Graphics Reference
In-Depth Information
Algorithm 5
A univariate feature ranking algorithm.
function
Ranking Algorithm(
x
- features,
U
- measure)
initialize:
list
L
={}
L
stores ordered features
for
each feature
x
i
,
i
∈{
1
,...,
M
}
do
v
i
=
compute(
x
i
,
U
)
position
x
i
into
L
according to
v
i
end for
return
L
in decreasing order of feature relevance.
end function
operation allows us to obtain a minimum subset of features and is usually adopted.
However, it is not recommended if the goal is to find the minimum feature subset.
7.3.1.2 Minimum Subset Techniques
The number of relevant features is a parameter that is often not known by the prac-
titioner. There must be a second category of techniques focused on obtaining the
minimum possible subset without ordering the features. An algorithm belonging to
this category returns a minimum feature subset and no difference is made for features
in the subset. So, whatever is relevant within the subset, is otherwise irrelevant.
The minimum subset algorithm is detailed in Algorithm 6. The
subsetGenerate()
function returns a subset following a certain search method, in which a stopping
criterion determines when
stop
is set to
true
; function
legitimacy()
returns
true
if subset
S
k
satisfies measure
U
. Function
subsetGenerate()
can take one of the
generation schemes.
Algorithm 6
A minimum subset algorithm.
function
Min- Set Algorithm(
x
- features,
U
- measure)
initialize:
L
={}
,stop=
false
S
holds the minimum set
repeat
S
k
= subsetGenerate(x)
stop
can be set here
if
legitimacy(
S
k
,
U
)is
true and
#
(
S
k
)<
#
(
S
)
then
S
=
S
k
S
is replaced by
S
k
end if
until
stop =
true
return
S
- the minimum subset of features
end function
7.3.2 Evaluation
Several aspects must be taken into account to evaluate a FS method. Among them, it
is important to known how well does a feature selector work or the conditions under
