FitOCTlib: Stan-based functions to fit OCT signals

Description Usage Arguments Details Value Author(s)

Define/estimate normal multivariate prior pdf for exponential decay parameters.

estimateExpPrior(
  x,
  uy,
  dataType = 2,
  priorType = "mono",
  out,
  ru_theta = 0.05,
  eps = 0.001,
  nb_chains = 4,
  nb_warmup = 800,
  nb_iter = nb_warmup + 200
)

`x`	numeric vector of depths
`uy`	numeric vector of uncertainties
`dataType`	integer defining the type of data (1:amplitude or 2:intensity) )
`priorType`	string defining the type of prior ('mono' or 'abc')
`out`	output list from `fitMonoExp`
`ru_theta`	optional positive real defining the relative uncertainty on parameters (priorType='mono')
`eps`	tolerance parameter for the moments matching method (priorType='abc')
`nb_chains`	number of MCMC chains (priorType='abc')
`nb_warmup`	number of warmup steps (priorType='abc')
`nb_iter`	number of steps (priorType='abc')

Provides two ways to buil a normal multivariate prior for the exponential decay parameters of the ExpGP model. The mean value is in both cases the MAP issued by fitMonoExp. For the covariance matrix one has two options:

priorType='mono': a covariance matrix is built from the correlation matrix estimated by the fitMonoExp model and a relative uncertainty on the parameters (ru_theta). This accounts for the fact that the parameters uncertainties provided by fitMonoExp are not reliable, as they are issued from an invalid model.
priorType='abc': the parameters uncertainties/variances are optimized by a moments matching strategy: (1) the 2-sigma prediction uncertainty has to match the 95-th quantile of the absolute errors of the fitMonoExp model (this statistics is weighted by uy); and (2) the standard deviation of the prediction uncertainty has to be as small as possible. This prior assumes a diagonal covariance matrix.

A list containing: the center, covariance matrix of the prior pdf and, for priorType = 'abc', a stats list containing the constraint Sobs and realized Ssim statistics.

Pascal PERNOT

ppernot/FitOCTlib documentation built on April 11, 2020, 1:55 a.m.