2024-12-08 update : changing center of prior.

inst/extdata/example_params.toml

@@ -3,8 +3,8 @@ uot = 50
 
 [infection_process]
 # spatial params
-log_phi_mu_prior = -1.386294 # log(0.25)
-log_phi_sd_prior = 0.75
+log_phi_mu_prior = 0 # log(1)
+log_phi_sd_prior = 0.4
 l = 1
 log_sigma_generalized_mu_prior = -5.298317 # log(0.05^4)
 log_sigma_generalized_sd_prior = 0.5

scratch/phi_prior_investig.R

@@ -0,0 +1,329 @@
+## Investigating if prior is identifiable / effect of changing prior
+# Library
+library(wwinference)
+library(tidyverse)
+library(patchwork)
+
+
+## Presets - Data Generation
+set.seed(2024)
+n_sites <- 4
+fake_locs <- data.frame(
+  x = runif(n_sites, 0, 100),
+  y = runif(n_sites, 0, 100),
+  ID = paste0("Site ", 1:n_sites)
+)
+r_in_weeks <- c(
+  rep(1.1, 5), rep(0.9, 5),
+  0.9 + 0.0005 * (1:16)^2
+)
+site <- 1:n_sites
+lab <- rep(1, n_sites)
+ww_pop_sites <- c(5e5, 7.5e5, 3.8e5, 4.5e5)
+pop_size <- 2 * sum(ww_pop_sites)
+global_rt_sd <- 0
+exp_corr_param <- list(
+  dist_matrix = as.matrix(
+    dist(
+      data.frame(
+        x = fake_locs$x,
+        y = fake_locs$y
+      ),
+      diag = TRUE,
+      upper = TRUE
+    )
+  ),
+  phi = 1.5,
+  l = 1
+)
+sigma_sqrd_generalized <- 0.01^n_sites
+
+# Data Generation - Exponential
+simulated_data <- withr::with_seed(1, {
+  wwinference::generate_simulated_data(
+    r_in_weeks = r_in_weeks, # nolint
+    n_sites = n_sites,
+    site = site,
+    lab = lab,
+    ww_pop_sites = ww_pop_sites,
+    pop_size = pop_size,
+    global_rt_sd = global_rt_sd,
+    use_spatial_corr = TRUE,
+    corr_function = exponential_decay_corr_func_r,
+    corr_fun_params = exp_corr_param,
+    sigma_sqrd_generalized = sigma_sqrd_generalized,
+    aux_site_bool = TRUE
+  )
+})
+hosp_data_preprocessed <- wwinference::preprocess_count_data(
+  simulated_data$hosp_data,
+  count_col_name = "daily_hosp_admits",
+  pop_size_col_name = "state_pop"
+)
+ww_data_preprocessed <- wwinference::preprocess_ww_data(
+  simulated_data$ww_data
+)
+ww_data_to_fit <- wwinference::indicate_ww_exclusions(
+  ww_data_preprocessed,
+  outlier_col_name = "flag_as_ww_outlier",
+  remove_outliers = TRUE
+)
+
+
+## Presets - Fit
+params_new <- get_params(
+  system.file("extdata", "example_params.toml",
+    package = "wwinference"
+  )
+)
+cat("New Log Phi Mean :", params_new$log_phi_mu_prior, "\n")
+params_old <- params_new
+params_old$log_phi_mu_prior <- log(0.25)
+params_old$log_phi_sd_prior <- 0.75
+cat("Old Log Phi Mean :", params_old$log_phi_mu_prior, "\n")
+forecast_date <- "2023-12-06"
+calibration_time <- 90
+forecast_horizon <- 28
+
+generation_interval <- wwinference::default_covid_gi
+inf_to_hosp <- wwinference::default_covid_inf_to_hosp
+
+# Assign infection feedback equal to the generation interval
+infection_feedback_pmf <- generation_interval
+
+model <- wwinference::compile_model()
+
+iter_warmup <- 250
+iter_sampling <- 500
+fit_options <- get_mcmc_options(
+  iter_warmup = iter_warmup,
+  iter_sampling = iter_sampling,
+  seed = 123
+)
+
+
+## Fitting
+fit_old <- wwinference::wwinference(
+  ww_data = ww_data_to_fit,
+  count_data = hosp_data_preprocessed,
+  forecast_date = forecast_date,
+  calibration_time = calibration_time,
+  forecast_horizon = forecast_horizon,
+  model_spec = get_model_spec(
+    generation_interval = generation_interval,
+    inf_to_count_delay = inf_to_hosp,
+    infection_feedback_pmf = infection_feedback_pmf,
+    params = params_old
+  ),
+  fit_opts = fit_options,
+  compiled_model = model,
+  dist_matrix = as.matrix(exp_corr_param$dist_matrix),
+  corr_structure_switch = 1
+)
+phi_draws_old <- fit_old$fit$result$draws(
+  variables = "phi",
+  format = "draws_array"
+)
+phi_draws_old <- phi_draws_old %>%
+  as.data.frame() %>%
+  pivot_longer(
+    everything(),
+    names_to = "chain",
+    values_to = "phi"
+  ) %>%
+  mutate(
+    chain = case_when(
+      chain == "1.phi" ~ 1,
+      chain == "2.phi" ~ 2,
+      chain == "3.phi" ~ 3,
+      chain == "4.phi" ~ 4
+    ),
+    chain = factor(chain)
+  )
+
+fit_new <- wwinference::wwinference(
+  ww_data = ww_data_to_fit,
