def test_extrapolation(self, df, config, columns):
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df, columns.date)
assert output_df.loc[7, columns.data] == 316.2
assert math.isnan(output_df.loc[7, columns.category])
config.pop("category_imputation_method")
resampler = Resampler(params)
output_df = resampler.transform(df, columns.date)
assert output_df.loc[7, columns.data] == 316.2
assert math.isnan(output_df.loc[7, columns.category])
config["extrapolation_method"] = "none"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df, columns.date)
assert math.isnan(output_df.loc[6, columns.category])
category_results = np.array(output_df[columns.category].values,
dtype=np.float64)
assert np.isnan(category_results).all()
config["extrapolation_method"] = "interpolation"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df, columns.date)
assert np.round(output_df.loc[7, columns.data], 3) == 316.003
assert math.isnan(output_df.loc[7, columns.category])
def test_no_categorical_impute(self, df, config, columns):
config.pop("category_imputation_method")
params_no_impute = get_resampling_params(config)
resampler_no_impute = Resampler(params_no_impute)
no_impute_df = resampler_no_impute.transform(df, "Date")
assert pd.isnull(no_impute_df[columns.category].values).all()
config["category_imputation_method"] = "empty"
params_with_impute = get_resampling_params(config)
resampler_with_impute = Resampler(params_with_impute)
impute_df = resampler_with_impute.transform(df, "Date")
assert pd.isnull(impute_df[columns.category].values).all()
def test_empty_identifiers(self, df, params, config, datetime_column):
resampler = Resampler(params)
datetime_column = config.get('datetime_column')
output_df = resampler.transform(df,
datetime_column,
groupby_columns=[])
assert output_df.shape == (8, 4)
output_df = resampler.transform(df, datetime_column)
assert output_df.shape == (8, 4)
output_df = resampler.transform(df,
datetime_column,
groupby_columns=None)
assert output_df.shape == (8, 4)
def test_missing_categorical(self, missing_row_df, config, columns):
config["time_unit"] = "weeks"
config["time_step"] = 12
config["category_imputation_method"] = "clip"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(missing_row_df, columns.date)
assert np.all(output_df.categorical.values == "second")
config["category_imputation_method"] = "previous"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(missing_row_df, columns.date)
assert math.isnan(output_df.loc[0, columns.category])
assert np.all(output_df.loc[1:, columns.category].values == "second")
def test_three_identifiers(self, long_df_3, params, config,
datetime_column):
resampler = Resampler(params)
groupby_columns = ["country", "item", "store"]
datetime_column = config.get('datetime_column')
output_df = resampler.transform(long_df_3,
datetime_column,
groupby_columns=groupby_columns)
np.testing.assert_array_equal(
output_df[datetime_column].values,
pd.DatetimeIndex([
"1959-02-01",
"1959-02-15",
"1959-03-01",
"1959-02-01",
"1959-02-15",
"1959-03-01",
"1959-02-01",
"1959-02-15",
"1959-03-01",
"1959-02-01",
"1959-02-15",
"1959-03-01",
]))
def test_no_category_values(self, df, config, columns):
config["category_imputation_method"] = "previous"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df_first = resampler.transform(df, columns.date)
np.testing.assert_array_equal(
output_df_first.categorical.values,
np.array([
'first', 'first', 'first', 'first', 'first', 'second',
'second', 'second'
]))
config["category_imputation_method"] = "empty"
params = get_resampling_params(config)
resampler = Resampler(params)
output_df_empty = resampler.transform(df, columns.date)
assert math.isnan(output_df_empty.loc[0, columns.category])
def test_df_multiple_dates(self, df_multiple_dates, config, columns):
config["category_imputation_method"] = "previous"
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df_multiple_dates, columns.date)
assert pd.isnull(output_df.loc[1, "date2"])
def test_mode_filling(self, df3, config, columns):
config["category_imputation_method"] = "mode"
config["time_unit"] = "weeks"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df3, columns.date)
