Two dimensional visualization of kultarr

library(kultarr)
library(randomForest)
#> randomForest 4.7-1.2
#> Type rfNews() to see new features/changes/bug fixes.
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following object is masked from 'package:randomForest':
#> 
#>     combine
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)
library(kumquat)
library(ggplot2)
#> 
#> Attaching package: 'ggplot2'
#> The following object is masked from 'package:randomForest':
#> 
#>     margin

set.seed(145)
data("d_multitwo")
train_data <- d_multitwo
rf_model <- randomForest(class ~ ., data = train_data)

model_func <- carrier::crate(function(data) {
  return(randomForest:::predict.randomForest(!!rf_model, data))
})

find_closest <- function(pt, data) {
  dst <- data |>
    mutate(dst = sqrt((x - pt$x)^2 + (y - pt$y)^2))
  return(which.min(dst$dst))
}
obs <- find_closest(tibble(x = 0, y = 0), train_data)

final_bounds <- make_anchors(
  dataset = train_data,
  cols = c("x", "y"),
  instance = train_data[obs, ],
  model_func = model_func,
  class_col = "class",
  verbose = FALSE
)
#> INFO [2026-08-28 17:10:34] setting up bin edges
#> INFO [2026-08-28 17:10:34] setting lower bounds
#> INFO [2026-08-28 17:10:34] Have 2 lower bounds with -0.054531305283308
#> INFO [2026-08-28 17:10:34] Have 2 lower bounds with -0.104531305283308
#> INFO [2026-08-28 17:10:34] setting upper bounds
#> INFO [2026-08-28 17:10:34] Have 2 upper bounds with 0.045468694716692
#> INFO [2026-08-28 17:10:34] Have 2 upper bounds with 0.095468694716692
#> INFO [2026-08-28 17:10:34] setting lower bounds
#> INFO [2026-08-28 17:10:34] Have 2 lower bounds with -0.046591470669955
#> INFO [2026-08-28 17:10:34] Have 2 lower bounds with -0.096591470669955
#> INFO [2026-08-28 17:10:34] setting upper bounds
#> INFO [2026-08-28 17:10:34] Have 2 upper bounds with 0.053408529330045
#> INFO [2026-08-28 17:10:34] Have 2 upper bounds with 0.103408529330045
#> INFO [2026-08-28 17:10:35] received precisions 1 , 0
#> INFO [2026-08-28 17:10:35] found new max_reward -1 and node 1:1:1:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.984126984126984 , 0.0158730158730159
#> INFO [2026-08-28 17:10:35] found new max_reward 1.4509776489234 and node 2:1:1:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 1 , 0
#> INFO [2026-08-28 17:10:35] found new max_reward 1.47832713531113 and node 1:2:1:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 1 , 0
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.968253968253968 , 0.0317460317460317
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.988304093567251 , 0.0116959064327485
#> INFO [2026-08-28 17:10:35] found new max_reward 1.50199805708667 and node 2:2:1:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.989795918367347 , 0.0102040816326531
#> INFO [2026-08-28 17:10:35] found new max_reward 1.53494710654256 and node 2:1:2:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.86734693877551 , 0.13265306122449
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 1 , 0
#> INFO [2026-08-28 17:10:35] found new max_reward 1.56529729695833 and node 1:2:2:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.979591836734694 , 0.0204081632653061
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.976608187134503 , 0.0233918128654971
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.992481203007519 , 0.0075187969924812
#> INFO [2026-08-28 17:10:35] found new max_reward 1.58136313645174 and node 2:2:2:1
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.902255639097744 , 0.0977443609022556
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.902255639097744 , 0.0977443609022556
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.984962406015038 , 0.0150375939849624
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2
#> INFO [2026-08-28 17:10:35] received precisions 0.92797783933518 , 0.07202216066482
#> INFO [2026-08-28 17:10:35] max_values
#> INFO [2026-08-28 17:10:35] 2:2:2:2

The returned object has several parts in it. For visualisation we would primarily need the final_anchor and the perturbations used in it.

kultarr has a convenience function called rule_rect_layers, which when given a tibble containing the bounds of either the anchors or the perturbations, will create a geom_rect object that can be used in ggplot visualisations.

This one is on the lower end

model_fit <- expand_grid(x = seq(-1, 1, length.out = 50), y = seq(-1, 1, length.out = 50))
model_fit$pred <- model_func(model_fit)

ggplot() +
  geom_point(data = model_fit, mapping = aes(x = x, y = y, colour = pred), alpha = 0.1) +
  rule_rect_layers(final_bounds$perturb_bounds, fill = "gray", alpha = 0.5) +
  rule_rect_layers(final_bounds$final_anchor, fill = "black", alpha = 0.8) +
  geom_point(data = train_data[obs, ], mapping = aes(x = x, y = y, colour = class), size = 2, shape = 19, fill = "black") +
  theme_minimal() +
  coord_equal()

obs <- find_closest(tibble(x = 0, y = 0.1), train_data)

