rrat
implements Robust Regression approach tailored to deal with Asymmetric Tail noise distribution.
The main idea is to learn most of the parameters of the model using conditional quantile estimators
(which are biased but robust estimators of the regression)
and then to learn a few remaining parameters to combine and correct these estimators,
in order to minimize the average squared error in an unbiased way in the end.
rrat(x, y, taus = 0.5, ...)
a data frame of predictors in training data. Similar as design matrix style but without intercept.
a numeric vector of response.
candidate quantiles of underlying robust regression.
optional parameters to be passed to the low level function rq
.
An object of class rrat
, which is a list with the
following components:
candidate quantiles of underlying robust regression.
a named vector of coefficients
a named matrix of coefficients from underlying quantile regressions bases
a number to correct the combinations of underlying quantile regressions bases
%% ~Describe the value returned %% If it is a LIST, use %% \item{comp1 }{Description of 'comp1'} %% \item{comp2 }{Description of 'comp2'} %% ...
Moreover than the original RRAT paper, our implementation is extended to be able to conduct feature regularization of LASSO by adding parameter method
as lasso
,
as well as tuning the value of the additional penalty parameter lambda
in the function.
Such feature regularization is employed in all the underlying base quantile models simultaneously. The architecture and combination of these underlying base quantile models should be designed according to the asymmetric tail pattern of the noise distribution.
Takeuchi, Ichiro & Bengio, Y. & Kanamori, Takafumi. (2002). ``Robust Regression with Asymmetric Heavy-Tail Noise Distributions''. Neural computation. 14. 2469-96. 10.1162/08997660260293300.
# NOT RUN {
rrat_1 <- rrat(x=iris[,2:4], y=iris[,1], taus = 0.5)
# }
Run the code above in your browser using DataLab