#
# Multiple training routines EPage 82
# Problem EPage 129
# 
# Written by:
# -- 
# John L. Weatherwax                2009-04-21
# 
# email: wax@alum.mit.edu
# 
# Please send comments and especially bug reports to the
# above email address.
# 
#-----

save_plots = F

library(caret) 

set.seed(0)

data(tecator) 

# Part (b):
#
pr = prcomp( absorp )

vars = pr$sdev^2
total_var = sum(vars)

if( save_plots ){ postscript("../../WriteUp/Graphics/Chapter6/chap_6_prob_1_scree_plot.eps", onefile=FALSE, horizontal=FALSE) }
plot( ( vars/total_var ) * 100, xlim=c(0,40), type='b', pch=16, xlab='number of components', ylab='percent of total variation' )
grid()
if( save_plots ){ dev.off() }

# Part (c):
# 
fat = endpoints[,2]
absorp = data.frame(absorp)

# For various models build and then compare performance:
#
set.seed(0)
lm_model = train( absorp, fat, method="lm", preProcess=c("center","scale"), trControl=trainControl(method="repeatedcv",repeats=5) )

# For rlm we cannot have a singular predictor covariance matrix thus we preprocess with PCA:
# 
set.seed(0)
rlm_model = train( absorp, fat, method="rlm", preProcess=c("pca"), trControl=trainControl(method="repeatedcv",repeats=5) )

set.seed(0)
pls_model = train( absorp, fat, method="pls",
                   # the default tuning grid evaluates comonents 1 ... tuneLength
                   tuneLength=40, 
                   preProcess=c("center","scale"), trControl=trainControl(method="repeatedcv",repeats=5) )

# Ridge regression training:
#
if( FALSE ){ 
    ridgeGrid = data.frame(.lambda=seq(0,1,length=20))
    set.seed(0)
    ridge_model = train( absorp, fat, method="ridge",
                         # fit the model over many penalty values
                         tuneGrid = ridgeGrid,
                         preProcess=c("center","scale"), trControl=trainControl(method="repeatedcv",repeats=5) )
}

# Elastic net training:
# 
enetGrid = expand.grid(.lambda=seq(0,1,length=20), .fraction=seq(0.05, 1.0, length=20))
set.seed(0)
enet_model = train( absorp, fat, method="enet",
                    # fit the model over many penalty values
                    tuneGrid = enetGrid,
                    preProcess=c("center","scale"), trControl=trainControl(method="repeatedcv",repeats=5) )

# EPage 82 
resamp = resamples( list(lm=lm_model,rlm=rlm_model,pls=pls_model,enet=enet_model) )
print( summary(resamp) )

if( save_plots ){ postscript("../../WriteUp/Graphics/Chapter6/chap_6_prob_1_resamp_dotplot.eps", onefile=FALSE, horizontal=FALSE) }
dotplot( resamp, metric="RMSE" )
if( save_plots ){ dev.off() }

print( summary(diff(resamp)) )