def rolling_day_ahead_beliefs_about_time_slot_events(
request, time_slot_sensor: DBSensor):
"""Define multiple day-ahead beliefs about an ex post time slot event."""
source = (session.query(DBBeliefSource).filter(
DBBeliefSource.id == request.param).one_or_none())
beliefs = []
for i in range(1, 11): # ten events
# Recent belief (horizon 48 hours)
belief = DBTimedBelief(
sensor=time_slot_sensor,
source=source,
value=10 + i,
belief_time=datetime(2050, 1, 1, 10, tzinfo=utc) +
timedelta(hours=i),
event_start=datetime(2050, 1, 3, 10, tzinfo=utc) +
timedelta(hours=i),
)
session.add(belief)
beliefs.append(belief)
# Slightly older beliefs (belief_time an hour earlier)
belief = DBTimedBelief(
sensor=time_slot_sensor,
source=source,
value=100 + i,
belief_time=datetime(2050, 1, 1, 10, tzinfo=utc) +
timedelta(hours=i - 1),
event_start=datetime(2050, 1, 3, 10, tzinfo=utc) +
timedelta(hours=i),
)
session.add(belief)
beliefs.append(belief)
return beliefs
def test_upsample(time_slot_sensor,
rolling_day_ahead_beliefs_about_time_slot_events):
"""Upsample from 15 minutes to 5 minutes."""
new_resolution = timedelta(minutes=5)
belief_df = DBTimedBelief.query(
session=session,
sensor=time_slot_sensor,
belief_before=datetime(2100, 1, 1, 13, tzinfo=utc),
)
belief_df = belief_df.resample_events(new_resolution)
assert belief_df.sensor.event_resolution == timedelta(minutes=15)
assert belief_df.event_resolution == new_resolution
def day_ahead_belief_about_ex_post_time_slot_event(
ex_post_time_slot_sensor: DBSensor, test_source_a: DBBeliefSource):
"""Define day-ahead belief about an ex post time slot event."""
belief = DBTimedBelief(
source=test_source_a,
sensor=ex_post_time_slot_sensor,
value=10,
belief_time=datetime(2018, 1, 1, 10, tzinfo=utc),
event_start=datetime(2018, 1, 2, 22, 45, tzinfo=utc),
)
session.add(belief)
return belief
def test_query_belief_by_belief_time(
ex_post_time_slot_sensor: DBSensor,
day_ahead_belief_about_ex_post_time_slot_event: DBTimedBelief,
):
belief_df = DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_before=datetime(2018, 1, 1, 13, tzinfo=utc),
)
# By calling a pandas Series for its values we lose the timezone (a pandas bug still present in version 0.23.4)
# This next test warns us when it has been fixed (if it fails, just replace != with ==).
assert belief_df.knowledge_times.values[0] != datetime(
2018, 1, 1, 11, 0, tzinfo=utc)
# And this test is just a workaround to test what we wanted to test.
assert pd.Timestamp(belief_df.knowledge_times.values[0]) == pd.Timestamp(
datetime(2018, 1, 1, 11, 0))
# Just one belief found
assert len(belief_df.index) == 1
# No beliefs a year earlier
assert (len(
DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_before=datetime(2017, 1, 1, 10, tzinfo=utc),
).index) == 0)
# No beliefs 2 months later
assert (len(
DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_not_before=datetime(2018, 1, 3, 10, tzinfo=utc),
).index) == 0)
# One belief after 10am UTC
assert (len(
DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_not_before=datetime(2018, 1, 1, 10, tzinfo=utc),
).index) == 1)
# No beliefs an hour earlier
assert (len(
DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_before=datetime(2018, 1, 1, 9, tzinfo=utc),
).index) == 0)
# No beliefs after 1pm UTC
assert (len(
DBTimedBelief.query(
session=session,
sensor=ex_post_time_slot_sensor,
belief_not_before=datetime(2018, 1, 1, 13, tzinfo=utc),
).index) == 0)
def multiple_day_ahead_beliefs_about_ex_post_time_slot_event(
ex_post_time_slot_sensor: DBSensor, test_source_a: DBBeliefSource):
"""Define multiple day-ahead beliefs about an ex post time slot event."""
n = 10
event_start = datetime(2025, 1, 2, 22, 45, tzinfo=utc)
beliefs = []
for i in range(n):
belief = DBTimedBelief(
source=test_source_a,
sensor=ex_post_time_slot_sensor,
value=10 + i,
belief_time=ex_post_time_slot_sensor.knowledge_time(event_start) -
timedelta(hours=i + 1),
event_start=event_start,
)
session.add(belief)
beliefs.append(belief)
return beliefs
def test_query_rolling_horizon(
time_slot_sensor: DBSensor,
rolling_day_ahead_beliefs_about_time_slot_events):
# belief db selects all beliefs but one recent one (made exactly at 15h)
belief_df = DBTimedBelief.query(
session=session,
sensor=time_slot_sensor,
belief_before=datetime(2050, 1, 1, 15, tzinfo=utc),
)
rolling_df = belief_df.rolling_viewpoint(belief_horizon=timedelta(
hours=49)) # select only the five older beliefs
assert len(rolling_df) == 5 # 5 older (made at 10,11,12,13,14 o'clock)
assert (rolling_df["event_value"].values == np.arange(101, 106)).all()
rolling_df = belief_df.rolling_viewpoint(
belief_horizon=timedelta(hours=48)) # select mostly recent beliefs
assert (len(rolling_df) == 5
) # 4 early (made at 11,12,13,14 o'clock), 1 late ( at 14 o'clock)
assert (rolling_df["event_value"].values == [11, 12, 13, 14, 105]).all()
def test_query_fixed_horizon(time_slot_sensor: DBSensor,
rolling_day_ahead_beliefs_about_time_slot_events):
belief_time = datetime(2050, 1, 1, 11, tzinfo=utc)
df = DBTimedBelief.query(
session=session,
sensor=time_slot_sensor,
belief_before=datetime(2050, 1, 1, 15, tzinfo=utc),
source=[1, 2],
)
df2 = df.fixed_viewpoint(belief_time=belief_time)
assert len(df2) == 2
assert df2[df2.index.get_level_values("belief_time") > belief_time].empty
assert (df2["event_value"].values == np.array([11, 102])).all()
df3 = df.fixed_viewpoint(belief_time_window=(belief_time -
timedelta(minutes=1),
belief_time))
assert len(
df3) == 2 # The belief formed at 10 AM is now considered too old
assert (df3["event_value"].values == np.array([11, 102])).all()
def test_query_belief_history(
ex_post_time_slot_sensor: DBSensor,
multiple_day_ahead_beliefs_about_ex_post_time_slot_event: List[
DBTimedBelief],
):
df = DBTimedBelief.query(session=session, sensor=ex_post_time_slot_sensor)
event_start = datetime(2025, 1, 2, 22, 45, tzinfo=utc)
df2 = df.belief_history(event_start).sort_index(level="belief_time",
ascending=False)
assert len(df2) == 10
assert (df2["event_value"].values == np.arange(10, 20)).all()
df3 = df.belief_history(
event_start,
belief_time_window=(
datetime(2025, 1, 1, 7, tzinfo=utc),
datetime(2025, 1, 1, 9, tzinfo=utc),
),
)
assert len(df3) == 3
df4 = df.belief_history(
event_start,
belief_horizon_window=(timedelta(weeks=-10), timedelta(hours=2.5))
) # Only 2 beliefs were formed up to 2.5 hours before knowledge_time, and none after
assert len(df4) == 2