assert np.all(output_df.categorical.values == "second")
def test_microseconds(self, config, columns):
config["time_unit"] = "microseconds"
config["time_step"] = 3
params = get_resampling_params(config)
resampler = Resampler(params)
df_DST = get_df_DST("U", columns)
output_df = resampler.transform(df_DST, columns.date)
expected_dates = pd.DatetimeIndex(['2019-01-31T00:59:00.000000000', '2019-01-31T00:59:00.000003000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_long_df_different_sizes(self, long_df_different_sizes, params,
config, datetime_column):
resampler = Resampler(params)
groupby_columns = ["country"]
datetime_column = config.get('datetime_column')
output_df = resampler.transform(long_df_different_sizes,
datetime_column,
groupby_columns=groupby_columns)
assert output_df.shape == (12, 4)
def test_clip_filling(self, long_df, config, columns):
config["category_imputation_method"] = "clip"
config["time_unit"] = "weeks"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(long_df,
columns.date,
groupby_columns=["country"])
assert output_df.loc[3, columns.category] == "first"
def test_year(self, config, columns):
config["time_unit"] = "years"
params = get_resampling_params(config)
resampler = Resampler(params)
df = get_df("Y", columns)
output_df = resampler.transform(df, columns.date)
assert np.mean(output_df[columns.data]) == 316.19
expected_dates = pd.DatetimeIndex(['1959-12-31T00:00:00.000000000', '1961-12-31T00:00:00.000000000',
'1963-12-31T00:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_empty_filling(self, df2, config, columns):
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df2, columns.date)
assert output_df.loc[0, columns.category] == "first"
assert math.isnan(output_df.loc[1, columns.category])
assert math.isnan(output_df.loc[2, columns.category])
assert output_df.loc[6, columns.category] == "second"
assert math.isnan(output_df.loc[7, columns.category])
def test_next_filling_long_format(self, long_df, config, columns):
config["category_imputation_method"] = "next"
config["time_unit"] = "weeks"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(long_df,
columns.date,
groupby_columns=["country"])
assert math.isnan(output_df.loc[4, columns.category])
assert output_df.loc[3, columns.category] == "second"
def test_next_filling(self, df2, config, columns):
config["category_imputation_method"] = "next"
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df2, columns.date)
assert output_df.loc[0, columns.category] == "first"
assert output_df.loc[1, columns.category] == "first"
assert output_df.loc[3, columns.category] == "first"
assert output_df.loc[5, columns.category] == "second"
assert output_df.loc[9, columns.category] == "third"
def test_hours_DST(self, config, columns):
config["time_unit"] = "hours"
params = get_resampling_params(config)
resampler = Resampler(params)
df_DST = get_df_DST("4H", columns)
output_df = resampler.transform(df_DST, columns.date)
assert np.mean(output_df[columns.data]) == 316.33428571428567
expected_dates = pd.DatetimeIndex(['2019-01-31T01:00:00.000000000', '2019-01-31T03:00:00.000000000',
'2019-01-31T05:00:00.000000000', '2019-01-31T07:00:00.000000000',
'2019-01-31T09:00:00.000000000', '2019-01-31T11:00:00.000000000',
'2019-01-31T13:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_constant_value_filling(self, df2, config, columns):
config["category_imputation_method"] = "constant"
config["category_constant_value"] = "myvalue"
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df2, columns.date)
assert output_df.loc[0, columns.category] == "first"
assert output_df.loc[1, columns.category] == "myvalue"
assert output_df.loc[2, columns.category] == "myvalue"
assert output_df.loc[6, columns.category] == "second"
assert output_df.loc[7, columns.category] == "myvalue"
def test_long_format(self, long_df, params, config, datetime_column):
resampler = Resampler(params)
groupby_columns = ["country"]
datetime_column = config.get('datetime_column')
output_df = resampler.transform(long_df,
datetime_column,
groupby_columns=groupby_columns)
np.testing.assert_array_equal(
output_df[datetime_column].values,
pd.DatetimeIndex([
