Thread Subject: Give me a Regression Problem

On Jul 16, 6:03=A0pm, paulvbi...@gmail.com wrote:
> > With a severely skewed distribution, maybe another summary
> > statistic (mode, median?) would be better.
>
> *******************************
>
> Yes that is a very good suggestion
>
> will see how I can do that
>
> thanks Greg a good one for sure
> **********************************
>
> > > As you may or may not know the Sensitivity S =A0=3D=3D> S(X) =A0so it=
 very
> > > much depends where you evaluate the sensitivity in the design space o=
f
> > > X. =A0I chose the average but would vary and sometimes considerably i=
f
> > > different areas of the X space
>
> > That is why I recommend averaging over the entire training set
> > or, if that is too expensive, averaging over cluster centers.
>
> ****************************
> It is not the expense, it is the fact or suppostion that you will get
> a whack of variation and then "who is on first"

The region/cluster that has the most data points will be weighted more
highly.

> that is my worry, you comment about the Median is a good one that I
> can buy
>
> for a given data set we will have an area of interest more than the
> other parts of the space and therefore
> the Sensitivity needs to be tied to where a customer has his/her
> highest interest,

That's where the weighted cluster center approach might be the
most useful.

Hope this helps.

Greg