tanh_scaler

tanh_scaler.py

import numpy as np
from sklearn.preprocessing import RobustScaler
from sklearn.base import TransformerMixin

class TanhScaler(TransformerMixin):
    def __init__(self,
                 factor=2., 
                 decim=1,
                 n_max=None,
                 shuffle=False,
                 with_centering=True,
                 with_scaling=True,
                 quantile_range=(25.0, 75.0),):
        self.factor = factor
        self.decim = decim
        self.n_max = n_max
        self.shuffle = shuffle
        self.quantile_range = quantile_range
    
    
    def fit(self, X, y=None):
        self.scaler = RobustScaler(
            with_centering=True,
            with_scaling=True,
            quantile_range=self.quantile_range,
            copy=True,
            unit_variance=True,
        )
        if self.decim:
            X = X[::decim]
        if self.n_max:
            if self.shuffle:
                select = np.random.permutation(len(X))[:self.n_max]
            else:
                select = slice(0, self.n_max)
            X = X[select]
        self.scaler.fit(X)
        return self
        
    def transform(self, X, y=None):
        X -= self.scaler.center_
        X /= self.scaler.scale_
        return np.tanh(X / self.factor)

x = np.random.randn(10000, 1)/2

x[:100] += 200

plt.subplot(131).plot(np.sort(x.ravel()))

y = RobustScaler(unit_variance=False).fit_transform(x)
plt.subplot(132).plot(np.sort(y.ravel()))

z = TanhScaler(10).fit_transform(x)
plt.subplot(133).plot(np.sort(z.ravel()))