Momocs (version 1.1.6)

calibrate_deviations: Quantitative calibration, through deviations, for Out and Opn objects

Description

Calculate deviations from original and reconstructed shapes using a range of harmonic number.

Usage

calibrate_deviations(x, method, id, range, norm.centsize, dist.method,
  interpolate.factor, dist.nbpts, plot)

Arguments

x

and Out or Opn object on which to calibrate_deviations

method

any method from c('efourier', 'rfourier', 'tfourier') and 'dfourier'.

id

the shape on which to perform calibrate_deviations

range

vector of harmonics (or degree for opoly and npoly on Opn) on which to perform calibrate_deviations. If not provided, the harmonics corresponding to 0.9, 0.95 and 0.99 are used.

norm.centsize

logical whether to normalize deviation by the centroid size

dist.method

a method such as edm_nearest to calculate deviations

interpolate.factor

a numeric to increase the number of points on the original shape (1 by default)

dist.nbpts

numeric the number of points to use for deviations calculations

plot

logical whether to print the graph (FALSE is you just want the calculations)

Value

a ggplot object and the full list of intermediate results. See examples.

Details

Note that from version 1.1, the calculation changed and fixed a problem. Before, the 'best' possible shape was calculated using the highest possible number of harmonics. This worked well for efourier but not for others (eg rfourier, tfourier) as they are known to be unstable with high number of harmonics. From now on, Momocs uses the 'real' shape, as it is (so it must be centered) and uses coo_interpolate to produce interpolate.factor times more coordinates as the shape has and using the default dist.method, eg edm_nearest, the latter finds the euclidean distance, for each point on the reconstructed shape, the closest point on this interpolated shape. interpolate.factor being set to 1 by default, no interpolation will be made in you do not ask for it. Note, that interpolation to decrease artefactual errors may also be done outside calibrate_deviations and will be probably be removed from it in further versions. For *poly methods on Opn objects, the deviations are calculated from a degree 12 polynom.

See Also

Other calibration: calibrate_harmonicpower, calibrate_r2, calibrate_reconstructions

Examples

data(bot)
calibrate_deviations(bot)

## Not run: ------------------------------------
# 
# # on Opn
# data(olea)
# calibrate_deviations(olea)
# 
# # lets customize the ggplot
# library(ggplot2)
# gg <- calibrate_deviations(bot, id=1:20)$gg
# gg + geom_hline(yintercept=c(0.001, 0.005), linetype=3)
# gg + labs(col="Number of harmonics", fill="Number of harmonics",
#            title="Harmonic power") + theme_bw()
# gg + coord_polar()
# 
# ### intermediate results can be accessed eg with:
# shp <- hearts[1] %>% coo_interpolate(360) %>% coo_samplerr(60) %>% Out()
# calibrate_deviations(shp, id=1, range=1:24, method="efourier") %$%
#    res %>% apply(1, mean) %>% plot(type="b")
# calibrate_deviations(shp, id=1, range=1:24, method="rfourier") %$%
#    res %>% apply(1, mean) %>% plot(type="b")
# calibrate_deviations(shp, id=1, range=1:24, method="tfourier") %$%
#    res %>% apply(1, mean) %>% plot(type="b")
# 
# # ... providing an illustration of the e vs r/t fourier approaches developped in the help page.
# 
# ### illustration of interpolate.factor
# interp <- c(1, 5, 25)
# res <- list()
# for (i in seq_along(interp))
# calibrate_deviations(shp, id=1, range=1:24,
#    method="tfourier", interpolate.factor=interp[i], plot=FALSE) %$%
#    res %>% apply(1, mean) -> res[[i]]
# 
#  ### int_5 is more accurate than no inteprolation
#  sign(res[[2]] - res[[1]])
#  ### int 25 is more accurate than int_5, etc.
#  sign(res[[3]] - res[[2]])
## ---------------------------------------------