"1959-02-01", "1959-02-15", "1959-03-01", "1959-02-01",
"1959-02-15", "1959-03-01"
]))
def test_weeks_monday_end(self, config, columns):
config["time_unit"] = "weeks"
config["time_unit_end_of_week"] = "MON"
params = get_resampling_params(config)
resampler = Resampler(params)
df = get_df("M", columns)
output_df = resampler.transform(df, columns.date)
assert np.mean(output_df[columns.data]) == 316.36625000000004
expected_dates = pd.DatetimeIndex(['1959-02-02T00:00:00.000000000', '1959-02-16T00:00:00.000000000',
'1959-03-02T00:00:00.000000000', '1959-03-16T00:00:00.000000000',
'1959-03-30T00:00:00.000000000', '1959-04-13T00:00:00.000000000',
'1959-04-27T00:00:00.000000000', '1959-05-11T00:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_mode_filling_long_format(self, long_df_mode, config, columns):
config["category_imputation_method"] = "mode"
config["time_unit"] = "weeks"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(long_df_mode,
columns.date,
groupby_columns=["country"])
assert np.all(output_df.loc[output_df.country == 0,
columns.category].values == "first")
assert np.all(output_df.loc[output_df.country == 1,
columns.category].values == "fourth")
def test_nanoseconds(self, config, columns):
config["time_unit"] = "nanoseconds"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
df = get_df_DST("3N", columns)
output_df = resampler.transform(df, columns.date)
expected_dates = pd.DatetimeIndex(['2019-01-31T00:59:00.000000000', '2019-01-31T00:59:00.000000001',
'2019-01-31T00:59:00.000000002', '2019-01-31T00:59:00.000000003',
'2019-01-31T00:59:00.000000004', '2019-01-31T00:59:00.000000005',
'2019-01-31T00:59:00.000000006', '2019-01-31T00:59:00.000000007',
'2019-01-31T00:59:00.000000008', '2019-01-31T00:59:00.000000009'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_bool_column(self, bool_df, config, columns):
config["category_imputation_method"] = "previous"
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(bool_df, columns.date)
np.testing.assert_array_equal(
output_df.categorical.values,
np.array([
True, True, True, True, True, True, False, False, False, False,
False
]))
def test_seconds(self, config, columns):
config["time_unit"] = "seconds"
config["time_step"] = 30
params = get_resampling_params(config)
resampler = Resampler(params)
df_DST = get_df_DST("min", columns)
output_df = resampler.transform(df_DST, columns.date)
assert np.mean(output_df[columns.data]) == 316.28999999999996
expected_dates = pd.DatetimeIndex(['2019-01-31T00:59:00.000000000', '2019-01-31T00:59:30.000000000',
'2019-01-31T01:00:00.000000000', '2019-01-31T01:00:30.000000000',
'2019-01-31T01:01:00.000000000', '2019-01-31T01:01:30.000000000',
'2019-01-31T01:02:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_previous_filling(self, df2, config, columns):
config["category_imputation_method"] = "previous"
config["time_unit"] = "hours"
config["time_step"] = 12
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(df2, columns.date)
np.testing.assert_array_equal(
output_df.categorical.values,
np.array([
'first', 'first', 'first', 'first', 'first', 'first', 'second',
'second', 'second', 'second', 'third'
]))
def test_days(self, config, columns):
config["time_unit"] = "days"
params = get_resampling_params(config)
resampler = Resampler(params)
df = get_df("W-TUE", columns)
output_df = resampler.transform(df, columns.date)
assert np.mean(output_df[columns.data]) == 316.3254545454545
expected_dates = pd.DatetimeIndex(['1959-01-06T00:00:00.000000000', '1959-01-08T00:00:00.000000000',
'1959-01-10T00:00:00.000000000', '1959-01-12T00:00:00.000000000',
'1959-01-14T00:00:00.000000000', '1959-01-16T00:00:00.000000000',
'1959-01-18T00:00:00.000000000', '1959-01-20T00:00:00.000000000',
'1959-01-22T00:00:00.000000000', '1959-01-24T00:00:00.000000000',
'1959-01-26T00:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_days_DST(self, config, columns):
config["time_unit"] = "days"
params = get_resampling_params(config)
resampler = Resampler(params)
df_DST = get_df_DST("W-WED", columns)
output_df = resampler.transform(df_DST, columns.date)
assert np.mean(output_df[columns.data]) == 316.3072727272727
