Page 166 -

8 The Importance of Ontological Structure: Why Validation by ‘Fit-to-Data’... 163

class), (continued)

p) each j,1,0)
D for
p used DD

model) “maximum”) model) “euclidean”) model) “minkowski”, model) “maximum”) sqrt(sum((vdata--model)ˆ2)/length(vdata)) being number one ifelse(model[i]

sum(abs(vdata--model)) rbind(vdata, D method sqrt(sum((vdata--model)ˆ2)) rbind(vdata, D method rbind(vdata, D method max(abs(vdata--model)) rbind(vdata, D method sum((vdata--model)ˆ2) with classes, model[i]<

code < dist(dm, < dist(dm, < dist(dm, < dist(dm, (i.e.
R or dm or dm dm a or dm variables j,1,0)and

vector vectors. differences implies Zero norm. the for nominal DD

model model Zero values. L 2 the for
and and the of of pair of corrected used ifelse(vdata[i]
data data power elements. any square norm fit be also
the fit the pth vector the L 2 perfect could
between perfect a between fit the of model between simply fit the is a implies case vdata[i]<

distance implies Zero distance perfect a sum the of and data difference fit perfect is error perfect a error Zero data. This variable) j: class

Description variable) absolute Total elements. Euclidean The implies Zero root pth The the between fit perfect maximum The a implies squared of Sum implies Zero Root-mean-squared the of size (Boolean each for

validation (cardinal 2f0, 1g Here,

of 2R (SSE) error class.
Measures model[i] squared model[i] each for

8.3 norm norm norm of error vdata[i], repeated
Table Metric vdata[i], L 1 L 2 norm p L 1 Sum RMS if