ghlestimator 1.0.0

Linear generalized Huber estimator compatible with scikit-learn.

ghlestimator

Linear Generalized Huber Regressor compatible with scikit-learn. A detailed explanation of the underlying generalized Huber objective function can be found here.

The Generalized Huber Regressor depends on the definition of a invertible link function g and optimizes a term proportional to (y - ginv(X'w))**2 for samples where |g(y) - (X'w)| <= epsilon and a term proportional to |y - ginv(X'w)| for samples where |g(y) - (X'w)| > epsilon, where w is to be optimized and ginv denotes the inverse of g.

Parameters

class GeneralizedHuberRegressor(epsilon=1.0,max_iter=100,tol=1e-5, scale=10,
fit_intercept=True, link_dict={'g':_log,'ginv':_loginv,'ginvp':_loginvp})

epsilon : float, default 1.0

The parameter epsilon defines the crossover between the rmse type of loss 
and the mae type of loss.  

max_iter : int, default 100

Maximum number of iterations that
scipy.optimize.minimize(method="L-BFGS-B") should run for.

fit_intercept : bool, default True

Whether or not to fit the intercept.

tol : float, default 1e-5

The iteration will stop when max{|proj g_i | i = 1, ..., n} <= tol
where pg_i is the i-th component of the projected gradient.

scale : float, default 10.0

Preconditioner for better numerical stability. Input array is internally 
divided by scale.

link_dict : dictionary, default {'g':_log,'ginv':_loginv,'ginvp':_loginvp}

The link function 'g', it's inverse 'ginv' and the derivative of the 
latter 'ginvp' have to be specified as callables. 
The default link function is g(x) = sign(x)log(1+|x|).              

Attributes

coef_ : array, shape (n_features,)

Fitted coefficients got by optimizing the generalized Huber loss.

intercept_ : float

The bias.

n_iter_ : int

Number of iterations that
scipy.optimize.minimize(method="L-BFGS-B") has run for.

Methods

fit(self, X, y)

Fit the model to the given training features X and target y both given as 
ndarrays.

predict(self, X)

Predict using the fitted linear model.

score(self, X, y)

Return the coefficient of determination R^2 of the prediction.

Installation

Use the package manager pip to install ghlestimator.

pip install ghlestimator

Usage

from ghlestimator import GeneralizedHuberRegressor

ghl = GeneralizedHuberRegressor() # initializes default ghl estimator 
ghl.fit(X, y) # fit on features X and target y
ghl.score(X, y) # compute the R^2 score
ghl.predict(X) # make pedictions

License

MIT