Information Technology Reference
In-Depth Information
In the case of a two-dimensional biplot the intersection L N ( µ ) follows by setting
all elements of ( y , y m + 1 ) except the first two equal to zero. Denote these two nonzero
coordinates as z = ( z 1 , z 2 ) .Then
d µ d n + 1 (µ)
1
λ 1 y 1
λ 2 y 2 z =−
n 1 d µ d n + 1 (µ)
1
1
.
(5.16)
t .
Multiplying the coefficient vector a and t with any constant will result in an equally
valid equation for the intersection space L N (µ) . Therefore we reparameterize the
equation as l (µ) z = t such that l 1 (µ) + l 2 (µ) = 1.
Thus the intersection space becomes a line with equation of the form a z
=
5.4.2.1 Normal projection
With normal projection, we find a trajectory that is normal to all intersection spaces
L N . If we denote the k th two-dimensional biplot trajectory at the marker µ by
b 1 (µ)
b 2 (µ)
,
β k (µ) =
then the direction of the tangent at
β k (µ)
is defined by
d µ b 1 (µ)
d
d
b 2 (µ)
µ
and the gradient of the tangent is given by
d
d
b 2 (µ)
µ
b 1 (µ) .
d
d µ
The tangent must be orthogonal to the direction of the intersection space L N (
µ)
with gradient
l 1 (µ)/
l 2 (µ)
, therefore
d
d
b 1 (µ)
l 1 (µ)
µ
l 2 (µ) =−
b 2 (µ) .
d
d
µ
Furthermore, since β k (µ) lies on the intersection space, it can be expressed as
l (µ) β k (µ) = t (µ) ,
which after differentiation becomes
d µ l 1 (µ)
b 1 (µ) +
d µ b 1 (µ)
l 1 (µ) +
d µ l 2 (µ)
b 2 (µ) +
d µ b 2 (µ)
l 2 (µ) =
d
d
t (µ).
µ
After some algebraic manipulation it follows that
b 1 (µ)
l 2 (µ)
d
d µ
t (µ)
l 2 (µ)
d
d µ
= l 1 (µ)
.
(5.17)
Search WWH ::




Custom Search