runMCMC: Run MCMC iterations

In dylanirion/MSctmm: Multi State Continous-Time Movement Modeling

Run MCMC iterations

Description

Run MCMC iterations

Usage

runMCMC(
  track,
  nbStates,
  nbIter,
  burnin,
  inits,
  fixed,
  priors,
  props,
  tunes,
  Hmat,
  updateState = TRUE,
  adapt = FALSE,
  model = NA
)

Arguments

track	data.frame. Data, with columns x, y, time, and ID
nbStates	integer. Number of states
nbIter	integer. Number of iterations for the MCMC
inits	list. Initial parameters: tau_pos: vector. Initial tau_{pos} for each state, of length nbStates tau_vel: vector. Initial tau_{vel} for each state, of length nbStates sigma: vector. Initial sigma for each state, of length nbStates for isotropic covariance or 3 * nbStates for anisotropic covariance mu: list. Initial OUF range centre coordinate pairs ⁠(x, y)⁠, with NA for IOU states Q: state: vector. Initial state sequence, length nrow(track) rateparam: vector. Length dependent on model choice
fixed	list of fixed parameters: tau_pos: vector. Fixed values of tau_{pos} for each state with NA for parameters to estimate. Length nbStates tau_vel: vector. Fixed values of tau_{vel} for each state with NA for parameters to estimate. Length nbStates sigma: vector. Fixed values of sigma for each state with NA for parameters to estimate. Length nbStates for isotropic covariance, or 3 * nbStates for anisotropic covariance mu: list. Fixed OUF range centre coordinate pairs ⁠(x, y)⁠, with NA for IOU states or pairs to estimate Q: Unused knownStates: vector. Known states with NA for those to estimate. Length nrow(track) kappa: integer. Maximum transition rate to bound rates when they are modelled. Length 1
priors	list. Parameters of prior distributions, with components: func: list of distribution function names, of length 5 * nbStates or 7 * nbStates for movement parameters + length(rateparam) when estimating rate parameters rateparam by MH args: list of list containg function args for distribution functions, of length 5 * nbStates or 7 * nbStates for movement parameters + length(rateparam) when estimating rate parameters rateparam by MH shape: vector. Shapes of gamma priors for the transition rates when Gibbs sampling rate: vector. Rates of gamma priors for the transition rates when Gibbs sampling con: vector. Concentrations of Dirichlet priors for transition probabilities when Gibbs sampling
props	list. Parameters of proposal distributions, with components: S: matrix. Initial value for the lower triangular matrix of RAM algorithm, so that the covariance matrix of the proposal distribution is ⁠SS'⁠. Dimensions ⁠(5 * nbStates, 5 * nbStates)⁠ when not modelling rate parameters (model is NA) and ⁠(5 * nbStates + length(rateparam), 5 * nbStates + length(rateparam))⁠ otherwise. updateLim: vector. Two values: min and max length of updated state sequence updateProbs: vector. Probabilities for each element of updateLim[1]:updateLim[2] (if NULL, all values are equiprobable)
tunes	list. Tuning parameters, with components: thinStates: integer. Thinning factor for the posterior state sequences (needed because of memory limitations)
Hmat	matrix. Observation error variance (four columns, and one row for each row of data)
updateState	logical. If FALSE, the state process is not updated (for exploratory analysis only, useful for testing single state models)
adapt	integer. If adapt > 0, use the the Robust Adaptive Metropolis (RAM) algorithm by Vihola (2012) to update the proposal distribution for each parameter at each iteration (up to adapt iterations) to a target acceptance rate of 23.4%.
model	experimental

References

Michelot, T., Blackwell, P.G. (2019). State‐switching continuous‐time correlated random walks. Methods Ecol Evol, 10: 637-649. doi:10.1111/2041-210X.13154

Vihola, M. (2012). Robust adaptive Metropolis algorithm with coerced acceptance rate. Stat Comput, 22: 997-1008. doi:10.1007/s11222-011-9269-5

dylanirion/MSctmm documentation built on Sept. 27, 2024, 3:41 a.m.