fit_model {disaggregation}R Documentation

Fit the disaggregation model

Description

fit_model function takes a disag_data object created by prepare_data and performs a Bayesian disaggregation fit.

Usage

fit_model(
  data,
  priors = NULL,
  family = "gaussian",
  link = "identity",
  iterations = 100,
  field = TRUE,
  iid = TRUE,
  hess_control_parscale = NULL,
  hess_control_ndeps = 1e-04,
  silent = TRUE
)

disag_model(
  data,
  priors = NULL,
  family = "gaussian",
  link = "identity",
  iterations = 100,
  field = TRUE,
  iid = TRUE,
  hess_control_parscale = NULL,
  hess_control_ndeps = 1e-04,
  silent = TRUE
)

Arguments

data

disag_data object returned by prepare_data function that contains all the necessary objects for the model fitting

priors

list of prior values

family

likelihood function: gaussian, binomial or poisson

link

link function: logit, log or identity

iterations

number of iterations to run the optimisation for

field

logical. Flag the spatial field on or off

iid

logical. Flag the iid effect on or off

hess_control_parscale

Argument to scale parameters during the calculation of the Hessian. Must be the same length as the number of parameters. See optimHess for details.

hess_control_ndeps

Argument to control step sizes during the calculation of the Hessian. Either length 1 (same step size applied to all parameters) or the same length as the number of parameters. Default is 1e-3, try setting a smaller value if you get NaNs in the standard error of the parameters. See optimHess for details.

silent

logical. Suppress verbose output.

Details

The model definition

The disaggregation model make predictions at the pixel level:

link(pred_i) = \beta_0 + \beta X + GP(s_i) + u_i

And then aggregates these predictions to the polygon level using the weighted sum (via the aggregation raster, agg_i):

cases_j = \sum_{i \epsilon j} pred_i \times agg_i

rate_j = \frac{\sum_{i \epsilon j} pred_i \times agg_i}{\sum_{i \epsilon j} agg_i}

The different likelihood correspond to slightly different models (y_j is the response count data):

Specify priors for the regression parameters, field and iid effect as a single list. Hyperpriors for the field are given as penalised complexity priors you specify \rho_{min} and \rho_{prob} for the range of the field where P(\rho < \rho_{min}) = \rho_{prob}, and \sigma_{min} and \sigma_{prob} for the variation of the field where P(\sigma > \sigma_{min}) = \sigma_{prob}. Also, specify pc priors for the iid effect

The family and link arguments are used to specify the likelihood and link function respectively. The likelihood function can be one of gaussian, poisson or binomial. The link function can be one of logit, log or identity. These are specified as strings.

The field and iid effect can be turned on or off via the field and iid logical flags. Both are default TRUE.

The iterations argument specifies the maximum number of iterations the model can run for to find an optimal point.

The silent argument can be used to publish/suppress verbose output. Default TRUE.

Value

A list is returned of class disag_model. The functions summary, print and plot can be used on disag_model. The list of class disag_model contains:

obj

The TMB model object returned by MakeADFun.

opt

The optimized model object returned by nlminb.

sd_out

The TMB object returned by sdreport.

data

The disag_data object used as an input to the model.

model_setup

A list of information on the model setup. Likelihood function (family), link function(link), logical: whether a field was used (field) and logical: whether an iid effect was used (iid).

Examples

## Not run: 
 polygons <- list()
 for(i in 1:100) {
  row <- ceiling(i/10)
  col <- ifelse(i %% 10 != 0, i %% 10, 10)
  xmin = 2*(col - 1); xmax = 2*col; ymin = 2*(row - 1); ymax = 2*row
  polygons[[i]] <- rbind(c(xmin, ymax), c(xmax,ymax), c(xmax, ymin), c(xmin,ymin))
 }

 polys <- do.call(raster::spPolygons, polygons)
 response_df <- data.frame(area_id = 1:100, response = runif(100, min = 0, max = 10))
 spdf <- sp::SpatialPolygonsDataFrame(polys, response_df)

 r <- raster::raster(ncol=20, nrow=20)
 r <- raster::setExtent(r, raster::extent(spdf))
 r[] <- sapply(1:raster::ncell(r), function(x) rnorm(1, ifelse(x %% 20 != 0, x %% 20, 20), 3))
 r2 <- raster::raster(ncol=20, nrow=20)
 r2 <- raster::setExtent(r2, raster::extent(spdf))
 r2[] <- sapply(1:raster::ncell(r), function(x) rnorm(1, ceiling(x/10), 3))
 cov_rasters <- raster::stack(r, r2)

 cl <- parallel::makeCluster(2)
 doParallel::registerDoParallel(cl)
 test_data <- prepare_data(polygon_shapefile = spdf, 
                           covariate_rasters = cov_rasters)
 parallel::stopCluster(cl)
 foreach::registerDoSEQ()
                         
 result <- fit_model(test_data, iterations = 2)
 
## End(Not run)


[Package disaggregation version 0.1.4 Index]