update docs, add_vars

R/run.R

@@ -144,8 +144,20 @@ run_gp = function(dat, ..., max_grid_size = 2000,
 #' @param lambda tradeoff between weighting posterior predictive variance and expected
 #'   improvement at grid values.
 #' @details The acquisition function is \code{lambda*f_star_var + (1-lambda)*exp_imp}.
-#' Higher values of lambda up-weight posterior predictive variance, leading to more
-#' exploration over exploitation. Lower lambda values up-weight expected improvement over \code{max(dat$rating) - offset}. 
+#'   Higher values of lambda up-weight posterior predictive variance, leading to more
+#'   exploration over exploitation. Lower lambda values up-weight expected improvement
+#'   over \code{max(dat$rating) - offset}.
+#'
+#'   For the sake of simplicity, the range of each parameter in the grid is linearly
+#'   scaled to an N-dimensional hypercube that spans -3 to 3 on each edge. So the model is
+#'   insensitive to the range of grid values. It won't make a difference if your grinder
+#'   shows different numbers or something.
+#'
+#'   It's normal for the sampling to slow down dramatically after the warmup phase. This
+#'   is because while it's fast to fit a GP to a tiny number of observations, it's much
+#'   more expensive to evaluate the GP over the parameter grid. This happens in the
+#'   generated quantities block of the model, which only gets evaluated in the sampling
+#'   phase, which is why it's slow. Whatever man. My CPU has 16 physical cores.
 #' @returns a list with elements:
 #' \itemize{
 #'   \item{draws_df}{a draws data frame of model parameters and grid point predictive draws f_star}
@@ -184,21 +196,6 @@ suggest_next = function(dat, ..., max_grid_size = 2000,
 
   if (all(exp_imp < .Machine$double.eps^0.5)) cli::cli_warn("All expected improvement values near zero. You may need to run the chains for longer or raise {.var offset}.")
 
-  # pred_g = x_grid[max_pred_dens,,drop=FALSE][,"gc"]
-
-  # exp_imp_post = data.table(variable = colnames(exp_imp),
-  #                       mean = exp_imp |> colMeans(),
-  #                       i = 1:ncol(exp_imp))
-  # 
-  # post_range = exp_imp_post$mean |> range()
-
-  # qDT(x_grid) |> mtt(i = 1:nrow(x_grid)) |>
-  #   sbt(dplyr::near(gc, pred_g)) |>
-  #   join(exp_imp_post, on = "i", validate = "1:1") |>
-  #   ggplot(aes(tc, bc)) +
-  #   geom_tile(aes(fill = mean)) +
-  #   scale_fill_viridis_c(limits = post_range)
-
   # posterior uncertainty ----
   f_mean_sd = f_mean_mat |> fsd()
 
@@ -209,7 +206,7 @@ suggest_next = function(dat, ..., max_grid_size = 2000,
   acq_df = data.table(post_sd = f_mean_sd,
                       exp_imp = exp_imp,
                       acq = combined_acq) |> 
-    cbind(x_grid) 
+    add_vars(x_grid) 
 
   suggest = acq_df |> sbt(whichv(acq, fmax(acq)))