+  count_data = hosp_data_preprocessed,
+  forecast_date = forecast_date,
+  calibration_time = calibration_time,
+  forecast_horizon = forecast_horizon,
+  model_spec = get_model_spec(
+    generation_interval = generation_interval,
+    inf_to_count_delay = inf_to_hosp,
+    infection_feedback_pmf = infection_feedback_pmf,
+    params = params_new
+  ),
+  fit_opts = fit_options,
+  compiled_model = model,
+  dist_matrix = as.matrix(exp_corr_param$dist_matrix),
+  corr_structure_switch = 1
+)
+phi_draws_new <- fit_new$fit$result$draws(
+  variables = "phi",
+  format = "draws_array"
+)
+phi_draws_new <- phi_draws_new %>%
+  as.data.frame() %>%
+  pivot_longer(
+    everything(),
+    names_to = "chain",
+    values_to = "phi"
+  ) %>%
+  mutate(
+    chain = case_when(
+      chain == "1.phi" ~ 1,
+      chain == "2.phi" ~ 2,
+      chain == "3.phi" ~ 3,
+      chain == "4.phi" ~ 4
+    ),
+    chain = factor(chain)
+  )
+
+
+
+## Visualizing Fits
+# Trace Plots
+phi_trace_new_plot <- ggplot(phi_draws_new) +
+  geom_line(
+    aes(
+      y = phi,
+      x = seq_along(phi),
+      colour = chain
+    )
+  ) +
+  labs(
+    title = "Trace Plot for Phi Draws w/ New Prior",
+    x = "iter",
+    y = "phi",
+    colour = "Chain"
+  ) +
+  scale_color_manual(
+    values = c("gold", "darkred", "darkgreen", "darkblue")
+  ) +
+  theme_bw()
+phi_trace_old_plot <- ggplot(phi_draws_old) +
+  geom_line(
+    aes(
+      y = phi,
+      x = seq_along(phi),
+      colour = chain
+    )
+  ) +
+  labs(
+    title = "Trace Plot for Phi Draws w/ Old Prior",
+    x = "iter",
+    y = "phi",
+    colour = "Chain"
+  ) +
+  scale_color_manual(
+    values = c("gold", "darkred", "darkgreen", "darkblue")
+  ) +
+  theme_bw()
+
+phi_trace_old_plot / phi_trace_new_plot +
+  plot_layout(
+    guides = "collect"
+  )
+
+# Prior and Posterior
+prior_old <- exp(rnorm(
+  n = 1000,
+  mean = params_old$log_phi_mu_prior,
+  sd = params_old$log_phi_sd_prior
+))
+prior_new <- exp(rnorm(
+  n = 1000,
+  mean = params_new$log_phi_mu_prior,
+  sd = params_new$log_phi_sd_prior
+))
+
+prior_post_old_plot <- ggplot() +
+  geom_vline(
+    xintercept = exp_corr_param$phi,
+    linetype = "dashed"
+  ) +
+  geom_density(
+    aes(
+      x = prior_old,
+      fill = "Prior",
+      colour = "Prior"
+    ),
+    alpha = .25
+  ) +
+  geom_density(
+    data = phi_draws_old %>%
+      filter(
+        chain == 1
+      ),
+    aes(
+      x = phi,
+      fill = "Posterior",
+      colour = "Posterior"
+    ),
+    alpha = .25
+  ) +
+  scale_color_manual(
+    breaks = c("Prior", "Posterior"),
+    values = c("dodgerblue", "purple")
+  ) +
+  scale_fill_manual(
+    breaks = c("Prior", "Posterior"),
+    values = c("dodgerblue", "purple")
+  ) +
+  labs(
+    title = "Prior & Posterior w/ Old Prior",
+    subtitle = "Vertical Dashed Line Actual",
+    fill = "Dist. Type",
+    colour = "Dist. Type",
+    y = "Density",
+    x = "Value"
+  ) +
+  theme_bw()
+
+prior_post_new_plot <- ggplot() +
+  geom_vline(
+    xintercept = exp_corr_param$phi,
+    linetype = "dashed"
+  ) +
+  geom_density(
+    aes(
+      x = prior_new,
+      fill = "Prior",
+      colour = "Prior"
+    ),
+    alpha = .25
+  ) +
+  geom_density(
+    data = phi_draws_new %>%
+      filter(
+        chain == 1
+      ),
+    aes(
+      x = phi,
+      fill = "Posterior",
+      colour = "Posterior"
+    ),
+    alpha = .25
+  ) +
+  scale_color_manual(
+    breaks = c("Prior", "Posterior"),
+    values = c("dodgerblue", "purple")
+  ) +
+  scale_fill_manual(
+    breaks = c("Prior", "Posterior"),
+    values = c("dodgerblue", "purple")
+  ) +
+  labs(
+    title = "Prior & Posterior w/ New Prior",
+    subtitle = "Vertical Dashed Line Actual",
+    fill = "Dist. Type",
+    colour = "Dist. Type",
+    y = "Density",
+    x = "Value"
+  ) +
+  theme_bw()
+
+prior_post_old_plot / prior_post_new_plot +
+  plot_layout(
+    guides = "collect"
+  )

vignettes/spatial_wwinference.Rmd

@@ -243,7 +243,7 @@ exp_corr_param <- list(
       upper = TRUE
     )
   ),
-  phi = 0.4,
+  phi = 1.2,
   l = 1
 )
 sigma_sqrd_generalized <- 0.01^n_sites