Validation data critically small (<50) for reliable validation

[Shujiang Kang]

Dear Tom,

First, really appreciate you for the wonderful package, GSIF. It is so convenient and friendly. Currently, I am using it for some detailed soil mapping in precision ag, and got the following error info as I use 'fit.gstatModel'  method = "quantregForest".  

"Generating predictions using the trend model (RK method)...

Running GLM cross-validation without any extra model-fitting...

Creating an object of class "SpatialPredictions"

Error in validObject(.Object) : 

  invalid class “SpatialPredictions” object: Validation data critically small (<50) for reliable validation"

I guess that my observation size is too small <50; however, we do need to map a small plot with limited sampled data. Could we adjust the threshold, < 50, somewhere or modify the package?

Thank you very much in advance,

Shujiang

[de Sousa, Luis]

Dear Shujiang,

Tom no longer works at ISRIC, therefore he might not be reading these messages, or might take a while before he answers back. I suggest you contact Tom directly for issues related to the GSIF package.

Best regards.?



--

Luís Moreira de Sousa

ISRIC - World Soil Information

Droevendaalsesteeg 3, Building 101

6708 PB Wageningen

The Netherlands

t: +31 643 84 40 91

e: luis.d...@wur.nl

Ring<https://ring.cx/>: luis_de_sousa

________________________________
From: global-soil...@googlegroups.com <global-soil...@googlegroups.com> on behalf of Shujiang Kang <skang2...@gmail.com>
Sent: 02 May 2018 16:42
To: Global Soil Information
Subject: Validation data critically small (<50) for reliable validation

--
You received this message because you are subscribed to the Google Groups "Global Soil Information" group.
To unsubscribe from this group and stop receiving emails from it, send an email to global-soil-infor...@googlegroups.com<mailto:global-soil-infor...@googlegroups.com>.
Visit this group at https://groups.google.com/group/global-soil-information.
For more options, visit https://groups.google.com/d/optout.