And just then I realized the problem. nknots need to be length(knots).
Otherwise knots are deleted.
I am not so sure this works equally well as my original loess fit
though. The fit I get with cobs is highly dependent on the "knot step
size". At 0.4 for example it seems ok. At 0.3 I get points with
basically zero width intervals. At 0.2 it breaks ("There is at least one
pair of adjacent knots that contains no observation."). Not so great for
my use case where I need to build many models unsupervised.
- Jan.
On 08/19/2014 09:44 AM, Jan Stanstrup wrote:
> Sorry. I have updated the code to have include the knot selection
> (https://github.com/stanstrup/retpred_shiny/blob/master/retdb_admin/make_predictions_CI_tests.R).
>
> I am working on the "Good data" at the moment.
>
>
> - Jan.
>
>
>
>
> On 08/18/2014 08:14 PM, David Winsemius wrote:
>> I had that result sometimes when testing as well. You don't offer any code
>> so there's nothing I can do to follow-up.
>>
>
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
