R6 class holding sf grid data with data and analysis functions

Description

R6 class holding sf grid data with data and analysis functions

R6 class holding sf grid data with data and analysis functions

Details

Grid data consists of the computational grid over the area of interest. Outcomes and covariates are projected onto the grid, which can then be sent to the LGCP model.

If zcol is not specified then only the geometry is plotted, otherwise the covariates specified will be plotted. The user can also use sf plotting functions on grid$grid_data directly.

Case counts are generated for each grid cell for each time period. The user can specify the length of each time period; currently day, week, and month are supported.

The user must also specify the number of time periods to include with the laglength argument. The total number of time periods is the specified lag length counting back from the most recent case. The columns in the output will be named t1, t2,... up to the lag length, where the highest number is the most recent period.

ADDING COVARIATES Spatially-varying data only cov_data is an sf object describing covariate values for a set of polygons over the area of interest. The values are mapped onto grid_data. For each grid cell in grid_data a weighted average of each covariate listed in zcols is generated with weights either equal to the area of intersection of the grid cell and the polygons in cov_data (weight_type="area"), or this area multiplied by the population density of the polygon for population weighted (weight_type="pop"). Columns with the names in zcols are added to the output.

Temporally-varying only data cov_data is a data frame with number of rows equal to the number of time periods. One of the columns must be called t and have values from 1 to the number of time periods. The other columns of the data frame have the values of the covariates for each time period. See get_dow() for day of week data. A total of length(zcols)*(number of time periods) columns are added to the output: for each covariate there will be columns appended with each time period number. For example, dayMon1, dayMon2, etc.

Spatially and temporally varying data There are two ways to add data that vary both spatially and temporally. The final output for use in analysis must have a column for each covariate and each time period with the same name appended by the time period number, e.g. covariateA1,covariateA2,... If the covariate values for different time periods are in separate sf objects, one can follow the method for spatially-varying only data above and append the time period number using the argument t_label. If the values for different time periods are in the same sf object then they should be named as described above and then can be added as for spatially-varying covariates, e.g. zcols=c("covariateA1","covariateA2").

BAYESIAN MODEL FITTING The grid data must contain columns t*, giving the case count in each time period (see points_to_grid), as well as any covariates to include in the model (see add_covariates) and the population density. Otherwise, if the data are regional data, then the outcome counts must be in self$region_data

Our statistical model is a Log Gaussian cox process, whose realisation is observed on the Cartesian area of interest A and time period T. The resulting data are relaisations of an inhomogeneous Poisson process with stochastic intensity function \{λ{s,t}:s\in A, t \in T\}. We specify a log-linear model for the intensity:

λ(s,t) = r(s,t)exp(X(s,t)'γ + Z(s,t))

where r(s,t) is a spatio-temporally varying Poisson offset. X(s,t) is a length Q vector of covariates including an intercept and Z(s,t) is a latent field. We use an auto-regressive specification for the latent field, with spatial innovation in each field specified as a spatial Gaussian process.

The argument approx specifies whether to use a full LGCP model (approx='none') or whether to use either a nearest neighbour approximation (approx='nngp') or a "Hilbert space" approximation (approx='hsgp'). For full details of NNGPs see XX and for Hilbert space approximations see references (1) and (2). #' Priors For Bayesian model fitting, the priors should be provided as a list to the griddata object:

griddata$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(-5,rep(0,7)),
  prior_linpred_sd=c(3,rep(1,7))
)

where these refer to the priors: prior_lscale: the length scale parameter has a half-normal prior N(a,b^2)I[0,∞). The vector is c(a,b). prior_var: the standard deviation term has a half normal prior σ ~ N(a,b^2)I[0,∞). The vector is c(a,b). prior_linpred_mean and prior_linpred_sd: The parameters of the linear predictor. If X is the nT x Q matrix of covariates, with the first column as ones for the intercept, then the linear prediction contains the term X'γ. Each parameter in γ has prior γ_q ~ N(a_q,b_q^2). prior_linpred_mean should be the vector (a_1,a_2,...,a_Q) and prior_linpred_sd should be (b_1,b_2,...,b_Q).

