fit_GP: Fit MLN Gaussian process model to a given host's series using...
In kimberlyroche/rulesoflife:

fit_GP

R Documentation

Fit MLN Gaussian process model to a given host's series using fido

Description

Fit MLN Gaussian process model to a given host's series using fido

Usage

fit_GP(
  sname,
  counts,
  metadata,
  output_dir,
  MAP = TRUE,
  days_to_min_autocorrelation = 90,
  diet_weight = 0,
  var_scale_taxa = 1,
  var_scale_samples = 1,
  concentration = 10,
  age_min = -Inf,
  age_max = Inf,
  use_adam = FALSE,
  scramble_sample = FALSE,
  scramble_spacing = FALSE,
  scramble_order = FALSE
)

Arguments

`sname`	host short name indicating which baboon's series to fit
`counts`	filtered 16S count table (taxa x samples)
`metadata`	annotations data.frame
`output_dir`	specifies the subdirectory of the model fits directory in which to save the fitted model output
`MAP`	flag indicating whether or not to return (single) MAP estimate
`days_to_min_autocorrelation`	used in the calculation of bandwidth for the kernel over samples that models autocorrelation; indicates the number of days at which the squared exponential correlation decays to a minimum (here, 0.1)
`diet_weight`	relative contribution of first three diet PCs to covariance across samples
`var_scale_taxa`	scale of the prior associated with the covariance over taxa
`var_scale_samples`	scale of the hyperparameter associated with the covariance over samples
`concentration`	optional concentration parameter for the prior over taxon-taxon covariance
`age_min`	optional mimimum age; samples collected before this age for this animal will be omitted
`age_max`	optional maximum age; samples collected after this age for this animal will be omitted
`use_adam`	optimize with Adam (occasionally this converges more reliably
`scramble_sample`	optional flag to scramble taxonomic identity within a sample that default L-BFGS)
`scramble_spacing`	optional flag to scramble a host's sampling frequency
`scramble_order`	optional flag to scramble a host's sample order