Datetime preprocessors

Appropiate handling of datetime features.

class DateLevel(Enum):
    """An enum representing different date levels."""

    YEAR = "year"
    QUARTER = "quarter"
    MONTH = "month"
    DAY = "day"
    HOUR = "hour"
    MINUTE = "minute"
    SECOND = "second"
    MICROSECOND = "microsecond"
    NANOSECOND = "nanosecond"
    WEEKDAY = "weekday"
    DAYOFYEAR = "dayofyear"
    DAYSINMONTH = "daysinmonth"

---

source

DateLevel

 DateLevel (value, names=None, module=None, qualname=None, type=None,
            start=1)

An enum representing different date levels.

---

source

DatetimeEncoder

 DatetimeEncoder (levels:Optional[Sequence[DateLevel]]=None,
                  fmt:Optional[str]=None)

An encoder for datetime columns that outputs integer features

levels is a list of DateLevel that define which date features to extract, i.e, [DateLevel.HOUR, DateLevel.MINUTE] will extract hours and minutes. If left to the default None, all available features will be extracted initially, but zero variance features will be dropped (for example, because the dates don’t have seconds).

	Type	Default	Details
levels	Optional[SequenceDateLevel]	None	Date features to extract.
fmt	Optional[str]	None	Date format for string conversion if inputs are note datetime-like objects. Follows standard Pandas/stdlib formatting, or example, ‘%Y-%m-%d %H:%M:%S’.

---

source

DatetimeEncoder.fit

 DatetimeEncoder.fit
                      (X:Union[pandas.core.frame.DataFrame,numpy.ndarray,L
                      ist], y=None)

Fit the DatetimeEncoder.

	Type	Default	Details
X	Union[pd.DataFrame, np.ndarray, List]		Datetime-like features..
y	NoneType	None	Unused.
Returns	DatetimeEncoder		Fitted DatetimeEncoder.

After fitting, the categories of each feature are held in the categories_ attribute.

---

source

DatetimeEncoder.transform

 DatetimeEncoder.transform
                            (X:Union[pandas.core.frame.DataFrame,numpy.nda
                            rray,List])

Apply transformation. Will ignore zero variance features seen during DatetimeEncoder.fit.

While this transformer is generally stateless, during DatetimeEncoder.fit it checks whether any of the extracted features have zero variance (only one unique value) and sets those levels to be ignored during DatetimeEncoder.transform.

	Type	Details
X	Union[pd.DataFrame, np.ndarray, List]	The data to encode.
Returns	np.ndarray	Transformed input.

---

TransformerMixin.fit_transform

 TransformerMixin.fit_transform (X, y=None, **fit_params)

Fit to data, then transform it.

Fits transformer to X and y with optional parameters fit_params and returns a transformed version of X.

	Type	Default	Details
X	array-like of shape (n_samples, n_features)		Input samples.
y	NoneType	None	Target values (None for unsupervised transformations).
fit_params
Returns	ndarray array of shape (n_samples, n_features_new)		Transformed array.

import pandas as pd

X = pd.DataFrame(
    {
        "hours": pd.date_range(start="2022-01-01", freq="H", periods=25),
        "days": pd.date_range(start="2022-01-01", freq="D", periods=25),
    }
)

encoder = DatetimeEncoder()
pd.DataFrame(encoder.fit_transform(X), columns=encoder.get_feature_names_out()).head()

	hours_day	hours_hour	hours_weekday	hours_dayofyear	days_day	days_weekday	days_dayofyear
0	1	0	5	1	1	5	1
1	1	1	5	1	2	6	2
2	1	2	5	1	3	0	3
3	1	3	5	1	4	1	4
4	1	4	5	1	5	2	5

Dates can be strings as well, but datetimes and strings cannot be combined.

date_format = "%Y-%m-%d"
X = pd.DataFrame(
    {
        "days": pd.date_range(start="2022-01-01", freq="D", periods=25).strftime(
            date_format
        ),
        "quarters": pd.date_range(start="2023-01-01", freq="Q", periods=25).strftime(
            date_format
        ),
    }
)

encoder = DatetimeEncoder(fmt=date_format)
pd.DataFrame(encoder.fit_transform(X), columns=encoder.get_feature_names_out()).head()

	days_day	days_weekday	days_dayofyear	quarters_year	quarters_quarter	quarters_month	quarters_day	quarters_weekday	quarters_dayofyear	quarters_daysinmonth
0	1	5	1	2023	1	3	31	4	90	31
1	2	6	2	2023	2	6	30	4	181	30
2	3	0	3	2023	3	9	30	5	273	30
3	4	1	4	2023	4	12	31	6	365	31
4	5	2	5	2024	1	3	31	6	91	31

Date levels may be chosen.

encoder = DatetimeEncoder(
    levels=[DateLevel.DAY, DateLevel.HOUR, DateLevel.MONTH], fmt=date_format
)
pd.DataFrame(encoder.fit_transform(X), columns=encoder.get_feature_names_out()).head()

	days_day	quarters_day	quarters_month
0	1	31	3
1	2	30	6
2	3	30	9
3	4	31	12
4	5	31	3