def test_add_time_features_df():
"""Tests add_time_features_df"""
# create indexed input
date_list = pd.date_range(start=datetime.datetime(2019, 1, 1),
periods=100,
freq="H").tolist()
df0 = pd.DataFrame({TIME_COL: date_list}, index=date_list)
df = add_time_features_df(df=df0,
time_col=TIME_COL,
conti_year_origin=2018)
assert df["year"][0] == 2019
assert df.shape[0] == df0.shape[0]
hourly_data = generate_df_with_reg_for_tests(
freq="H",
periods=24 * 500,
train_start_date=datetime.datetime(2018, 7, 1),
conti_year_origin=2018)
cols = [TIME_COL, "regressor1", "regressor_bool", "regressor_categ"]
train_df = hourly_data["train_df"]
df = add_time_features_df(df=train_df[cols],
time_col=TIME_COL,
conti_year_origin=2018)
assert df["year"][0] == 2018
assert (df["dow_hr"][:3] == ["7_00", "7_01", "7_02"]).all()
assert df.shape[0] == train_df.shape[0]
def test_add_daily_events():
# generate events dictionary
countries = ["US", "India", "UK"]
event_df_dict = get_holidays(countries, year_start=2015, year_end=2025)
original_col_names = [
event_df_dict[country].columns[1] for country in countries
]
# generate temporal data
date_list = pd.date_range(start=dt(2019, 1, 1), periods=100,
freq="H").tolist()
df0 = pd.DataFrame({"ts": date_list})
df = add_time_features_df(df0, time_col="ts", conti_year_origin=2018)
df_with_events = add_daily_events(df=df,
event_df_dict=event_df_dict,
date_col="date")
assert df_with_events[f"{EVENT_PREFIX}_India"].values[
0] == "New Year's Day"
assert df_with_events[f"{EVENT_PREFIX}_US"].values[25] == ""
# makes sure the function does not modify the input
new_col_names = [
event_df_dict[country].columns[1] for country in countries
]
assert original_col_names == new_col_names
def test_get_custom_changepoints():
"""Tests get_custom_changepoints and get_changepoint_features"""
date_list = pd.date_range(start=dt(2019, 1, 1), periods=20,
freq="D").tolist()
time_col = "custom_time_col"
df0 = pd.DataFrame({time_col: date_list})
df = add_time_features_df(df0,
time_col="custom_time_col",
conti_year_origin=2018)
# dates as datetime
changepoint_dates = pd.to_datetime(
["2018-01-01", "2019-01-02-16", "2019-01-03", "2019-02-01"])
result = get_custom_changepoints_values(
df=df,
changepoint_dates=changepoint_dates,
time_col=time_col, # pd.Timestamp type
continuous_time_col=time_col # makes checking the result easier
)
# 2018-01-01 is mapped to 2019-01-01. Duplicates mapped to "2019-01-03" are merged
# Last requested changepoint is not found
assert np.all(result == pd.to_datetime(["2019-01-01", "2019-01-03"]))
# dates as strings
changepoint_dates = [
"2018-01-01", "2019-01-02-16", "2019-01-03", "2019-02-01"
]
result = get_custom_changepoints_values(
df=df,
changepoint_dates=changepoint_dates,
time_col="date", # datetime.date type
continuous_time_col=time_col)
assert np.all(result == pd.to_datetime(["2019-01-01", "2019-01-03"]))
# continuous_time_col different from time_col
# check using timestamps from last `result`
changepoints = get_custom_changepoints_values(
df=df,
changepoint_dates=changepoint_dates,
time_col=time_col,
continuous_time_col="ct1")
assert np.all(changepoints == df[df[time_col].isin(result)]["ct1"].values)
# generated features, using changepoints from above
changepoint_df = get_changepoint_features(
df,
changepoints,
