empml.transformers

Object	Description
Identity	Pass-through transformer.
AvgFeatures	Compute mean across multiple features row-wise.
MaxFeatures	Compute max across multiple features row-wise.
MinFeatures	Compute min across multiple features row-wise.
StdFeatures	Compute standard deviation across multiple features row-wise.
MedianFeatures	Compute median across multiple features row-wise.
ModuleFeatures	Compute Euclidean norm (module) of two features.
InteractionFeatures	Create pairwise multiplication features from feature pairs.
MeanTargetEncoder	Target encoding using mean of target variable.
StdTargetEncoder	Target encoding using standard deviation of target variable.
MaxTargetEncoder	Target encoding using max of target variable.
MinTargetEncoder	Target encoding using min of target variable.
MedianTargetEncoder	Target encoding using median of target variable.
KurtTargetEncoder	Target encoding using kurtosis of target variable.
SkewTargetEncoder	Target encoding using skewness of target variable.
OrdinalEncoder	Encode categorical features as ordinal integers.
DummyEncoder	One-hot encode categorical features.
FrequencyEncoder	Encode categorical features by their frequency or proportion.
StandardScaler	Standardize features by removing the mean and scaling to unit variance.
MinMaxScaler	Scale features to [0, 1] range using min-max normalization.
RobustScaler	Scale features using median and interquartile range (IQR).
Log1pFeatures	Apply log(1+x) transformation.
Expm1Features	Apply exp(x-1) transformation.
PowerFeatures	Apply power transformation.
InverseFeatures	Apply inverse transformation (1/x).
QuantileBinning	Discretize continuous features into quantile-based bins.
RankFeatures	Convert features to percentile rank based on training distribution.
SimpleImputer	Impute missing values using mean or median.
FillNulls	Fill null and NaN values with a constant.
GenerateLags	Generate lagged features for time series data.
KMeansCluster	Assign cluster labels using KMeans on selected features.
PCATransformer	Reduce dimensionality using Principal Component Analysis.

Identity

Pass-through transformer that returns data unchanged. No parameters required.

AvgFeatures

Compute mean across multiple features row-wise.

Methods

def __init__(self, features: list[str], new_feature: str):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to average.
    new_feature : str
        Name of the output column.
    """
    pass

MaxFeatures

Compute max across multiple features row-wise.

Methods

def __init__(self, features: list[str], new_feature: str):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to compute max over.
    new_feature : str
        Name of the output column.
    """
    pass

MinFeatures

Compute min across multiple features row-wise.

Methods

def __init__(self, features: list[str], new_feature: str):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to compute min over.
    new_feature : str
        Name of the output column.
    """
    pass

StdFeatures

Compute standard deviation across multiple features row-wise.

Methods

def __init__(self, features: list[str], new_feature: str):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to compute std over.
    new_feature : str
        Name of the output column.
    """
    pass

MedianFeatures

Compute median across multiple features row-wise.

Methods

def __init__(self, features: list[str], new_feature: str):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to compute median over.
    new_feature : str
        Name of the output column.
    """
    pass

ModuleFeatures

Compute Euclidean norm (module) of two features: sqrt(f1^2 + f2^2).

Methods

def __init__(self, features: tuple[str, str], new_feature: str):
    """
    Parameters:
    -----------
    features : tuple[str, str]
        Tuple of two column names.
    new_feature : str
        Name of the output column.
    """
    pass

InteractionFeatures

Create pairwise multiplication features from feature pairs. For each pair (f1, f2), creates a column f1 * f2.