expected_dates = pd.DatetimeIndex(['2019-02-05T23:00:00.000000000', '2019-02-07T23:00:00.000000000',
'2019-02-09T23:00:00.000000000', '2019-02-11T23:00:00.000000000',
'2019-02-13T23:00:00.000000000', '2019-02-15T23:00:00.000000000',
'2019-02-17T23:00:00.000000000', '2019-02-19T23:00:00.000000000',
'2019-02-21T23:00:00.000000000', '2019-02-23T23:00:00.000000000',
'2019-02-25T23:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_mix_identifiers(self, long_df_4, params, config, datetime_column):
resampler = Resampler(params)
groupby_columns = ["country", "item", "store"]
datetime_column = config.get('datetime_column')
output_df = resampler.transform(long_df_4,
datetime_column,
groupby_columns=groupby_columns)
expected_dates = pd.DatetimeIndex([
'2020-02-02T00:00:00.000000000', '2020-02-16T00:00:00.000000000',
'2020-03-01T00:00:00.000000000', '2020-02-29T00:00:00.000000000',
'2020-01-31T00:00:00.000000000', '2020-02-02T00:00:00.000000000',
'2020-02-16T00:00:00.000000000', '2020-03-01T00:00:00.000000000',
'2020-02-02T00:00:00.000000000', '2020-02-16T00:00:00.000000000',
'2020-03-01T00:00:00.000000000'
])
np.testing.assert_array_equal(output_df[datetime_column].values,
expected_dates)
def test_month(self, config, columns):
config["time_unit"] = "months"
params = get_resampling_params(config)
resampler = Resampler(params)
df = get_df("Y", columns)
output_df = resampler.transform(df, columns.date)
assert np.mean(output_df[columns.data]) == 316.32550000000003
expected_dates = pd.DatetimeIndex(['1959-12-31T00:00:00.000000000', '1960-02-29T00:00:00.000000000',
'1960-04-30T00:00:00.000000000', '1960-06-30T00:00:00.000000000',
'1960-08-31T00:00:00.000000000', '1960-10-31T00:00:00.000000000',
'1960-12-31T00:00:00.000000000', '1961-02-28T00:00:00.000000000',
'1961-04-30T00:00:00.000000000', '1961-06-30T00:00:00.000000000',
'1961-08-31T00:00:00.000000000', '1961-10-31T00:00:00.000000000',
'1961-12-31T00:00:00.000000000', '1962-02-28T00:00:00.000000000',
'1962-04-30T00:00:00.000000000', '1962-06-30T00:00:00.000000000',
'1962-08-31T00:00:00.000000000', '1962-10-31T00:00:00.000000000',
'1962-12-31T00:00:00.000000000', '1963-02-28T00:00:00.000000000'])
np.testing.assert_array_equal(output_df[columns.date].values, expected_dates)
def test_previous_filling_long_format(self, long_df, config, columns):
config["category_imputation_method"] = "previous"
config["time_unit"] = "weeks"
config["time_step"] = 1
params = get_resampling_params(config)
resampler = Resampler(params)
output_df = resampler.transform(long_df,
columns.date,
groupby_columns=["country"])
expected_dates = pd.DatetimeIndex([
'1959-02-01T00:00:00.000000000', '1959-02-08T00:00:00.000000000',
'1959-02-15T00:00:00.000000000', '1959-02-22T00:00:00.000000000',
'1959-03-01T00:00:00.000000000', '1959-02-01T00:00:00.000000000',
'1959-02-08T00:00:00.000000000', '1959-02-15T00:00:00.000000000',
'1959-02-22T00:00:00.000000000', '1959-03-01T00:00:00.000000000'
])
np.testing.assert_array_equal(output_df[columns.date].values,
expected_dates)
expected_categorical = np.array([
'first', 'first', 'first', 'first', 'second', 'third', 'third',
'third', 'third', 'fourth'
])
np.testing.assert_array_equal(output_df.categorical.values,
expected_categorical)
from dataiku.customrecipe import get_recipe_config
from dku_timeseries import Resampler
from io_utils import get_input_output
from recipe_config_loading import check_and_get_groupby_columns, check_time_column_parameter, check_python_version, get_resampling_params
check_python_version()
# --- Setup
(input_dataset, output_dataset) = get_input_output()
recipe_config = get_recipe_config()
input_dataset_columns = [
column["name"] for column in input_dataset.read_schema()
]
check_time_column_parameter(recipe_config, input_dataset_columns)
groupby_columns = check_and_get_groupby_columns(recipe_config,
input_dataset_columns)
datetime_column = recipe_config.get('datetime_column')
params = get_resampling_params(recipe_config)
# --- Run
df = input_dataset.get_dataframe()
resampler = Resampler(params)
output_df = resampler.transform(df,
datetime_column,
groupby_columns=groupby_columns)
# --- Write output
output_dataset.write_with_schema(output_df)
