Scale

Defined in fynance.features.scale

class Scale(X, kind='std', a=0., b=1., axis=0, **kwargs)

Bases: object

Fit/transform-style scaler for time-series data.

Wraps the four scaling primitives (standardize, normalize, roll_standardize, roll_normalize) behind a uniform fit / scale / revert API. Parameters are fitted once at construction and reused on subsequent calls — the typical pipeline pattern of fitting on a training window and applying the same transform to the test window, which avoids leaking test-period statistics into training.

The revert method inverts the transformation, useful when the target of an ML model was scaled and the prediction must be converted back to the original units.

Parameters:

Xnp.ndarray[dtype, ndim=1 or 2]

Data to fit the parameters of scale transformation.

kindstr, optional

• “std” : Standardized scale transformation (default), see standardize.

• “norm” : Normalized scale transformation, see normalize.

• “raw” : No scale is apply.

• “roll_std” : Standardized scale transformation, computed with rolling mean and standard deviation (see roll_standardize).

• “roll_norm” : Normalized scale transformation, computed with roling minimum and maximum (see roll_normalize).

a, bfloat or array_like, optional

Some scale factors to apply after the transformation. By default is respectively 0 and 1.

axisint, optional

Axis along which compute the scale parameters. Default is 0.

**kwargskeyword arguments for particular functions

E.g: for rolling function set w the lagged window (see roll_normalize) or for rolling standardization set kind_moment={"s", "w", "e"} (see roll_standardize).

Attributes:

funccallable

The scale function.

revert_funccallable

The revert scale function.

paramsdict

Parameters of the scale transformation.

axisint

The axis along which is computed the scale parameters.

kindstr

The kind of scale transformation.

Methods

fit(X[, kind, a, b, axis])	Compute the parameters of the scale transformation.
scale(X[, axis])	Scale the data with the fitted parameters.
revert(X[, axis])	Revert the transformation of the scale with the fitted parameters.

See also

normalize, standardize, roll_standardize, roll_normalize

__call__(X, axis=None)

Callable method to scale data with fitted parameters.

Parameters:

Xnp.ndarray[dtype, ndim=1 or 2]

Data to scale.

Returns:

np.ndarray[dtype, ndim=1 or 2]

Scalled data.

fit(X, kind=None, a=0., b=1., axis=0, **kwargs)

Compute the parameters of the scale transformation.

Parameters:

Xnp.ndarray[dtype, ndim=1 or 2]

Data to fit the parameters of scale transformation.

kindstr, optional

• “std” : Standardized scale transformation (default), see standardize.

• “norm” : Normalized scale transformation, see normalize.

• “raw” : No scale is apply.

• “roll_std” : Standardized scale transformation, computed with rolling mean and standard deviation (see roll_standardize).

• “roll_norm” : Normalized scale transformation, computed with roling minimum and maximum (see roll_normalize).

a, bfloat or array_like, optional

Some scale factors to apply after the transformation. By default is respectively 0 and 1.

axisint, optional

Axis along which compute the scale parameters. Default is 0.

**kwargskeyword arguments for particular functions

E.g: for rolling function set w the lagged window (see roll_normalize) or for rolling standardization set kind_moment={"s", "w", "e"} (see roll_standardize).

revert(X, axis=None)

Revert the transformation of the scale with the fitted parameters.

Parameters:

Xnp.ndarray[dtype, ndim=1 or 2]

Data to revert the scale.

Returns:

np.ndarray[dtype, ndim=1 or 2]

The revert transformed data.

scale(X, axis=None)

Scale the data with the fitted parameters.

Parameters:

Xnp.ndarray[dtype, ndim=1 or 2]

Data to scale.

Returns:

np.ndarray[dtype, ndim=1 or 2]

Scalled data.