def test_compute_time_features_none():
index = pd.date_range("2017-01-01", periods=168, freq="H", tz="UTC")
with pytest.raises(ValueError):
compute_time_features(index,
hour_of_week=False,
day_of_week=False,
hour_of_day=False)
def create_caltrack_hourly_preliminary_design_matrix(meter_data, temperature_data):
"""A helper function which calls basic feature creation methods to create an
input suitable for use in the first step of creating a CalTRACK hourly model.
Parameters
----------
meter_data : :any:`pandas.DataFrame`
Hourly meter data in eemeter format.
temperature_data : :any:`pandas.Series`
Hourly temperature data in eemeter format.
Returns
-------
design_matrix : :any:`pandas.DataFrame`
A design matrix with meter_value, hour_of_week, hdd_50, and cdd_65 features.
"""
time_features = compute_time_features(
meter_data.index, hour_of_week=True, hour_of_day=False, day_of_week=False
)
temperature_features = compute_temperature_features(
meter_data.index,
temperature_data,
heating_balance_points=[50],
cooling_balance_points=[65],
degree_day_method="hourly",
)
design_matrix = merge_features(
[meter_data.value.to_frame("meter_value"), temperature_features, time_features]
)
return design_matrix
def test_compute_occupancy_feature(even_occupancy):
index = pd.date_range("2017-01-01", periods=1000, freq="H", tz="UTC")
time_features = compute_time_features(index, hour_of_week=True)
hour_of_week = time_features.hour_of_week
occupancy = compute_occupancy_feature(hour_of_week, even_occupancy)
assert occupancy.name == "occupancy"
assert occupancy.shape == (1000, )
assert occupancy.sum().sum() == 500
def test_compute_occupancy_feature_hour_of_week_has_nan(even_occupancy):
index = pd.date_range("2017-01-01", periods=72, freq="H", tz="UTC")
time_features = compute_time_features(index, hour_of_week=True)
hour_of_week = time_features.hour_of_week
hour_of_week.iloc[-1] = np.nan
occupancy = compute_occupancy_feature(hour_of_week, even_occupancy)
assert occupancy.name == "occupancy"
assert occupancy.shape == (72, )
assert occupancy.sum() == 36
def segmented_data():
index = pd.date_range(start="2017-01-01", periods=24, freq="H", tz="UTC")
time_features = compute_time_features(index)
segmented_data = pd.DataFrame(
{
"hour_of_week": time_features.hour_of_week,
"temperature_mean": np.linspace(0, 100, 24),
"meter_value": np.linspace(10, 70, 24),
"weight": np.ones((24, )),
},
index=index,
)
return segmented_data
def test_compute_occupancy_feature_with_nans(even_occupancy):
"""If there are less than 168 periods, the NaN at the end causes problems"""
index = pd.date_range("2017-01-01", periods=100, freq="H", tz="UTC")
time_features = compute_time_features(index, hour_of_week=True)
hour_of_week = time_features.hour_of_week
hour_of_week[-1] = np.nan
# comment out line below to see the error from not dropping na when
# calculationg _add_weights when there are less than 168 periods.
# TODO (ssuffian): Refactor so get_missing_hours_warnings propogates.
# right now, it will error if the dropna below isn't used.
hour_of_week.dropna(inplace=True)
occupancy = compute_occupancy_feature(hour_of_week, even_occupancy)
def occupancy_precursor(il_electricity_cdd_hdd_hourly):
meter_data = il_electricity_cdd_hdd_hourly["meter_data"]
temperature_data = il_electricity_cdd_hdd_hourly["temperature_data"]
time_features = compute_time_features(meter_data.index)
temperature_features = compute_temperature_features(
meter_data.index,
temperature_data,
heating_balance_points=[50],
cooling_balance_points=[65],
degree_day_method="hourly",
)
return merge_features([
meter_data.value.to_frame("meter_value"), temperature_features,
time_features
])
def test_compute_time_features_all():
index = pd.date_range("2017-01-01", periods=168, freq="H", tz="UTC")
features = compute_time_features(index)
assert list(
features.columns) == ["day_of_week", "hour_of_day", "hour_of_week"]
assert features.shape == (168, 3)
assert features.sum().sum() == 16464.0
with pytest.raises(TypeError): # categoricals
features.day_of_week.sum()
with pytest.raises(TypeError):
features.hour_of_day.sum()
with pytest.raises(TypeError):
features.hour_of_week.sum()
assert features.day_of_week.astype("float").sum() == sum(range(7)) * 24
assert features.hour_of_day.astype("float").sum() == sum(range(24)) * 7
assert features.hour_of_week.astype("float").sum() == sum(range(168))
assert features.index[0] == index[0]
assert features.index[-1] == index[-1]
def occupancy_precursor_only_nan(il_electricity_cdd_hdd_hourly):
meter_data = il_electricity_cdd_hdd_hourly["meter_data"]
meter_data = meter_data[datetime(2017, 1, 4):datetime(2017, 6, 1)]
meter_data.iloc[-1] = np.nan
# Simulates a segment where there is only a single nan value
temperature_data = il_electricity_cdd_hdd_hourly["temperature_data"]
time_features = compute_time_features(meter_data.index)
temperature_features = compute_temperature_features(
meter_data.index,
temperature_data,
heating_balance_points=[50],
cooling_balance_points=[65],
degree_day_method="hourly",
)
return merge_features([
meter_data.value.to_frame("meter_value"), temperature_features,
time_features
])
def create_caltrack_hourly_preliminary_design_matrix(meter_data,
temperature_data):
time_features = compute_time_features(meter_data.index,
hour_of_week=True,
hour_of_day=False,
day_of_week=False)
temperature_features = compute_temperature_features(
meter_data.index,
temperature_data,
heating_balance_points=[50],
cooling_balance_points=[65],
degree_day_method="hourly",
)
design_matrix = merge_features([
meter_data.value.to_frame("meter_value"), temperature_features,
time_features
])
return design_matrix
def test_compute_time_features_bad_freq():
index = pd.date_range("2017-01-01", periods=168, freq="D", tz="UTC")
with pytest.raises(ValueError):
compute_time_features(index)
def partial_hour_of_week_feature():
index = pd.date_range("2017-01-01", periods=84, freq="H", tz="UTC")
time_features = compute_time_features(index, hour_of_week=True)
hour_of_week_feature = time_features.hour_of_week
return hour_of_week_feature
