fetch: Extract estimates from model output.

In StatisticsNZ/demest: Bayesian Demographic Estimation and Forecasting

Extract estimates from model output.

Description

Functions estimateModel, estimateCounts, and estimateAccount, and their predict counterparts, send their output to a simple database specified by filename. Function fetch extracts estimates from this database.

Usage

fetch(filename, where = character(), iterations = NULL, impute = TRUE)

Arguments

filename	The filename used by the estimate function.
where	A character vector describing the path to the item to be extracted.
iterations	A vector of positive integers giving the iterations to be extracted if an item has multiple iterations.
impute	Logical. Whether to impute missing values. Defaults to TRUE.

Details

Estimates are stored in a hierarchical structure. The structure be viewed with function listContents. Function fetch uses a description of the path through the hierarchical structure to locate estimates. The where argument to fetch is a character vector, giving names of nodes in the hierarchy. Partial matching is used with the names of the nodes, so names can be shortened, provided they are still long enough to uniquely identify a node. (To make code self-documenting, it is still best to use the full names in production code.)

Using the iterations to select a subset of iterations can be useful if the object being extracted is large relative to memory or if calculations are running slowly.

Where possible, estimateModel, estimateCounts, and estimateAccount avoid imputing missing values during model fitting, since the imputed values are typically highly correlated with other unknown quantities, which can slow convergence. If a batch of estimates has missing values and if impute is TRUE the missing values will be imputed at the time they are fetched.

Value

Parameters that were estimated from the data typically have class DemographicArray and have a dimension with dimtype "iteration". Other elements stored in object have a variety of classes.

References

The weak identification of main effects and interactions is discussed in

Gelman, A. (2005). Analysis of variance: Why it is more important than ever. Annals of Statistics, 33(1):1-53.

and

Nelder, J. A. (1994). The statistics of linear models: Back to basics. Statistics and Computing. 4: 221-234.

See Also

fetch is used to extract output from functions estimateModel, estimateCounts, and estimateAccount. Function listContents shows the internal structure of the output, which is useful for constructing the where argument to fetch.

Examples

deaths <- demdata::VADeaths2
popn <- demdata::VAPopn
deaths <- round(deaths)
deaths <- Counts(deaths)
popn <- Counts(popn)
filename <- tempfile()
estimateModel(Model(y ~ Poisson(mean ~ age * sex)),
              y = deaths,
              exposure = popn,
              filename = filename,
              nBurnin = 20,
              nSim = 20,
              nChain = 2,
              parallel = FALSE)
## the model hasn't been run for long enough,
## but we'll extract some results anyway

## examine structure
listContents(filename)

## extract means from likelihood model
rate <- fetch(filename,
              where = c("model", "likelihood", "rate"))
plot(rate)

## only supply enough of component of 'where' to identify
rate <- fetch(filename,
              where = c("mo", "l", "r"))

## extract all iterations of intercept
fetch(filename,
      where = c("model", "prior", "(Intercept)"))

## extract every fifth iteration
fetch(filename,
      where = c("model", "prior", "(Intercept)"),
      iterations = seq(from = 5, by = 5, to = 40))

StatisticsNZ/demest documentation built on Nov. 2, 2023, 7:56 p.m.