# 
# Written by:
# -- 
# John L. Weatherwax                2007-07-05
# 
# email: wax@alum.mit.edu
# 
# Please send comments and especially bug reports to the
# above email address.
# 
#-----

yf = seq( -2, +2, length.out=100 )

#postscript("../../WriteUp/Graphics/Chapter10/dup_fig_10_4.eps", onefile=FALSE, horizontal=FALSE)

# Missclassification loss:
# 
miss_classification_loss = rep( 0, length(yf) )
miss_classification_loss[ yf<0 ] = 1.0
plot( yf, miss_classification_loss, type="l", col="gray", xlim=c(-2,+2), ylim=c(0,3), xlab="y f", ylab="loss" )

# Exponential loss:
# 
exp_loss = exp( - yf )
lines( yf, exp_loss, type="l", col="cyan" )

# Binomial deviance (try all possible "scalings"):
# 
lines( yf, log( 1 + exp( -2 * yf ) )/log(2), type="l", col="orange" ) # here we "scale" binomial deviance

###t = + 0.5 * log( exp(1) - 1 ) 
###lines( yf, log( 1 + exp( -2 * (yf-t) ) ), type="l", col="orange" ) # here we shift the x-axis of binomial deviance

###lines( yf, log( 1 + exp( -2 * yf ) ) - log(2) + 1, type="l", col="orange" ) # here we shift the y-axis of binomial deviance

# Squared Error loss:
# 
lines( yf, ( 1 - yf )^2, type="l", col="red" )

# Hinge loss:
# 
hinge_loss = 1 - yf
hinge_loss[ hinge_loss<0 ] = 0
lines( yf, hinge_loss, type="l", col="darkgreen" )

#dev.off()