def get_seasonal_dummies(df):
"""Accepts a time-indexed df of hourly data, returns hourly and weekday dummies as a df
to passed as exogenous variables in a SARIMAX model"""
columns = df.columns
new_df = df.copy()
new_df['time'] = new_df.index
# create weekday dummy generator
wday_dumgen = ppc.DateFeaturizer(column_name='time',
with_day_of_month=False)
# since all have the same index, we can use any column in the df to generate the day_dums
_, wday_dums = wday_dumgen.fit_transform(new_df[columns[0]], new_df)
# drop the columns that aren't dummies
wday_dums = wday_dums[wday_dums.columns[-7:]]
# set the index for easy merging
wday_dums.set_index(new_df.index, inplace=True)
# create hourly dummy generator
hourly_dumgen = CalendarSeasonality('H', 'D')
# generate dummies
hourly_dums = hourly_dumgen.in_sample(new_df.index)
# merge results
full_dums = wday_dums.merge(hourly_dums, on='time')
return full_dums
def test_calendar_seasonal_period_q():
period = "Q"
index = pd.date_range("2000-01-01", freq="M", periods=600)
cs = CalendarSeasonality("M", period=period)
terms = cs.in_sample(index)
assert np.all(terms.sum(1) == 1.0)
for i in range(index.shape[0]):
assert terms.iloc[i, i % 3] == 1.0
def test_calendar_seasonality(time_index, forecast_index, freq_period):
freq, period = freq_period
cs = CalendarSeasonality(period, freq)
cs.in_sample(time_index)
steps = 83 if forecast_index is None else len(forecast_index)
cs.out_of_sample(steps, time_index, forecast_index)
assert isinstance(cs.period, str)
assert isinstance(cs.freq, str)
str(cs)
repr(cs)
hash(cs)
cs2 = CalendarSeasonality(period, freq)
assert cs == cs2
def test_calendar_seasonal_period_w():
period = "W"
index = pd.date_range("2000-01-03", freq="H", periods=600)
cs = CalendarSeasonality("H", period=period)
terms = cs.in_sample(index)
assert np.all(terms.sum(1) == 1.0)
for i in range(index.shape[0]):
assert terms.iloc[i, i % 168] == 1.0
index = pd.date_range("2000-01-03", freq="B", periods=600)
cs = CalendarSeasonality("B", period=period)
terms = cs.in_sample(index)
assert np.all(terms.sum(1) == 1.0)
for i in range(index.shape[0]):
assert terms.iloc[i, i % 5] == 1.0
index = pd.date_range("2000-01-03", freq="D", periods=600)
cs = CalendarSeasonality("D", period=period)
terms = cs.in_sample(index)
assert np.all(terms.sum(1) == 1.0)
for i in range(index.shape[0]):
assert terms.iloc[i, i % 7] == 1.0
def test_invalid_freq_period(time_index):
with pytest.raises(ValueError, match="The combination of freq="):
CalendarSeasonality("H", "A")
cs = CalendarSeasonality("B", "W")
with pytest.raises(ValueError, match="freq is B but index contains"):
cs.in_sample(pd.date_range("2000-1-1", periods=10, freq="D"))