Methods

def __init__(self, feature_pairs: list[tuple[str, str]], separator: str = '_x_'):
    """
    Parameters:
    -----------
    feature_pairs : list[tuple[str, str]]
        List of (col1, col2) tuples to multiply.
    separator : str
        String between feature names in the output column name (default: '_x_').
        Output column name: '{col1}{separator}{col2}'.
    """
    pass

MeanTargetEncoder

Encode categorical features with the mean of a target variable. Unseen categories during transform are filled with the global mean.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'mean_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    """
    Parameters:
    -----------
    features : list[str]
        Categorical columns to encode.
    encoder_col : str
        Target column to aggregate.
    prefix : str
        Prefix for encoded column names (default: 'mean_').
    suffix : str
        Suffix for encoded column names (default: '_encoded').
    replace_original : bool
        If True, drop original columns and use their names
        for encoded columns, ignoring prefix/suffix (default: False).
    """
    pass

StdTargetEncoder

Encode categorical features with the standard deviation of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'std_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

MaxTargetEncoder

Encode categorical features with the max of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'max_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

MinTargetEncoder

Encode categorical features with the min of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'min_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

MedianTargetEncoder

Encode categorical features with the median of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'median_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

KurtTargetEncoder

Encode categorical features with the kurtosis of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'kurt_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

SkewTargetEncoder

Encode categorical features with the skewness of a target variable. Same interface as MeanTargetEncoder.

Methods

def __init__(
    self,
    features: list[str],
    encoder_col: str,
    prefix: str = 'skew_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    pass

OrdinalEncoder

Encode categorical features with ordinal integers based on sorted order. Null values are encoded as -99, unknown categories (not seen during fit) as -9999.

Methods

def __init__(
    self,
    features: list[str],
    suffix: str = '_ordinal_encoded',
    replace_original: bool = False
):
    """
    Parameters:
    -----------
    features : list[str]
        Categorical columns to encode.
    suffix : str
        Suffix for encoded column names (default: '_ordinal_encoded').
    replace_original : bool
        If True, drop original columns and use their names
        for encoded columns, ignoring suffix (default: False).
    """
    pass

DummyEncoder

One-hot encode categorical features. Creates binary columns for each category, plus dedicated columns for null and unknown values.

Methods

def __init__(self, features: list[str]):
    """
    Parameters:
    -----------
    features : list[str]
        Categorical columns to encode.
    """
    pass

FrequencyEncoder

Encode categorical features by their frequency (count) or proportion. Unseen categories during transform are filled with 0.

Methods

def __init__(
    self,
    features: list[str],
    normalize: bool = True,
    prefix: str = 'freq_',
    suffix: str = '_encoded',
    replace_original: bool = False
):
    """
    Parameters:
    -----------
    features : list[str]
        Categorical columns to encode.
    normalize : bool
        If True, encode as proportion (0-1); if False, as raw count (default: True).
    prefix : str
        Prefix for encoded column names (default: 'freq_').
    suffix : str
        Suffix for encoded column names (default: '_encoded').
    replace_original : bool
        If True, drop original columns and use their names
        for encoded columns, ignoring prefix/suffix (default: False).
    """
    pass

StandardScaler

Standardize features by removing the mean and scaling to unit variance (z-score normalization). Handles zero standard deviation by returning 0.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to standardize.
    suffix : str
        Suffix for scaled column names (default: '' overwrites original).
    """
    pass

MinMaxScaler

Scale features to [0, 1] range using min-max normalization. Handles zero range by returning 0.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to scale.
    suffix : str
        Suffix for scaled column names (default: '' overwrites original).
    """
    pass

RobustScaler

Scale features using median and interquartile range (IQR): (x - median) / (Q75 - Q25). Less sensitive to outliers than StandardScaler. Handles zero IQR by returning 0.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to scale.
    suffix : str
        Suffix for scaled column names (default: '' overwrites original).
    """
    pass

Log1pFeatures

Apply log(1+x) transformation to features.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to transform.
    suffix : str
        Suffix for output column names (default: '' overwrites original).
    """
    pass

Expm1Features

Apply exp(x-1) transformation to features.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to transform.
    suffix : str
        Suffix for output column names (default: '' overwrites original).
    """
    pass

PowerFeatures

Raise features to a specified power.

Methods

def __init__(self, features: list[str], suffix: str = '', power: float = 2):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to transform.
    suffix : str
        Suffix for output column names (default: '' overwrites original).
    power : float
        Exponent for power transformation (default: 2).
    """
    pass

InverseFeatures

Apply inverse (1/x) transformation to features.

