MARSS (version 3.9)

MARSSmcinit: Monte Carlo Initialization

Description

Performs a Monte Carlo search for optimal initial conditions iterative maximization algorithms (MARSSkem and MARSSoptim). This is a utility function in the MARSS-package.

Usage

MARSSmcinit(MLEobj)

Arguments

MLEobj

An object of class marssMLE.

Value

A list with 8 matrices Z, A, R, B, U, Q, x0, V0, specifying initial values for parameters for iteration 1 of the EM algorithm. Note the output is the initial values for a marssMODEL in marss form.

Details

It is recommended that initial parameter values be set using MARSSmcinit(), particularly if the model is not a good fit to the data. This requires more compuation time, but reduces the chance of the algorithm terminating at a local maximum and not reaching the true MLEs.

Options for MARSSmcinit() may be set using MLEobj$control, as follows:

MLEobj$control$numInits

Number of random initial value draws.

MLEobj$control$numInitSteps

Maximum number of EM iterations for each random initial value draw.

MLEobj$control$boundsInits

Length 6 list. Each component is a length 2 vector of bounds on the uniform distributions from which initial values will be drawn (for A, B, U, and Z). For R and Q, variance-covariance matrices are generated from a wishart distribution with df=bound[1] and S=diag(bound[2],m). Note, random initial conditions are only used for parameters that are not fixed.

The default values for these are given in MARSSsettings.R and listed in MARSS.

References

The user guide: Holmes, E. E., E. J. Ward, and M. D. Scheuerell (2012) Analysis of multivariate time-series using the MARSS package. NOAA Fisheries, Northwest Fisheries Science Center, 2725 Montlake Blvd E., Seattle, WA 98112 Type RShowDoc("UserGuide",package="MARSS") to open a copy.

See Also

MARSSkem marssMLE MARSS

Examples

# NOT RUN {
  #Note doing a Monte-Carlo search takes a long, long time
  dat = t(harborSeal)
  dat = dat[c(2,nrow(dat)),]
  fit1=MARSS(dat, control=list(MCInit=TRUE))
  fit1
  #Show the inits that were used
  fit1$start
  #Try fewer initial start locations 
  #and different mean variance (0.1 instead of 1) for R and Q
  cntl.list = list(MCInit=TRUE, numInits=10, 
                   numInitSteps = 10, 
                   boundsInits=list(Q=c(1,0.1),R=c(1,0.1)))
  fit2=MARSS(dat, control=cntl.list)
  fit2
  #Show the inits that were used
  fit2$start
  #ignore the values for Z,B, and V0; those parameters are fixed
# }
# NOT RUN {
  
# }