Negative mse values

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
3 messages Options
Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Negative mse values

R-sig-geo mailing list
Hi  dear friends.

I was trying to smooth a spatial point pattern with kernel function but I am a bit confused .
I need some  explanation about negative MSE values obtained ( used for selecting optimum bandwidth)  by using mse2d function of "splancs " package.
As the definition of mse function  for one dimension I know it might not take negative values.
However I have no idea for the two dimensional definition of mse for a spatial point pattern. Could the mse values be negative in this case? I found an example taking minimum negative value of mse fort he optimum bandwidth which is the lowest negative value
Thanks in advance.


Res. Asst. Cenk Icoz
Statistics Department ,Anadolu University, Turkey



        [[alternative HTML version deleted]]

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Negative mse values

Rolf Turner

On 27/03/17 21:28, Cenk İÇÖZ via R-sig-Geo wrote:

> Hi  dear friends.
>
> I was trying to smooth a spatial point pattern with kernel function
> but I am a bit confused .
> I need some explanation about negative MSE values obtained ( used
> forselecting optimum bandwidth) by using mse2d function of "splancs " package.
> As the definition of mse function for one dimension I know it might
> not take negative values.
> However I have no idea for the two dimensional definition of mse for
> aspatial point pattern. Could the mse values be negative in this case? I
> found an example taking minimum negative value of mse fort he optimum
> bandwidth which is the lowest negative value

The short answer is that the value returned by mse2d() is not actually
the MSE but rather MSE minus a data dependent constant.  So this can
"legitimately" be negative.

The off-setting constant "does no harm" since interest lies in
determining the optimum bandwidth, so it is the relative sizes of the
values produced that are of interest.

It has been suggested to me in the past (by Barry Rowlingson, one of the
original authors of splancs) that the function bw.diggle() from the
spatstat package may be more reliable than mse2d().  The former function
uses a somewhat different calculation procedure, whence the results of
the two functions are not exactly comparable.

Note that bw.diggle() is expressed in terms of "sigma" whereas mse2d()
is expressed in terms of "h" where sigma = h/2. So if mse2d() gives an
"optimal" value of 3, one would *roughly* expect bw.diggle() to give an
optimal value of 1.5.

Note also that estimating an "optimal" bandwidth is a pretty inexact
endeavour under the best of circumstances.  The smooth surface to be
fitted is an ill-determined creature and the bandwidth that gives the
best fit is even more ill-determined.

cheers,

Rolf Turner

--
Technical Editor ANZJS
Department of Statistics
University of Auckland
Phone: +64-9-373-7599 ext. 88276

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Negative mse values

R-sig-geo mailing list
Thanks Rolf. Now it is settled. I get it.
I agree they are all ill determined endeavours.
Nonetheless, some has to determine a bandwidth to see the big picture.
What else we can do? . Parametric methods (point processes) or others do not depict the characteristic of the pattern exactly.
Thanks again.


Res. Asst. Cenk Icoz
Statistics Department , Anadolu University, Turkey


-----Original Message-----
From: Rolf Turner [mailto:[hidden email]]
Sent: Monday, March 27, 2017 1:24 PM
To: Cenk İÇÖZ <[hidden email]>
Cc: [hidden email]
Subject: Re: [R-sig-Geo] Negative mse values


On 27/03/17 21:28, Cenk İÇÖZ via R-sig-Geo wrote:

> Hi  dear friends.
>
> I was trying to smooth a spatial point pattern with kernel function
> but I am a bit confused .
> I need some explanation about negative MSE values obtained ( used
> forselecting optimum bandwidth) by using mse2d function of "splancs " package.
> As the definition of mse function for one dimension I know it might
> not take negative values.
> However I have no idea for the two dimensional definition of mse for
> aspatial point pattern. Could the mse values be negative in this case? I
> found an example taking minimum negative value of mse fort he optimum
> bandwidth which is the lowest negative value

The short answer is that the value returned by mse2d() is not actually
the MSE but rather MSE minus a data dependent constant.  So this can
"legitimately" be negative.

The off-setting constant "does no harm" since interest lies in
determining the optimum bandwidth, so it is the relative sizes of the
values produced that are of interest.

It has been suggested to me in the past (by Barry Rowlingson, one of the
original authors of splancs) that the function bw.diggle() from the
spatstat package may be more reliable than mse2d().  The former function
uses a somewhat different calculation procedure, whence the results of
the two functions are not exactly comparable.

Note that bw.diggle() is expressed in terms of "sigma" whereas mse2d()
is expressed in terms of "h" where sigma = h/2. So if mse2d() gives an
"optimal" value of 3, one would *roughly* expect bw.diggle() to give an
optimal value of 1.5.

Note also that estimating an "optimal" bandwidth is a pretty inexact
endeavour under the best of circumstances.  The smooth surface to be
fitted is an ill-determined creature and the bandwidth that gives the
best fit is even more ill-determined.

cheers,

Rolf Turner

--
Technical Editor ANZJS
Department of Statistics
University of Auckland
Phone: +64-9-373-7599 ext. 88276
_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Loading...