Drawing of the Epanechnikov core
The Epanechnikov core (according to WA Jepanetschnikow) is the core that fulfills the following properties for a compact carrier :
k
(
x
)
≥
0
{\ displaystyle k (x) \ geq 0}
for all
x
∈
R.
{\ displaystyle x \ in \ mathbb {R}}
∫
k
(
x
)
d
x
=
1
{\ displaystyle \ int k (x) \, {\ mathrm {d}} x = 1}
∫
x
2
k
(
x
)
d
x
=
1
{\ displaystyle \ int x ^ {2} k (x) \, {\ mathrm {d}} x = 1}
∫
k
2
(
x
)
d
x
{\ displaystyle \ int k ^ {2} (x) \, {\ mathrm {d}} x}
is minimized.
With these properties, the Epanechnikov kernel minimizes the mean square deviation of the corresponding kernel density estimator among all kernels . It is a polynomial of the form .
a
+
b
x
2
{\ displaystyle a + bx ^ {2}}
We want to put the numerical factors of the core in context. First consider the normalized family , whose terms assume a hill shape in the interval and which converges to the rectangular distribution of the height for large n :
a
,
b
{\ displaystyle a, b}
k
n
,
d
(
x
)
{\ displaystyle k_ {n, d} (x)}
[
-
d
,
d
]
{\ displaystyle [-d, d]}
1
2
d
{\ displaystyle {\ tfrac {1} {2d}}}
k
n
,
d
(
x
)
=
{
1
2
d
(
1
+
1
2
n
)
(
1
-
(
x
d
)
2
n
)
,
|
x
|
≤
d
0
,
|
x
|
>
d
{\ displaystyle k_ {n, d} (x) = {\ begin {cases} {\ frac {1} {2d}} \ left (1 + {\ frac {1} {2n}} \ right) \ left ( 1- \ left ({\ frac {x} {d}} \ right) ^ {2n} \ right) &, | x | \ leq d \\ 0 &, | x |> d \ end {cases}}}
For this applies
∫
-
∞
∞
x
2
n
k
n
,
d
(
x
)
d
x
=
d
2
n
4th
n
+
1
.
{\ displaystyle \ int _ {- \ infty} ^ {\ infty} x ^ {2n} k_ {n, d} (x) \, {\ mathrm {d}} x = {\ frac {d ^ {2n} } {4n + 1}}.}
The kernel given by Epanechnikov himself normalizes this integral for to one. So for we choose :
n
=
1
{\ displaystyle n = 1}
(
d
4th
+
1
)
2
=
1
{\ displaystyle \ left ({\ tfrac {d} {\ sqrt {4 + 1}}} \ right) ^ {2} = 1}
k
E.
: =
k
1
,
5
{\ displaystyle k_ {E}: = k_ {1, {\ sqrt {5}}}}
k
E.
(
x
)
=
{
3
4th
5
(
1
-
x
2
5
)
,
|
x
|
≤
5
0
,
|
x
|
>
5
{\ displaystyle k_ {E} (x) = {\ begin {cases} {\ frac {3} {4 {\ sqrt {5}}}} \ left (1 - {\ frac {x ^ {2}} { 5}} \ right) &, | x | \ leq {\ sqrt {5}} \\ 0 &, | x |> {\ sqrt {5}} \ end {cases}}}
Sometimes the core is also referred to as the Epanechnikov core, which accordingly does not meet property 3:
d
=
1
{\ displaystyle d = 1}
k
E.
(
x
)
=
{
3
4th
(
1
-
x
2
)
,
|
x
|
≤
1
0
,
|
x
|
>
1
{\ displaystyle k_ {E} (x) = {\ begin {cases} {\ frac {3} {4}} (1-x ^ {2}) &, | x | \ leq 1 \\ 0 &, | x |> 1 \ end {cases}}}
Web links
swell
↑ VA Epanechnikov: Non-Parametric Estimation of a Multivariate Probability Density . In: Theory of Probability and its Applications , 1969, p. 156
<img src="https://de.wikipedia.org/wiki/Special:CentralAutoLogin/start?type=1x1" alt="" title="" width="1" height="1" style="border: none; position: absolute;">