Methods

def __init__(self, features: list[str], suffix: str = ''):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to transform.
    suffix : str
        Suffix for output column names (default: '' overwrites original).
    """
    pass

QuantileBinning

Discretize continuous features into quantile-based bins. During fit, computes quantile bin edges. During transform, assigns bin indices (0 to num_bins-1).

Methods

def __init__(
    self,
    features: list[str],
    num_bins: int = 10,
    suffix: str = '_qbin',
    labels: list[str] | None = None
):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to discretize.
    num_bins : int
        Number of quantile bins (default: 10).
    suffix : str
        Suffix for binned column names (default: '_qbin').
    labels : list[str] | None
        Optional string labels for bins; length must equal num_bins.

    Raises:
    -------
    ValueError
        If labels length does not match num_bins.
    """
    pass

RankFeatures

Convert features to percentile rank [0, 1] based on the training distribution.

Methods

def __init__(
    self,
    features: list[str],
    suffix: str = '_rank',
    method: Literal['average', 'min', 'max', 'dense'] = 'average'
):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to rank.
    suffix : str
        Suffix for ranked column names (default: '_rank').
    method : str
        Ranking method (default: 'average'):
        - 'average': average of min and max rank positions
        - 'min': lowest rank position
        - 'max': highest rank position
        - 'dense': like 'min' but ranks always increase by 1
    """
    pass

SimpleImputer

Impute missing values (null and NaN) using mean or median strategy.

Methods

def __init__(self, features: list[str], strategy: str = 'mean'):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to impute.
    strategy : str
        Imputation strategy: 'mean' or 'median' (default: 'mean').

    Raises:
    -------
    ValueError
        If strategy is not 'mean' or 'median'.
    """
    pass

FillNulls

Fill null and NaN values with a constant.

Methods

def __init__(self, features: list[str], value: float = -9999):
    """
    Parameters:
    -----------
    features : list[str]
        Columns to fill.
    value : float
        Constant to use for filling (default: -9999).
    """
    pass

GenerateLags

Generate lagged features for time series data by joining shifted dates.

Methods

def __init__(
    self,
    ts_index: str,
    date_col: str,
    lag_col: str,
    lag_frequency: str = 'days',
    lag_min: int = 1,
    lag_max: int = 1,
    lag_step: int = 1
):
    """
    Parameters:
    -----------
    ts_index : str
        Column for time series identifier (e.g., entity ID).
    date_col : str
        Date/time column (must be Polars Date or Datetime type).
    lag_col : str
        Column to lag.
    lag_frequency : str
        Time unit for lags (default: 'days').
        Options: 'weeks', 'days', 'hours', 'minutes', 'seconds',
        'milliseconds', 'microseconds', 'nanoseconds'.
    lag_min : int
        Minimum lag period (default: 1).
    lag_max : int
        Maximum lag period (default: 1).
    lag_step : int
        Step size between lags (default: 1).

    Raises:
    -------
    ValueError
        If lag_frequency is not a valid time unit.
    """
    pass

KMeansCluster

Assign cluster labels using KMeans on selected features. Uses scikit-learn's KMeans internally. Missing values are imputed with column means before clustering.

Methods

def __init__(
    self,
    features: list[str],
    num_clusters: int = 8,
    new_feature: str = 'kmeans_cluster',
    random_state: int = 42
):
    """
    Parameters:
    -----------
    features : list[str]
        Numeric columns to use for clustering.
    num_clusters : int
        Number of clusters (default: 8).
    new_feature : str
        Name of the output cluster column (default: 'kmeans_cluster').
    random_state : int
        Random seed for reproducibility (default: 42).
    """
    pass

PCATransformer

Reduce dimensionality using Principal Component Analysis. Uses scikit-learn's PCA internally. Missing values are imputed with column means before PCA.

Methods

def __init__(
    self,
    features: list[str],
    n_components: int = 2,
    prefix: str = 'pc_'
):
    """
    Parameters:
    -----------
    features : list[str]
        Numeric columns to use for PCA.
    n_components : int
        Number of principal components to keep (default: 2).
    prefix : str
        Prefix for output column names (default: 'pc_').
        Columns are named '{prefix}0', '{prefix}1', etc.
    """
    pass