MAXIMUM LIKELIHOOD MODEL FITTING The grid data must contain columns t*, giving the case count in each time period (see points_to_grid), as well as any covariates to include in the model (see add_covariates) and the population density. Otherwise, if the data are regional data, then the outcome counts must be in self$region_data. See lgcp_bayes() for more details on the model.

The argument approx specifies whether to use a full LGCP model (approx='none') or whether to use either a nearest neighbour approximation (approx='nngp')

Model fitting uses MCMC Maximum likelihood, which has three steps:

1. Sample random effects using MCMC. Using cmdstanr is recommended as it is much faster. The arguments mcmc_warmup and mcmc_sampling specify the warmup and sampling iterations for this step.

2. Fit fixed effect parameters using expectation maximisation.

3. Fit covariance parameters using expectation maximisation. This third step is the slowest. The NNGP approximation provides some speed improvements. Otherwise this step can be skipped if the covaraince parameters are "known".

EXTRACTING PREDICTIONS Three outputs can be extracted from the model fit, which will be added as columns to grid_data:

Predicted incidence: If type includes pred then pred_mean_total and pred_mean_total_sd provide the predicted mean total incidence and its standard deviation, respectively. pred_mean_pp and pred_mean_pp_sd provide the predicted population standardised incidence and its standard deviation.

Relative risk: if type includes rr then the relative risk is reported in the columns rr and rr_sd. The relative risk here is the exponential of the latent field, which describes the relative difference between expexted mean and predicted mean incidence.

Incidence risk ratio: if type includes irr then the incidence rate ratio (IRR) is reported in the columns irr and irr_sd. This is the ratio of the predicted incidence in the last period (minus t_lag) to the predicted incidence in the last period minus irr_lag (minus t_lag). For example, if the time period is in days then setting irr_lag to 7 and leaving t_lag=0 then the IRR is the relative change in incidence in the present period compared to a week prior.

Public fields

Methods

Public methods

• grid$new()

• grid$print()

• grid$plot()

• grid$points_to_grid()

• grid$add_covariates()

• grid$get_dow()

• grid$add_time_indicators()

• grid$lgcp_bayes()

• grid$lgcp_ml()

• grid$extract_preds()

• grid$hotspots()

• grid$aggregate_output()

• grid$scale_conversion_factor()

• grid$get_region_data()

• grid$variogram()

• grid$reorder()

• grid$data()

• grid$get_random_effects()

• grid$clone()

Method new()

Create a new griddata object

Produces a regular grid over an area of interest as an sf object

Given a contiguous boundary describing an area of interest, which is stored as an sf object of a regular grid within the limits of the boundary at $grid_data. The boundary is also stored in the object as $boundary

Usage

grid$new(poly, cellsize, verbose = TRUE)

Arguments

Returns

NULL

Examples

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)

Method print()

Prints the $grid_data sf object

Usage

grid$print()

Method plot()

Plots the grid data

Usage

grid$plot(zcol)

Arguments

Returns

A plot

Examples

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
g1$plot()

Method points_to_grid()

Generates case counts of points over the grid

Counts the number of cases in each time period in each grid cell

Given the sf object with the point locations and date output from create_points(), the functions will add columns to grid_data indicating the case count in each cell in each time period.

Usage

grid$points_to_grid(
  point_data,
  t_win = c("day"),
  laglength = 14,
  verbose = TRUE
)

Arguments

Returns

NULL

Examples

b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
g1$points_to_grid(dp, laglength=5)

Method add_covariates()

Adds covariate data to the grid

Maps spatial, temporal, or spatio-temporal covariate data onto the grid

Usage

grid$add_covariates(
  cov_data,
  zcols,
  weight_type = "area",
  popdens = NULL,
  verbose = TRUE,
  t_label = NULL
)

Arguments

Returns

NULL

Examples

b1 <-  sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1$grid_data,
                  zcols="cov",
                  verbose = FALSE)

Method get_dow()

Generate day of week data

Create data frame with day of week indicators

Generates a data frame with indicator variables for each day of the week for use in the add_covariates() function.

Usage

grid$get_dow()

Returns

data.frame with columns t, day, and dayMon to daySun

Examples

b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
g1$points_to_grid(dp, laglength=5)
g1$get_dow()

Method add_time_indicators()

Adds time period indicators to the data

Adds indicator variables for each time period to the data. To include these in a model fitting procedure use, for example, covs = c("time1i, time2i,...)

