Class ExponentialMovingFeature

• Jump right in for a hands-on

Import

from NitroFE import ExponentialMovingFeature

The exponential moving average is caluclated as

\[ \operatorname{ema[0]} = dataframe[0] \]

\[ \operatorname{ema[t]} = (1-alpha)*ema[t-1] + alpha*x[t] \]

if you want you calculate via the traditional way, in which the ema[0] isnt the first value in the dataframe (usually a simple moving average over the first few values ),

you can use the paramters 'initialize_using_operation' and 'initialize_span' , in which case the exponential moving avergae will be calculated as

\[ \operatorname{ema}[0 \to (initialize\_span - 2)] = Nan \]

\[ \operatorname{ema}[initialize\_span - 1] = operation( dataframe[0 \to (initialize\_span-1) ] ) \]

\[ \operatorname{ema[t]} = (1-alpha)*ema[t-1] + alpha*x[t] \]

Methods

Provided dataframe must be in ascending order.

__init__(self, alpha=None, operation='mean', initialize_using_operation=False, initialize_span=None, com=None, span=None, halflife=None, min_periods=0, ignore_na=False, axis=0, times=None) special

Parameters:

Name	Type	Description	Default
alpha	float	Specify smoothing factor directly, by default None	None
operation	str	operation to be performed for the moving feature,available operations are 'mean','var','std', by default 'mean'	'mean'
initialize_using_operation	bool	If True, then specified 'operation' is performed on the first 'initialize_span' values, and then the exponential moving average is calculated, by default False	False
initialize_span	int	the span over which 'operation' would be performed for initialization, by default None	None
com	float	Specify decay in terms of center of mass, by default None	None
span	int	specify decay in terms of span , by default None	None
halflife	float	Specify decay in terms of half-life, by default None	None
min_periods	int	Minimum number of observations in window required to have a value (otherwise result is NA), by default 0	0
ignore_na	bool	Ignore missing values when calculating weights; specify True to reproduce pre-0.15.0 behavior, by default False	False
axis	int	The axis to use. The value 0 identifies the rows, and 1 identifies the columns, by default 0	0
times	str	Times corresponding to the observations. Must be monotonically increasing and datetime64[ns] dtype, by default None	None

Source code in nitrofe\time_based_features\moving_average_features\moving_average_features.py

def __init__(
    self,
    alpha: float = None,
    operation: str = "mean",
    initialize_using_operation: bool = False,
    initialize_span: int = None,
    com: float = None,
    span: int = None,
    halflife: float = None,
    min_periods: int = 0,
    ignore_na: bool = False,
    axis: int = 0,
    times: str = None,
):
    """
    Parameters
    ----------
    alpha : float, optional
        Specify smoothing factor  directly, by default None
    operation : str, {'mean','var','std'}
        operation to be performed for the moving feature,available operations are 'mean','var','std', by default 'mean'
    initialize_using_operation : bool, optional
        If True, then specified 'operation' is performed on the first 'initialize_span' values, and then the exponential moving average is calculated, by default False
    initialize_span : int, optional
        the span over which 'operation' would be performed for initialization, by default None
    com : float, optional
        Specify decay in terms of center of mass, by default None
    span : float, optional
        specify decay in terms of span , by default None
    halflife : float, optional
        Specify decay in terms of half-life, by default None
    min_periods : int, optional
        Minimum number of observations in window required to have a value (otherwise result is NA), by default 0
    ignore_na : bool, optional
        Ignore missing values when calculating weights; specify True to reproduce pre-0.15.0 behavior, by default False
    axis : int, optional
        The axis to use. The value 0 identifies the rows, and 1 identifies the columns, by default 0
    times : str, optional
        Times corresponding to the observations. Must be monotonically increasing and datetime64[ns] dtype, by default None
    """

    self.com = com
    self.span = span
    self.halflife = halflife
    self.alpha = alpha
    self.min_periods = min_periods if min_periods != None else 0
    self.adjust = False
    self.ignore_na = ignore_na
    self.axis = axis
    self.times = times
    self.operation = operation
    self.last_values_from_previous_run = None
    self.initialize_using_operation = initialize_using_operation
    self.initialize_span = initialize_span

fit(self, dataframe, first_fit=True)

For your training/initial fit phase (very first fit) use fit_first=True, and for any production/test implementation pass fit_first=False

Parameters:

Name	Type	Description	Default
dataframe	Union[pandas.core.frame.DataFrame, pandas.core.series.Series]	dataframe containing column values to create exponential moving feature over	required
first_fit	bool	Moving features require past values for calculation. Use True, when calculating for training data (very first fit) Use False, when calculating for subsequent testing/production data { in which case the values, which were saved during the last phase, will be utilized for calculation }, by default True	True

Source code in nitrofe\time_based_features\moving_average_features\moving_average_features.py

def fit(self, dataframe: Union[pd.DataFrame, pd.Series], first_fit: bool = True):
    """
    For your training/initial fit phase (very first fit) use fit_first=True, and for any production/test implementation pass fit_first=False

    Parameters
    ----------
    dataframe : Union[pd.DataFrame, pd.Series]
        dataframe containing column values to create exponential moving feature over
    first_fit : bool, optional
        Moving features require past values for calculation.
        Use True, when calculating for training data (very first fit)
        Use False, when calculating for subsequent testing/production data { in which case the values, which
        were saved during the last phase, will be utilized for calculation }, by default True
    """
    if not first_fit:
        if self.last_values_from_previous_run is None:
            raise ValueError(
                "First fit has not occured before. Kindly run first_fit=True for first fit instance,"
                "and then proceed with first_fit=False for subsequent fits "
            )
        self.adjust = False
        dataframe = pd.concat(
            [self.last_values_from_previous_run, dataframe], axis=0
        )
    else:

        if self.initialize_using_operation:
            self.min_periods = 0
            if (self.initialize_span is None) and (self.span is None):
                raise ValueError(
                    "For initialize_using_operation=True,"
                    "either initialize_span or span value is required"
                )
            elif (self.initialize_span is None) and (self.span is not None):
                self.initialize_span = self.span

            first_frame = self._perform_temp_operation(
                dataframe[: self.initialize_span].rolling(
                    window=self.initialize_span
                )
            )
            dataframe = pd.concat([first_frame, dataframe[self.initialize_span :]])
        else:
            if self.initialize_span is not None:
                raise ValueError(
                    "In order to use initialize_span, initialize_using_operation must be True"
                )

    _dataframe = dataframe.ewm(
        com=self.com,
        span=self.span,
        halflife=self.halflife,
        alpha=self.alpha,
        min_periods=self.min_periods,
        adjust=self.adjust,
        ignore_na=self.ignore_na,
        axis=self.axis,
        times=self.times,
    )
    _return = self._perform_temp_operation(_dataframe)

    if not first_fit:
        _return = _return.iloc[1:]
    self.last_values_from_previous_run = _return.iloc[-1:]
    return _return

References

• pydata, "ewm"

  https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.ewm.html