problem about GATK indel VQSR

[wanguan2000]

java -Xmx4g -jar GenomeAnalysisTK-1.1-36-g367bbee/GenomeAnalysisTK.jar \
-T VariantRecalibrator \
-R ucsc.hg19.fasta \
-B:input,VCF s1_raw_indel.vcf \
-B:training,VCF,known=true,training=true,truth=true,prior=12.0 indels_mills_devine.hg19.sites.vcf \
-an QD -an FS -an HaplotypeScore -an ReadPosRankSum \
-mode INDEL \
-recalFile s1_raw_indel.recal \
-tranchesFile s1_raw_indel.tranches \
-rscriptFile s1_raw_indel_plots.R

when I run this GATK command followed : http://www.broadinstitute.org/gsa/wi...th_the_GATK_v3
Indel specific recommendations

We use the (Mills, Devine, Genome Research, 2011) dataset as training data when modeling indels with the VQSR. This dataset is available in the GATK resource bundle.
Arguments for VariantReacalibrator:
-B:training,VCF,known=true,training=true,truth=true,prior=12.0 indels_mills_devine.b37.sites.vcf \
-an QD -an FS -an HaplotypeScore -an ReadPosRankSum -an InbreedingCoeff \
-mode INDEL \

I got this error:
##### ERROR MESSAGE: NaN LOD value assigned. Clustering with this few variants and these annotations is unsafe.

please help me,and the snp VariantRecalibrator is all right.

[wanguan2000]

ps: s1_raw_indel.vcf file contains total of 8454 indels, are the number is few variants and these annotations is unsafe?

[liu_xt005]

Gaussian modeling might not be appropriate for indels. I did manual filtering following a pipeline for exome sequence posted here.