example code for getting surfaces with some good properties.

examplesurfaces.R

load("/home/tim/Desktop/Hom_viv_muj.rdata")
ls()
library(RColorBrewer)
# select a suitable ramp (sequential for rates, divergent for differences or ratios)
# display.brewer.all()
ramp <- colorRampPalette(brewer.pal(9, "RdPu"), space = "Lab")

years <- 2000:2017
ages  <- c(0,1,seq(5,105,by=5))

# probably log rates, but not necessarily for homicide.
breaks <- pretty(MxF_hom[ages[1:nrow(MxF_hom)] <= 80,], n = 20)
# give it a ceiling
breaks <- c(breaks,max(range(pretty(MxF_hom))))
# gives good breaks for absolutes, not so good for log


pdf("test.pdf")
par(mfrow=c(1,3))

# surf 1
image(  x = years,
		y = ages, 
		t(MxF_hom), 
		breaks = breaks, # breaks from above
		col = ramp(length(breaks) - 1), # one less color than there are intervals
		asp = 1) # age = time

contour( x = 2000.5:2016.5,  # midpoints!!
		y = c(.5,3,seq(7.5,102.5,by=5)), 
		t(MxF_hom),
		add = TRUE) # draws on top

# surf 2
image(  x = years,
		y = ages, 
		t(MxF_hom), 
		breaks = breaks, 
		col = ramp(length(breaks) - 1),
		asp = 1)

contour( x = 2000.5:2016.5,
		y = c(.5,3,seq(7.5,102.5,by=5)), 
		t(MxF_hom),
		add = TRUE)

# surf 3
image(  x = years,
		y = ages, 
		t(MxF_hom), 
		breaks = breaks, 
		col = ramp(length(breaks) - 1), 
		asp = 1)  

contour( x = 2000.5:2016.5,
		y = c(.5,3,seq(7.5,102.5,by=5)), 
		t(MxF_hom),
		add = TRUE)

dev.off()

library(MortalitySmooth)
# use Mort1Dsmooth()
# https://www.rdocumentation.org/packages/MortalitySmooth/versions/2.3.4/topics/Mort1Dsmooth

# suggestion: do image() on raw data, as above, but draw contours based on the smoothed surface,
# where each year is smoothed independently. Reason: there are ruptures to preserve.

# for ratios visualize the log, 
# before: 
hist(runif(100)/runif(100),n=20)
# after: ( < 0 after logging = < 1 of ratio).
# now negatives are on same scale as positives, and even steps are *proportional*
hist(log(runif(100)/runif(100)),n=20)
# choose breaks from this:
# however, to get good breaks, you can think in the analog:
example_breaks <- c(1/1000,1/100,1/10,1,10,100,1000)
# for log ratios, you probably won't go as big as 1000, but you'll want to think in terms of 
# equal jumps in any case, mirrored. Then you can use these as absolute breaks for colors
# and contours.

# 2) choose a ramp that is DIVERGENT, and make sure the neutral midpoint color is on 0,
# You can get that by have the same number and spread of low and high values, like 
# the above example.