proposal for hook for loop function that allows plotting

R/bayesopt_ego.R

@@ -30,6 +30,9 @@
 #' For example, if `random_interleave_iter = 2`, random interleaving is performed in the second,
 #' fourth, sixth, ... iteration.
 #' Default is `0`, i.e., no random interleaving is performed at all.
+#' @param hook_fun ([function])\cr
+#' [Function] to be called in each iteration of the loop, before evaluating the next proposed point.
+#' See examples.
 #'
 #' @note
 #' * The `acq_function$surrogate`, even if already populated, will always be overwritten by the `surrogate`.
@@ -82,6 +85,42 @@
 #'
 #' optimizer$optimize(instance)
 #'
+#' # same as above, but plot files with information at each iteration of the loop
+#' library(ggplot2)
+#' library(gridExtra)
+#' 
+#' myPlot = function(xdt, instance, surrogate, acq_function, acq_optimizer) {
+#' data.plot = data.table(x = seq(instance$objective$domain$lower, instance$objective$domain$upper, length.out = 100))
+#' data.plot$y = instance$objective$eval_dt(data.plot)$y
+#' data.plot$acq_ei = acq_optimizer$acq_function$eval_dt(data.table(x = data.plot$x))$acq_ei
+#' data.plot = data.table(data.plot, surrogate$predict(data.plot))
+#' 
+#' p1 = ggplot(data.plot, aes(x = x, y = y)) +
+#' geom_ribbon(aes(ymin = mean - se, ymax = mean + se), fill = "lightgray") +
+#' geom_line() +
+#' geom_line(aes(y = mean), linetype = "dashed") +
+#' geom_point(data = xdt, aes(x = x, y = surrogate$predict(data.table(x = x))$mean), color = "red") +
+#' geom_point(data = instance$archive$data, aes(x = x, y = y), color = "darkgreen") +
+#' theme(axis.title.x = element_blank(),
+#' axis.text.x = element_blank(),
+#' axis.ticks.x = element_blank())
+#' 
+#' p2 = ggplot(data.plot, aes(x = x, y = acq_ei)) +
+#' geom_line()
+#' 
+#' p = grid.arrange(p1, p2, ncol = 1)
+#' ggsave(p, file = paste("mbo-", instance$archive$n_evals, ".pdf", sep = ""))
+#' }
+#' 
+#' optimizer = opt("mbo",
+#' loop_function = bayesopt_ego,
+#' surrogate = surrogate,
+#' acq_function = acqfun,
+#' acq_optimizer = acqopt,
+#' args = list("hook_fun" = myPlot))
+#' 
+#' optimizer$optimize(instance)
+#'
 #' # expected improvement per second example
 #' fun = function(xs) {
 #' list(y = xs$x ^ 2, time = abs(xs$x))
@@ -112,7 +151,8 @@ bayesopt_ego = function(
  acq_function,
  acq_optimizer,
  init_design_size = NULL,
- random_interleave_iter = 0L
+ random_interleave_iter = 0L,
+ hook_fun = function(...) {}
  ) {
 
  # assertions
@@ -154,6 +194,8 @@ bayesopt_ego = function(
  generate_design_random(search_space, n = 1L)$data
  })
 
+ hook_fun(xdt, instance, surrogate, acq_function, acq_optimizer)
+
  instance$eval_batch(xdt)
  if (instance$is_terminated) break
  }

R/bayesopt_emo.R

@@ -30,6 +30,9 @@
 #' For example, if `random_interleave_iter = 2`, random interleaving is performed in the second,
 #' fourth, sixth, ... iteration.
 #' Default is `0`, i.e., no random interleaving is performed at all.
+#' @param hook_fun ([function])\cr
+#' [Function] to be called in each iteration of the loop, before evaluating the next proposed point.
+#' See examples in [mlr_loop_functions_ego].
 #'
 #' @note
 #' * The `acq_function$surrogate`, even if already populated, will always be overwritten by the `surrogate`.
@@ -85,7 +88,8 @@ bayesopt_emo = function(
  acq_function,
  acq_optimizer,
  init_design_size = NULL,
- random_interleave_iter = 0L
+ random_interleave_iter = 0L,
+ hook_fun = function(...) {}
  ) {
 
