modelMethods: Model Function Methods: Parameter Names, Bounds and Initial...

modelMethodsR Documentation

Model Function Methods: Parameter Names, Bounds and Initial Values

Description

A large number of functions for obtaining and modifying the parameters of likelihood models made in paleotree. These functions allow users to obtain or set parameter names, or obtain and set parameter bounds, both of which are treated as an attribute of the function class used by paleotree. In practice, this allows users to quickly obtain parameter names and upper and lower values for use in bounded optimizers, including reasonable starting values.

Usage

parnames(x, ...)

## S3 method for class 'paleotreeFunc'
parnames(x, ...)

## S3 method for class 'constrained'
parnames(x, ...)

parnames(x) <- value

## S3 replacement method for class 'constrained'
parnames(x) <- value

## S3 replacement method for class 'paleotreeFunc'
parnames(x) <- value

parbounds(x, ...)

## S3 method for class 'paleotreeFunc'
parbounds(x, ...)

## S3 method for class 'constrained'
parbounds(x, ...)

parbounds(x) <- value

## S3 replacement method for class 'constrained'
parbounds(x) <- value

## S3 replacement method for class 'paleotreeFunc'
parbounds(x) <- value

parLower(x, ...)

## S3 method for class 'constrained'
parLower(x, ...)

## S3 method for class 'paleotreeFunc'
parLower(x, ...)

parLower(x) <- value

## S3 replacement method for class 'constrained'
parLower(x) <- value

## S3 replacement method for class 'paleotreeFunc'
parLower(x) <- value

parUpper(x, ...)

## S3 method for class 'constrained'
parUpper(x, ...)

## S3 method for class 'paleotreeFunc'
parUpper(x, ...)

parUpper(x) <- value

## S3 replacement method for class 'constrained'
parUpper(x) <- value

## S3 replacement method for class 'paleotreeFunc'
parUpper(x) <- value

parInit(x, ...)

## S3 method for class 'constrained'
parInit(x, ...)

## S3 method for class 'paleotreeFunc'
parInit(x, ...)

Arguments

x

A function of S3 class 'paleotreeFunc' with all necessary attributes expected of that class, which include parameter names and upper and lower bounds. As I have deliberately not exported the function which creates this class, it should be impossible for regular users to obtain such objects easily without using one of the make functions, which automatically output a function of the appropriate class and attributes.

...

'Ignored arguments to future methods' (i.e. for diversitree). Kept here only so constrainParPaleo is kept as close to the parent method in ddiversitree as possible.

value

The new value with which to replace the parameter names or bounds. Must be a vector of the same length as the number of parameters. For parbounds, must be a list composed of two vectors.

Details

Parameter names cannot be changed for a constrained function.

The parInit function calls the bounds for each parameter and gives a randomly selected value selected from a uniform distribution, using the parameter bounds for each parameter as the bounds on the uniform distribution. This users a shorthand to quickly generate initial parameter values which are within the set bounds, for use in functions such as optim. The random sampling of initial values allows a user to quickly assess if initial parameter values affect the optimization by simply rerunning the function on new values. Infinite initial parameter values (resulting from infinite bounds) are discarded, and replaced with the lower bound value (assuming only upper bounds are infinite...). Some randomly selected initial parameter values may be too high (due to the liberal upper bounds I set for parameters in many of the likelihood functions) and thus users should always try slightly different values to see if the resulting maximum likelihood parameter values change.

As parInit depends on the upper and lower bounds attribute, no function is offered to allow it to be replaced (as there is nothing to replace!).

Value

Returns the sought parameter names, bounds or initial values or (for the replacement methods) returns a modified function with the respective attributes altered.

Author(s)

These functions are strongly based on or inspired by the argnames functions provided for handling models in Rich Fitzjohn's library diversitree, but the functions presented here are derivations written by David Bapst.

See Also

These model methods were introduced to interact with the new model framework introduced in paleotree version >1.9, in particular to interface with constrainParPaleo.

Examples

#example with make_durationFreqCont
set.seed(444)
record <- simFossilRecord(p = 0.1, q = 0.1, nruns = 1,
	nTotalTaxa = c(30,40), nExtant = 0)
taxa <- fossilRecord2fossilTaxa(record)
rangesCont <- sampleRanges(taxa,r = 0.5)
likFun <- make_durationFreqCont(rangesCont)

#get parameter names
parnames(likFun)

#get the bounds for those parameters
parbounds(likFun)

#can also get these seperately
parLower(likFun)
parUpper(likFun)

#initial parameter values
parInit(likFun)   #arbitrary midway value between par bounds

#can then use these in optimizers, such as optim with L-BFGS-B
#see the example for make_durationFreqCont

#renaming parameter names
likFun2 <- likFun
parnames(likFun2) <- c("extRate","sampRate")
parnames(likFun2)
#test if reset correctly
parnames(likFun2) == c("extRate","sampRate")
#also works for constrained functions
constrainFun <- constrainParPaleo(likFun,q.1~r.1)
parnames(constrainFun)
#also modified the parameter bounds, see!
parbounds(constrainFun)
parInit(constrainFun)
#but cannot rename parameter for constrained function!


dwbapst/paleotree documentation built on Aug. 30, 2022, 6:44 a.m.