Usage

grid$add_time_indicators()

Returns

Nothing. Called for effects.

Method lgcp_bayes()

Fit an (approximate) log-Gaussian Cox Process model using Bayesian methods

Usage

grid$lgcp_bayes(
  popdens,
  covs = NULL,
  covs_grid = NULL,
  approx = "nngp",
  m = 10,
  L = 1.5,
  model = "exp",
  known_theta = NULL,
  dir = NULL,
  iter_warmup = 500,
  iter_sampling = 500,
  chains = 3,
  parallel_chains = 3,
  verbose = TRUE,
  vb = FALSE,
  use_cmdstanr = TRUE,
  ...
)

Arguments

Returns

A stanfit or a CmdStanMCMC object

Examples

\dontrun{
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_bayes(popdens="cov")
}

Method lgcp_ml()

Fit an (approximate) log-Gaussian Cox Process model using Maximum Likelihood

Usage

grid$lgcp_ml(
  popdens,
  covs = NULL,
  covs_grid = NULL,
  approx = "nngp",
  m = 10,
  L = 1.5,
  model = "exp",
  known_theta = NULL,
  starting_values = NULL,
  lower_bound = NULL,
  upper_bound = NULL,
  formula_1 = NULL,
  formula_2 = NULL,
  algo = 1,
  tol = 0.01,
  max.iter = 30,
  iter_warmup = 100,
  iter_sampling = 250,
  trace = 1,
  use_cmdstanr = TRUE
)

Arguments

Returns

A mcmlrts model fit object

Examples

\dontrun{
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
res <- g1$lgcp_ml(popdens="cov")
}

Method extract_preds()

Extract predictions

Extract incidence and relative risk predictions

Usage

grid$extract_preds(
  fit,
  type = c("pred", "rr", "irr"),
  irr.lag = NULL,
  t.lag = 0,
  popdens = NULL,
  verbose = TRUE
)

Arguments

Returns

NULL

Examples

\dontrun{
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$extract_preds(res,
                 type=c("pred","rr"),
                 popdens="cov")
}

Method hotspots()

Hotspots

Generate hotspot probabilities

Given a definition of a hotspot in terms of threshold(s) for incidence, relative risk, and/or incidence rate ratio, returns the probabilities each area is a "hotspot". See Details of extract_preds. Columns will be added to grid_data. Note that for incidence threshold, the threshold should be specified as the per individual incidence.

Usage

grid$hotspots(
  fit,
  incidence.threshold = NULL,
  irr.threshold = NULL,
  irr.lag = NULL,
  rr.threshold = NULL,
  popdens,
  col_label = NULL
)

Arguments

Returns

None, called for effects. Columns are added to grid or region data.

Examples

\dontrun{
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$hotspots(res,
            incidence.threshold=1,
            popdens="cov")
}

Method aggregate_output()

Aggregate output

Aggregate lgcp_fit output to another geography

Usage

grid$aggregate_output(
  new_geom,
  zcols,
  weight_type = "area",
  popdens = NULL,
  verbose = TRUE
)

Arguments

Returns

An sf object identical to new_geom with additional columns with the variables specified in zcols

Examples

\dontrun{
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$extract_preds(res,
                 type=c("pred","rr"),
                 popdens="cov")
new1 <- g1$aggregate_output(cov1,
                            zcols="rr")
}

Method scale_conversion_factor()

Returns scale conversion factor

Coordinates are scaled to [-1,1] for lgcp_fit(). This function returns the scaling factor for this conversion.

Usage

grid$scale_conversion_factor()

Returns

numeric

Examples

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
g1$scale_conversion_factor()

Method get_region_data()

Returns summary data of the region/grid intersections

Information on the intersection between the region areas and the computational grid including the number of cells intersecting each region (n_cell), the indexes of the cells intersecting each region in order (cell_id), and the proportion of each region's area covered by each intersecting grid cell (q_weights).

Usage

grid$get_region_data()

Returns

A named list

Method variogram()

Plots the empirical semi-variogram

Usage

grid$variogram(popdens, yvar, nbins = 20)

Arguments

Returns

A ggplot plot is printed and optionally returned

Method reorder()

Re-orders the computational grid

The quality of the nearest neighbour approximation can depend on the ordering of the grid cells. This function reorders the grid cells. If this is a region data model, then the intersections are recomputed.

