RobustScaler

• Remueve la mediana y escala los datos de acuerdo con el rango intercuartil.

• Es un método de escalamiento robusto a outliers.

import seaborn as sns

penguins = sns.load_dataset("penguins")

data = penguins[["flipper_length_mm"]]
data = data.rename(columns={"flipper_length_mm": "original"})

from sklearn.preprocessing import RobustScaler

robustScaler = RobustScaler(
    # -------------------------------------------------------------------------
    # If True, center the data before scaling.
    with_centering=True,
    # -------------------------------------------------------------------------
    # If True, scale the data to interquartile range.
    with_scaling=True,
    # -------------------------------------------------------------------------
    # Quantile range used to calculate scale_.
    quantile_range=(0.25, 0.75),
    # -------------------------------------------------------------------------
    # If True, scale data so that normally distributed features have a variance
    # of 1.
    unit_variance=False,
)

robustScaler.fit(data[["original"]])

data["transformed"] = robustScaler.transform(data[["original"]])

g = sns.jointplot(x="original", y="transformed", data=data, kind="scatter")
g.fig.set_figwidth(3)
g.fig.set_figheight(3)

display(
    robustScaler.center_,
    robustScaler.scale_,
)

array([197.])

array([3.41])