library(mixdist)
##################### CJKE #######################
pdf("Supplemental_figures_CJKE.pdf", paper="us", width=8, height=11)

# load the data
load("cjke.rda");

# use only first fixation duration
ffd<-cjke$ffd
# word length
wlen<-cjke$wlen
# landing position
landpos<-cjke$llpos
#type of next saccade, forward, refixation, or regression
next_fixtype<-cjke$next_fixtype

#set up graphs
fitresult<-list()
nrow<-9
ncol<-3

# ready, go
for (wl in 3:8) {
#	windows()
nf <- layout(matrix(1:(ncol*nrow),nrow,ncol,byrow = TRUE), 
		widths =matrix(rep(2,ncol*nrow),nrow,ncol), 
		heights=matrix(c(rep(1,ncol*(nrow-1)), rep(1.2, ncol)),nrow,ncol),
		respect=TRUE)
#layout.show(nf)
gr=1
  for (s in seq(0,wl,1)) {	

    for (ft in levels(next_fixtype)) {
      if(missing(ft)) next

	#par(mfrow = c(wl, 1))
	gtitle=sprintf("CJKE SingleFD, WordLen= %i, LandPos=%i ",wl, s)
	print(gtitle) 
	f<- ffd[!is.na(ffd) & ffd>0 & wlen==wl & landpos==s & next_fixtype==ft];
	if(length(f)==0) next
	f<-f[!is.na(f)]

	# get data in shape, 10ms bin
	mixgroup(f[f>0],breaks=c(0,seq(60,600,20),10000)) ->fdt

	# clamp mean and 2 of 3 sigma
	# means for #1 and #3 are fixed, the mean for #2 is free, see fixmu parameter below
	# note the means are different from Dundee corpus; there are many differences between the 2 studies
	# see text for more details
	meanpara=c(85, 160, 200)
	# these sigmas are fixed, see fixsigma parameter below
	sigmapara=c(30, 30, 70)
	# proportions are NOT fixed, see fixpi parameter
	pipara=c(0.10, 0.10, 0.80)
	dump<- mix(fdt,mixparam(meanpara,sigmapara, pipara), dist="lnorm", emsteps=10, 
		mixconstr(conpi="PFX", fixpi=c(F, F, F), conmu="MFX", fixmu=c(T,F,T), consigma="SFX", fixsigma=c(T,T, T))
	)

	# plot(dump, root=F, main=gtitle, ytop=0.01, xlab="Fixation Duration")
	par(mar=c(0,0,0,0))
	if (gr%%ncol==1) par(mar=c(0,2,0,0))
	if (gr%%(ncol*nrow)>ncol*(nrow-1) | gr%%(ncol*nrow)==0)  par(mar=c(2,0,0,0))
	if (gr%%ncol==1 & (gr%%(ncol*nrow)>ncol*(nrow-1) | gr%%(ncol*nrow)==0))  par(mar=c(2,2,0,0))
	plot(dump, root=F, ytop=0.01, sub="", ylab="", xlab="", axes=FALSE)
	# add Y axis only
	if (gr%%ncol==1) Axis(side=2)
	#if (gr%%nrow==0) Axis(side=1)
	if (gr%%(ncol*nrow)>ncol*(nrow-1) | gr%%(ncol*nrow)==0) Axis(side=1)
	if (s==wl) Axis(side=1)
	gr=gr+1

	# add kernal density estimate
	lines(density(f), col="blue", lty=2)
	# add verticle line to mark the C component, IF IT's >0
	if (dump$parameters$pi[2]>0.02) lines(c(dump$parameters$mu[2], dump$parameters$mu[2]), c(0,1), col="red")
	text(100,0.009, sprintf("%i (%s)",s, ft), cex = .8)
	text(printloc[1],0.009, sprintf("N=%i, m=%5.2f, se=%5.2f",length(f), mean(f),stderr(f)), cex = .8)
	text(printloc[1],0.007, sprintf("C: m=%5.2f, w=%5.2f",dump$parameters$mu[2], dump$parameters$pi[2]), cex = .8)
	text(printloc[1],0.005, sprintf("R: m=%5.2f, w=%5.2f",dump$parameters$mu[3],dump$parameters$pi[3]), cex = .8)
	text(printloc[1],0.003, sprintf("E: m=%5.2f, w=%5.2f",dump$parameters$mu[1],dump$parameters$pi[1]), cex = .8)
#	summary(dump)

# save data
	dump$wl <- wl
	dump$wdlp <-s
	dump$next_fixtype <-ft
	dump$N <-length(f)
	dump$mean <-mean(f)
	dump$se <-stderr(f)

	fitresult <-c(fitresult, list(dump))
  }
 }
}
graphics.off()
cjke200fixpi_fit_nextfixtype<-fitresult