
@ -63,7 +67,11 @@ def good_predict_at():
or `train_horizon=2` works.
"""
return datetime.datetime(
test_config.END.year, test_config.END.month, test_config.END.day, 12, 0,
test_config.END.year,
test_config.END.month,
test_config.END.day,
test_config.NOON,
0,
)
@ -76,7 +84,7 @@ def bad_predict_at():
"""
predict_day = test_config.END - datetime.timedelta(weeks=1, days=1)
return datetime.datetime(
predict_day.year, predict_day.month, predict_day.day, 12, 0,
predict_day.year, predict_day.month, predict_day.day, test_config.NOON, 0,
)
@ -282,3 +290,156 @@ class TestMakeVerticalTimeSeries:
predict_day=good_predict_at.date(),
train_horizon=999,
)
class TestMakeRealTimeTimeSeries:
"""Test the `OrderHistory.make_real_time_time_series()` method."""
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_wrong_pixel(self, order_history, good_predict_at, train_horizon):
"""A `pixel_id` that is not in the `grid`."""
with pytest.raises(LookupError):
order_history.make_real_time_time_series(
pixel_id=999_999,
predict_at=good_predict_at,
train_horizon=train_horizon,
)
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_time_series_are_dataframes(
self, order_history, good_pixel_id, good_predict_at, train_horizon,
):
"""The time series come in a one-column `pd.DataFrame`."""
result = order_history.make_real_time_time_series(
pixel_id=good_pixel_id,
predict_at=good_predict_at,
train_horizon=train_horizon,
)
training_df, _, actual_df = result
assert isinstance(training_df, pd.DataFrame)
assert training_df.columns == ['total_orders']
assert isinstance(actual_df, pd.DataFrame)
assert actual_df.columns == ['total_orders']
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_time_series_have_correct_length1(
self, order_history, good_pixel_id, good_predict_at, train_horizon,
):
"""The length of a training time series is the product of the ...
... weekly time steps (i.e., product of `7` and the number of daily time steps)
and the `train_horizon` in weeks; however, this assertion only holds if
we predict the first `time_step` of the day.
The time series with the actual order counts always holds `1` value.
"""
predict_at = datetime.datetime(
good_predict_at.year,
good_predict_at.month,
good_predict_at.day,
config.SERVICE_START,
0,
)
result = order_history.make_real_time_time_series(
pixel_id=good_pixel_id, predict_at=predict_at, train_horizon=train_horizon,
)
training_df, _, actual_df = result
n_daily_time_steps = (
60
* (config.SERVICE_END - config.SERVICE_START)
// test_config.LONG_TIME_STEP
)
assert len(training_df) == 7 * n_daily_time_steps * train_horizon
assert len(actual_df) == 1
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_time_series_have_correct_length2(
self, order_history, good_pixel_id, good_predict_at, train_horizon,
):
"""The length of a training time series is the product of the ...
... weekly time steps (i.e., product of `7` and the number of daily time steps)
and the `train_horizon` in weeks; however, this assertion only holds if
we predict the first `time_step` of the day. Predicting any other `time_step`
means that the training time series becomes longer by the number of time steps
before the one being predicted.
The time series with the actual order counts always holds `1` value.
"""
assert good_predict_at.hour == test_config.NOON
result = order_history.make_real_time_time_series(
pixel_id=good_pixel_id,
predict_at=good_predict_at,
train_horizon=train_horizon,
)
training_df, _, actual_df = result
n_daily_time_steps = (
60
* (config.SERVICE_END - config.SERVICE_START)
// test_config.LONG_TIME_STEP
)
n_time_steps_before = (
60 * (test_config.NOON - config.SERVICE_START) // test_config.LONG_TIME_STEP
)
assert (
len(training_df)
== 7 * n_daily_time_steps * train_horizon + n_time_steps_before
)
assert len(actual_df) == 1
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_frequency_is_number_number_of_weekly_time_steps(
self, order_history, good_pixel_id, good_predict_at, train_horizon,
):
"""The `frequency` is the number of weekly time steps."""
result = order_history.make_real_time_time_series(
pixel_id=good_pixel_id,
predict_at=good_predict_at,
train_horizon=train_horizon,
)
_, frequency, _ = result # noqa:WPS434
n_daily_time_steps = (
60
* (config.SERVICE_END - config.SERVICE_START)
// test_config.LONG_TIME_STEP
)
assert frequency == 7 * n_daily_time_steps
@pytest.mark.parametrize('train_horizon', test_config.TRAIN_HORIZONS)
def test_no_long_enough_history1(
self, order_history, good_pixel_id, bad_predict_at, train_horizon,
):
"""If the `predict_at` day is too early in the `START`-`END` horizon ...
... the history of order totals is not long enough.
"""
with pytest.raises(RuntimeError):
order_history.make_real_time_time_series(
pixel_id=good_pixel_id,
predict_at=bad_predict_at,
train_horizon=train_horizon,
)
def test_no_long_enough_history2(
self, order_history, good_pixel_id, good_predict_at,
):
"""If the `train_horizon` is longer than the `START`-`END` horizon ...
... the history of order totals can never be long enough.
"""
with pytest.raises(RuntimeError):
order_history.make_real_time_time_series(
pixel_id=good_pixel_id, predict_at=good_predict_at, train_horizon=999,
)