# Copyright (c) 2014  Takafumi Kanamori [kanamori@is.nagoya-u.ac.jp]
# All rights reserved.  See the file COPYING for license terms.

#library(MASS)
library(robustbase)
source('functions.r')

## setup of sample distribution
d  <- 5               ## dimension of covariate variable
ntrain <-  100        ## num. of observations
ntest  <- 1000        ## num. of test samples
contam_ratio <- 0.05  ## ratio of outliers: 1-c
in_sd <- 1            ## sd of target conditional distribution
out_mean <- 5         ## mean of outlier distribution
out_sd   <- 1         ## std deviation of outlier distribution

theta <- c(0,rnorm(d))    ## true regression parameter
## observations
x <- matrix(rnorm(ntrain*d),ntrain,d)
y <- as.vector(cbind(1,x) %*% theta)
m <- sum(rbinom(ntrain,1,prob=contam_ratio)) ## num of outliers
n <- ntrain-m                                ## number of inliers
## contamination
if(m>0){
  out_idx  <- sample(ntrain,m)
  y[out_idx]  <- y[out_idx]  + rnorm(m,mean=out_mean,sd=out_sd)
  y[-out_idx] <- y[-out_idx] + rnorm(n,sd=in_sd)
}else{
  y  <- y + rnorm(n,sd=in_sd)
}
## test data
xtest <- matrix(rnorm(ntest*d),ntest,d)
ytest <- as.vector(cbind(1,xtest) %*% theta)+rnorm(ntest,sd=in_sd)

## estimate
res <- reg_unnormalizedLSmodel(x,y,gamma=0.5)                        ## gamma=0.5, linear model
print(res)                                                           ## print estimated result
cat("RMSE =",sqrt(mean((ytest-cbind(1,xtest) %*% res$par)^2)),'\n')  ## print root mean square error