continuous_time_col="ct1",
growth_func=lambda x: x,
)
assert changepoint_df.shape == (df.shape[0], len(result))
assert round(changepoint_df.iloc[15, 0], 3) == 0.041
assert round(changepoint_df.iloc[16, 1], 3) == 0.038
# no matching dates
changepoint_dates = ["2019-02-01"]
result = get_custom_changepoints_values(
df=df,
changepoint_dates=changepoint_dates,
time_col=time_col,
continuous_time_col=time_col)
assert result is None
# hourly data, provided changepoints at daily level
date_list = pd.date_range(start=dt(2019, 1, 1), periods=20 * 24,
freq="H").tolist()
time_col = "custom_time_col"
df0 = pd.DataFrame({time_col: date_list})
df = add_time_features_df(df0,
time_col="custom_time_col",
conti_year_origin=2018)
# dates as datetime
changepoint_dates = pd.to_datetime(
["2018-01-01", "2019-01-02-16", "2019-01-03", "2019-02-01"])
result = get_custom_changepoints_values(
df=df,
changepoint_dates=changepoint_dates,
time_col=time_col, # pd.Timestamp type
continuous_time_col=time_col # makes checking the result easier
)
# 2018-01-01 is mapped to 2019-01-01-00. Mapped to -00 if no hour provided
# Last requested changepoint is not found
assert np.all(result == pd.to_datetime(
["2019-01-01-00", "2019-01-02-16", "2019-01-03-00"]))
def test_get_evenly_spaced_changepoint_values():
df = pd.DataFrame({
"time_col":
np.arange(1, 11),
"ts":
pd.date_range(start="2020-01-01", periods=10, freq="AS")
})
changepoints = get_evenly_spaced_changepoints_values(df,
"time_col",
n_changepoints=3)
changepoint_dates = get_changepoint_dates_from_changepoints_dict(
changepoints_dict={
"method": "uniform",
"n_changepoints": 3
},
df=df,
time_col="ts")
# linear growth
changepoint_df = get_changepoint_features(
df,
changepoints,
continuous_time_col="time_col",
growth_func=None,
changepoint_dates=changepoint_dates)
expected = pd.DataFrame({
"changepoint0_2022_01_01_00": [0, 0, 0, 1, 2, 3, 4, 5, 6, 7],
"changepoint1_2025_01_01_00": [0, 0, 0, 0, 0, 0, 1, 2, 3, 4],
"changepoint2_2027_01_01_00": [0, 0, 0, 0, 0, 0, 0, 0, 1, 2]
})
assert changepoint_df.equals(expected)
# quadratic growth
changepoint_df = get_changepoint_features(
df,
changepoints,
continuous_time_col="time_col",
growth_func=lambda x: x**2,
changepoint_dates=changepoint_dates)
expected = pd.DataFrame({
"changepoint0_2022_01_01_00": [0, 0, 0, 1, 4, 9, 16, 25, 36, 49],
"changepoint1_2025_01_01_00": [0, 0, 0, 0, 0, 0, 1, 4, 9, 16],
"changepoint2_2027_01_01_00": [0, 0, 0, 0, 0, 0, 0, 0, 1, 4]
})
assert changepoint_df.equals(expected)
# real example
n_changepoints = 3
date_list = pd.date_range(start=dt(2019, 1, 1), periods=20,
freq="H").tolist()
df0 = pd.DataFrame({"ts": date_list})
df = add_time_features_df(df0, time_col="ts", conti_year_origin=2018)
changepoints = get_evenly_spaced_changepoints_values(
df, "ct1", n_changepoints=n_changepoints)
changepoint_dates = get_changepoint_dates_from_changepoints_dict(
changepoints_dict={
"method": "uniform",
"n_changepoints": 3
},
df=df,
time_col="ts")
changepoint_df = get_changepoint_features(
df,
changepoints,
continuous_time_col="ct1",
growth_func=lambda x: x,
changepoint_dates=changepoint_dates)
assert changepoint_df.shape == (df.shape[0], n_changepoints)
assert changepoint_df.iloc[15, 2] == 0.0
assert changepoint_df.iloc[16, 2] == 0.00011415525113989133