  # assertions
@@ -127,6 +131,8 @@ bayesopt_emo = function(
  generate_design_random(search_space, n = 1L)$data
  })
 
+ hook_fun(xdt, instance, surrogate, acq_function, acq_optimizer)
+
  instance$eval_batch(xdt)
  if (instance$is_terminated) break
  }

R/bayesopt_mpcl.R

@@ -38,6 +38,9 @@
 #' For example, if `random_interleave_iter = 2`, random interleaving is performed in the second,
 #' fourth, sixth, ... iteration.
 #' Default is `0`, i.e., no random interleaving is performed at all.
+#' @param hook_fun ([function])\cr
+#' [Function] to be called in each iteration of the loop, before evaluating the next proposed point.
+#' See examples in [mlr_loop_functions_ego].
 #'
 #' @note
 #' * The `acq_function$surrogate`, even if already populated, will always be overwritten by the `surrogate`.
@@ -102,7 +105,8 @@ bayesopt_mpcl = function(
  init_design_size = NULL,
  q = 2L,
  liar = mean,
- random_interleave_iter = 0L
+ random_interleave_iter = 0L,
+ hook_fun = function(...) {}
  ) {
 
  # assertions
@@ -175,6 +179,8 @@ bayesopt_mpcl = function(
 
  acq_function$surrogate$archive = instance$archive
 
+ hook_fun(xdt, instance, surrogate, acq_function, acq_optimizer)
+
  instance$eval_batch(xdt)
 
  if (instance$is_terminated) break

R/bayesopt_parego.R

@@ -40,6 +40,9 @@
 #' For example, if `random_interleave_iter = 2`, random interleaving is performed in the second,
 #' fourth, sixth, ... iteration.
 #' Default is `0`, i.e., no random interleaving is performed at all.
+#' @param hook_fun ([function])\cr
+#' [Function] to be called in each iteration of the loop, before evaluating the next proposed point.
+#' See examples in [mlr_loop_functions_ego].
 #'
 #' @note
 #' * The `acq_function$surrogate`, even if already populated, will always be overwritten by the `surrogate`.
@@ -105,7 +108,8 @@ bayesopt_parego = function(
  q = 1L,
  s = 100L,
  rho = 0.05,
- random_interleave_iter = 0L
+ random_interleave_iter = 0L,
+ hook_fun = function(...) {}
  ) {
 
  # assertions
@@ -169,6 +173,8 @@ bayesopt_parego = function(
  })
  }, .fill = TRUE)
 
+ hook_fun(xdt, instance, surrogate, acq_function, acq_optimizer)
+
  instance$eval_batch(xdt)
  if (instance$is_terminated) break
  }

R/bayesopt_smsego.R

@@ -29,6 +29,9 @@
 #' For example, if `random_interleave_iter = 2`, random interleaving is performed in the second,
 #' fourth, sixth, ... iteration.
 #' Default is `0`, i.e., no random interleaving is performed at all.
+#' @param hook_fun ([function])\cr
+#' [Function] to be called in each iteration of the loop, before evaluating the next proposed point.
+#' See examples in [mlr_loop_functions_ego].
 #'
 #' @note
 #' * The `acq_function$surrogate`, even if already populated, will always be overwritten by the `surrogate`.
@@ -90,7 +93,8 @@ bayesopt_smsego = function(
  acq_function,
  acq_optimizer,
  init_design_size = NULL,
- random_interleave_iter = 0L
+ random_interleave_iter = 0L,
+ hook_fun = function(...) {}
  ) {
 
  # assertions
@@ -134,6 +138,8 @@ bayesopt_smsego = function(
  generate_design_random(search_space, n = 1L)$data
  })
 
+ hook_fun(xdt, instance, surrogate, acq_function, acq_optimizer)
+
  instance$eval_batch(xdt)
  if (instance$is_terminated) break
  }