sklearn.linear model.HuberRegressor

Jump to: navigation, search

A sklearn.linear model.HuberRegressor is an Huber Regression System within sklearn.linear_model class.

1) Import Huber Regression model from scikit-learn : from sklearn.linear_model import HuberRegressor
2) Create design matrix X and response vector Y
3) Create Huber Regression object: Hreg=HuberRegressor([epsilon=1.35, max_iter=100, alpha=0.0001, warm_start=False, fit_intercept=True, tol=1e-05])
4) Choose method(s):
Input: Output:
#Importing modules
from sklearn.linear_model import HuberRegressor
from sklearn.model_selection import cross_val_predict
from sklearn.datasets import load_boston
from sklearn.metrics import explained_variance_score, mean_squared_error
import numpy as np
import pylab as pl
boston = load_boston() #Loading boston datasets
x = # Creating Regression Design Matrix
y = # Creating target dataset
Hreg= HuberRegressor(epsilon=1.0) # Create Huber regression object,y) # predicted values

#Calculaton of RMSE and Explained Variances

yp_cv = cross_val_predict(Hreg, x, y, cv=10) #Calculation 10-Fold CV
RMSE =np.sqrt(mean_squared_error(y,yp))
RMSECV =sqrt(mean_squared_error(y,yp_cv)

# Printing Results

print('Method: Huber Regression')
print('RMSE on the dataset: %.4f' %RMSE)
print('RMSE on 10-fold CV: %.4f' %RMSECV)
print('Explained Variance Regression Score on the dataset: %.4f' %Evariance)
print('Explained Variance Regression 10-fold CV: %.4f' %Evariance_cv)

#plotting real vs predicted data

pl.plot(yp, y,'ro')
pl.plot(yp_cv, y,'bo', alpha=0.25, label='10-folds CV')
pl.title('Huber Regression, epsilon=1.0')
huber boston10foldepsilon1.0.png
(blue dots correspond to 10-Fold CV)

Method: Huber Regression
RMSE on the dataset: 5.1709
RMSE on 10-fold CV: 6.4916
Explained Variance Regression Score on the dataset: 0.6932
Explained Variance Regression 10-fold CV: 0.5027



Linear regression model that is robust to outliers.
The Huber Regressor optimizes the squared loss for the samples where |(y - X'w) / sigma| < epsilon and the absolute loss for the samples where |(y - X'w) / sigma| > epsilon, where w and sigma are parameters to be optimized. The parameter sigma makes sure that if y is scaled up or down by a certain factor, one does not need to rescale epsilon to achieve the same robustness. Note that this does not take into account the fact that the different features of X may be of different scales.
This makes sure that the loss function is not heavily influenced by the outliers while not completely ignoring their effect.