Bayesian Model Averaging

Share:

Description

Bayesian Model Averaging

Perform Bayesian Model Averaging. We concentrate on the chain with temperature=1 , i.e the untempered posterior, to study the distribution over the model choices and perform model averaging. We consider as present the species that have a posterior probability greater than 0.9. We then fit the mixture model with these species in order to obtain relative abundances and read classification probabilities. A tab seperated file that has a species summary is produced, as well as log-likelihood traceplots and cumulative histogram plots.

bayes.model.aver.explicit is the same function as bayes.model.aver with a more involved syntax.

Usage

1
2
3
4
bayes.model.aver(step2, step3, taxon.name.map = NULL, poster.prob.thr = 0.9)

bayes.model.aver.explicit(result, pij.sparse.mat, read.weights, outDir,
  gen.prob.unknown, taxon.name.map = NULL, poster.prob.thr = 0.9)

Arguments

step3

list. The output from parallel.temper(), i.e the third step of the pipeline. Alternatively, it can be a character string containing the path name of the ".RData" file where step3 list was saved.

step2

list. The output from reduce.space(), i.e the second step of the pipeline. Alternatively, it can be a character string containing the path name of the ".RData" file where step2 list was saved.

taxon.name.map

The 'names.dmp' taxonomy names file, mapping each taxon identifier to the corresponding scientific name. It can be downloaded from ftp://ftp.ncbi.nih.gov/pub/taxonomy/taxdump.tar.gz

poster.prob.thr

Posterior probability of presence of species threshold for reporting in the species summary.

result

The list produced by parallel.temper() (or paraller.temper.nucl()) . It holds a detailed record for each chain, what moves were proposed, which were accepted and which were rejected as well the log-likelihood through the iterations.

pij.sparse.mat

see ?reduce.space

read.weights

see ?reduce.space

gen.prob.unknown

see ?reduce.space

outDir

see ?reduce.space

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
## See vignette for more details

## Not run: 
# Either load the object created by previous steps
data(step2)   ## example output of step2, i.e reduce.space()
data(step3)   ## example ouput of step3, i.e  parallel.temper()
step4<-bayes.model.aver(step2=step2, step3=step3, taxon.name.map="pathtoFile/taxon.file")

# or alternatively point to the location of the step2.RData and step3.RData objects
step4<-bayes.model.aver(step2="pathtoFile/step2.RData", step3="pathtoFile/step3.RData",
                        taxon.name.map="pathtoFile/taxon.file")


## End(Not run)