[R] How does predict.loess work?

[David Winsemius]

On Apr 3, 2012, at 4:12 PM, Recher She wrote:

> Dear R community,
>
> I am trying to understand how the predict function, specifically, the
> predict.loess function works.
>
> I understand that the loess function calculates regression  
> parameters at
> each data point in 'data'.
>
> lo <- loess ( y~x, data)

Well, it produces a fitted-value at each point. Whether there are  
parameters at each point might depend on the degree of the fit.

>
> p <- predict (lo, newdata)
>
> I understand that the predict function predicts values for 'newdata'
> according to the loess regression parameters. How does predict.loess  
> do
> this in the case that 'newdata' is different from the original data  
> x? How
> does the interpolation take place?

Type this at your console:

getAnywhere(predict.loess)

And after seeing that an additional functions is called type this:

getAnywhere(predLoess)

And then you will see that you have descended into a C function called  
'C_loess_dfitse'.

>
> 	[[alternative HTML version deleted]]
-- 

David Winsemius, MD
West Hartford, CT