Usage

grid$reorder(option = "y", verbose = TRUE)

Arguments

Returns

No return, used for effects.

Method data()

A list of prepared data

The class prepares data for use in the in-built estimation functions. The same data could be used for alternative models. This is a utility function to facilitate model fitting for custom models.

Usage

grid$data(m, approx, popdens, covs, covs_grid)

Arguments

Returns

A named list of data items used in model fitting

Method get_random_effects()

Returns the random effects stored in the object (if any) after using MCMCML fitting. For example, if a fitting procedure is stopped, the random effects can still be returned.

Usage

grid$get_random_effects()

Returns

A matrix of random effects samples if a MCMCML model has been initialised, otherwise returns FALSE

Method clone()

The objects of this class are cloneable with this method.

Usage

grid$clone(deep = FALSE)

Arguments

References

(1) Solin A, Särkkä S. Hilbert space methods for reduced-rank Gaussian process regression. Stat Comput. 2020;30:419–46. doi:10.1007/s11222-019-09886-w.

(2) Riutort-Mayol G, Bürkner P-C, Andersen MR, Solin A, Vehtari A. Practical Hilbert space approximate Bayesian Gaussian processes for probabilistic programming. 2020. http://arxiv.org/abs/2004.11408.

See Also

create_points

points_to_grid, add_covariates

points_to_grid, add_covariates

Examples

## ------------------------------------------------
## Method `grid$new`
## ------------------------------------------------

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)

## ------------------------------------------------
## Method `grid$plot`
## ------------------------------------------------

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
g1$plot()

## ------------------------------------------------
## Method `grid$points_to_grid`
## ------------------------------------------------

b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
g1$points_to_grid(dp, laglength=5)

## ------------------------------------------------
## Method `grid$add_covariates`
## ------------------------------------------------

b1 <-  sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1$grid_data,
                  zcols="cov",
                  verbose = FALSE)

## ------------------------------------------------
## Method `grid$get_dow`
## ------------------------------------------------

b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
g1$points_to_grid(dp, laglength=5)
g1$get_dow()

## ------------------------------------------------
## Method `grid$lgcp_bayes`
## ------------------------------------------------

## Not run: 
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_bayes(popdens="cov")

## End(Not run)

## ------------------------------------------------
## Method `grid$lgcp_ml`
## ------------------------------------------------

## Not run: 
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
res <- g1$lgcp_ml(popdens="cov")

## End(Not run)

## ------------------------------------------------
## Method `grid$extract_preds`
## ------------------------------------------------

## Not run: 
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$extract_preds(res,
                 type=c("pred","rr"),
                 popdens="cov")

## End(Not run)

## ------------------------------------------------
## Method `grid$hotspots`
## ------------------------------------------------

## Not run: 
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$hotspots(res,
            incidence.threshold=1,
            popdens="cov")

## End(Not run)

## ------------------------------------------------
## Method `grid$aggregate_output`
## ------------------------------------------------

## Not run: 
b1 <- sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
dp <- data.frame(y=runif(10,0,3),x=runif(10,0,3),date=paste0("2021-01-",11:20))
dp <- create_points(dp,pos_vars = c('y','x'),t_var='date')
cov1 <- grid$new(b1,0.8)
cov1$grid_data$cov <- runif(nrow(cov1$grid_data))
g1$add_covariates(cov1,
                  zcols="cov",
                  verbose = FALSE)
g1$points_to_grid(dp, laglength=5)
g1$priors <- list(
  prior_lscale=c(0,0.5),
  prior_var=c(0,0.5),
  prior_linpred_mean=c(0),
  prior_linpred_sd=c(5)
  )
res <- g1$lgcp_fit(popdens="cov")
g1$extract_preds(res,
                 type=c("pred","rr"),
                 popdens="cov")
new1 <- g1$aggregate_output(cov1,
                            zcols="rr")

## End(Not run)

## ------------------------------------------------
## Method `grid$scale_conversion_factor`
## ------------------------------------------------

b1 = sf::st_sf(sf::st_sfc(sf::st_polygon(list(cbind(c(0,3,3,0,0),c(0,0,3,3,0))))))
g1 <- grid$new(b1,0.5)
g1$scale_conversion_factor()