Replace default priors with user specified priors

This function is used internally by epinowcast to replace default model priors with users specified ones (restricted to normal priors with specified mean and standard deviations). A common use would be extracting the posterior from a previous epinowcast() run (using summary(nowcast, type = fit)) and using this a prior.

Usage

enw_replace_priors(priors, custom_priors)

Arguments

priors

A data.frame with the following variables: variable, mean, sd describing normal priors. Priors in the appropriate format are returned by enw_reference() as well as by other similar model specification functions.

custom_priors

A data.frame with the following variables: variable, mean, sd describing normal priors. Priors in the appropriate format are returned by enw_reference() as well as by other similar model specification functions. Priors in this data.frame replace the default priors. Note that currently vectorised prior names (i.e those of the form variable[n] will be treated as variable).

Value

A data.table of prior definitions (variable, mean and sd).

See also

Functions used to help convert models into the format required for stan enw_formula_as_data_list(), enw_model(), enw_priors_as_data_list(), enw_sample(), remove_profiling(), write_stan_files_no_profile()

Examples

# Update priors from a data.frame
priors <- data.frame(variable = c("x", "y"), mean = c(1, 2), sd = c(1, 2))
custom_priors <- data.frame(variable = "x[1]", mean = 10, sd = 2)
enw_replace_priors(priors, custom_priors)
#>    variable mean sd
#> 1:        y    2  2
#> 2:        x   10  2

# Update priors from a previous model fit
default_priors <- enw_reference(
 distribution = "lognormal",
 data = enw_example("preprocessed"),
)$priors
print(default_priors)
#>             variable
#> 1:     refp_mean_int
#> 2:       refp_sd_int
#> 3: refp_mean_beta_sd
#> 4:   refp_sd_beta_sd
#>                                                       description
#> 1:         Log mean intercept for parametric reference date delay
#> 2: Log standard deviation for the parametric reference date delay
#> 3:    Standard deviation of scaled pooled parametric mean effects
#> 4:      Standard deviation of scaled pooled parametric sd effects
#>             distribution mean sd
#> 1:                Normal  1.0  1
#> 2: Zero truncated normal  0.5  1
#> 3: Zero truncated normal  0.0  1
#> 4: Zero truncated normal  0.0  1

fit_priors <- summary(
 enw_example("nowcast"), type = "fit",
 variables = c("refp_mean_int", "refp_sd_int", "sqrt_phi")
)
fit_priors
#>            variable      mean    median         sd        mad        q5
#> 1: refp_mean_int[1] 1.4065186 1.3627850 0.31867081 0.28263545 0.9784832
#> 2:   refp_sd_int[1] 1.9488911 1.9301000 0.17462563 0.16212231 1.6974820
#> 3:      sqrt_phi[1] 0.3253364 0.3242025 0.03123022 0.03085661 0.2758690
#>          q20       q80       q95     rhat  ess_bulk ess_tail
#> 1: 1.1410640 1.6335340 2.0573970 1.012034  341.9545 526.7013
#> 2: 1.7996520 2.0693360 2.2809700 1.008008  404.0572 428.0146
#> 3: 0.2998764 0.3513736 0.3794238 1.002608 1325.7827 735.9718

enw_replace_priors(default_priors, fit_priors)
#>             variable
#> 1: refp_mean_beta_sd
#> 2:   refp_sd_beta_sd
#> 3:     refp_mean_int
#> 4:       refp_sd_int
#> 5:          sqrt_phi
#>                                                    description
#> 1: Standard deviation of scaled pooled parametric mean effects
#> 2:   Standard deviation of scaled pooled parametric sd effects
#> 3:                                                        <NA>
#> 4:                                                        <NA>
#> 5:                                                        <NA>
#>             distribution      mean         sd
#> 1: Zero truncated normal 0.0000000 1.00000000
#> 2: Zero truncated normal 0.0000000 1.00000000
#> 3:                  <NA> 1.4065186 0.31867081
#> 4:                  <NA> 1.9488911 0.17462563
#> 5:                  <NA> 0.3253364 0.03123022