final_bounds <- make_anchors(
  dataset = train_data,
  cols = c("x", "y"),
  instance = train_data[obs, ],
  model_func = model_func,
  class_col = "class",
  verbose = FALSE
)
#> INFO [2026-08-28 17:10:36] setting up bin edges
#> INFO [2026-08-28 17:10:36] setting lower bounds
#> INFO [2026-08-28 17:10:36] Have 2 lower bounds with -0.0616270487196744
#> INFO [2026-08-28 17:10:36] Have 2 lower bounds with -0.111627048719674
#> INFO [2026-08-28 17:10:36] setting upper bounds
#> INFO [2026-08-28 17:10:36] Have 2 upper bounds with 0.0383729512803256
#> INFO [2026-08-28 17:10:36] Have 2 upper bounds with 0.0883729512803257
#> INFO [2026-08-28 17:10:36] setting lower bounds
#> INFO [2026-08-28 17:10:36] Have 2 lower bounds with 0.061310753505677
#> INFO [2026-08-28 17:10:36] Have 2 lower bounds with 0.011310753505677
#> INFO [2026-08-28 17:10:36] setting upper bounds
#> INFO [2026-08-28 17:10:36] Have 2 upper bounds with 0.161310753505677
#> INFO [2026-08-28 17:10:36] Have 2 upper bounds with 0.211310753505677
#> INFO [2026-08-28 17:10:36] received precisions 0.395061728395062 , 0.604938271604938
#> INFO [2026-08-28 17:10:36] found new max_reward -1 and node 1:1:1:1
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.253968253968254 , 0.746031746031746
#> INFO [2026-08-28 17:10:36] found new max_reward 0.971125141612951 and node 2:1:1:1
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.579365079365079 , 0.420634920634921
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.611111111111111 , 0.388888888888889
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.253968253968254 , 0.746031746031746
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.426900584795322 , 0.573099415204678
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.459183673469388 , 0.540816326530612
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.163265306122449 , 0.836734693877551
#> INFO [2026-08-28 17:10:36] found new max_reward 1.16088389827451 and node 2:1:1:2
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.729591836734694 , 0.270408163265306
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.387755102040816 , 0.612244897959184
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.450292397660819 , 0.549707602339181
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.586466165413534 , 0.413533834586466
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.285714285714286 , 0.714285714285714
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.342105263157895 , 0.657894736842105
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.548872180451128 , 0.451127819548872
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2
#> INFO [2026-08-28 17:10:36] received precisions 0.440443213296399 , 0.559556786703601
#> INFO [2026-08-28 17:10:36] max_values
#> INFO [2026-08-28 17:10:36] 2:2:2:2

This one is right on the boundary

ggplot() +
  geom_point(data = model_fit, mapping = aes(x = x, y = y, colour = pred), alpha = 0.1) +
  rule_rect_layers(final_bounds$perturb_bounds, fill = "gray", alpha = 0.5) +
  rule_rect_layers(final_bounds$final_anchor, fill = "black", alpha = 0.8) +
  geom_point(data = train_data[obs, ], mapping = aes(x = x, y = y, colour = class), size = 2, shape = 19, fill = "black") +
  theme_minimal() +
  coord_equal()

obs <- find_closest(tibble(x = 0, y = 0.2), train_data)

final_bounds <- make_anchors(
  dataset = train_data,
  cols = c("x", "y"),
  instance = train_data[obs, ],
  model_func = model_func,
  class_col = "class",
  verbose = FALSE
)
#> INFO [2026-08-28 17:10:37] setting up bin edges
#> INFO [2026-08-28 17:10:37] setting lower bounds
#> INFO [2026-08-28 17:10:37] Have 2 lower bounds with -0.0558166175149381
#> INFO [2026-08-28 17:10:37] Have 2 lower bounds with -0.105816617514938
#> INFO [2026-08-28 17:10:37] setting upper bounds
#> INFO [2026-08-28 17:10:37] Have 2 upper bounds with 0.0441833824850619
#> INFO [2026-08-28 17:10:37] Have 2 upper bounds with 0.0941833824850619
#> INFO [2026-08-28 17:10:37] setting lower bounds
#> INFO [2026-08-28 17:10:37] Have 2 lower bounds with 0.137909724656492
#> INFO [2026-08-28 17:10:37] Have 2 lower bounds with 0.0879097246564925
#> INFO [2026-08-28 17:10:37] setting upper bounds
#> INFO [2026-08-28 17:10:37] Have 2 upper bounds with 0.237909724656492
#> INFO [2026-08-28 17:10:37] Have 2 upper bounds with 0.287909724656492
#> INFO [2026-08-28 17:10:37] received precisions 0 , 1
#> INFO [2026-08-28 17:10:37] found new max_reward -1 and node 1:1:1:1
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0 , 1
#> INFO [2026-08-28 17:10:37] found new max_reward 1.4509776489234 and node 2:1:1:1
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0634920634920635 , 0.936507936507937
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0873015873015873 , 0.912698412698413
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0 , 1
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0467836257309941 , 0.953216374269006
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0561224489795918 , 0.943877551020408
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0 , 1
#> INFO [2026-08-28 17:10:37] found new max_reward 1.50199805708667 and node 2:1:1:2
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.219387755102041 , 0.780612244897959
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0408163265306122 , 0.959183673469388
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.064327485380117 , 0.935672514619883
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.161654135338346 , 0.838345864661654
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0300751879699248 , 0.969924812030075
#> INFO [2026-08-28 17:10:37] found new max_reward 1.58136313645174 and node 2:2:1:2
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.0413533834586466 , 0.958646616541353
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.161654135338346 , 0.838345864661654
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2
#> INFO [2026-08-28 17:10:37] received precisions 0.119113573407202 , 0.880886426592798
#> INFO [2026-08-28 17:10:37] max_values
#> INFO [2026-08-28 17:10:37] 2:2:2:2

ggplot() +
  geom_point(data = model_fit, mapping = aes(x = x, y = y, colour = pred), alpha = 0.1) +
  rule_rect_layers(final_bounds$perturb_bounds, fill = "gray", alpha = 0.5) +
  rule_rect_layers(final_bounds$final_anchor, fill = "black", alpha = 0.7) +
  geom_point(data = train_data[obs, ], mapping = aes(x = x, y = y, colour = class), size = 2, shape = 19, fill = "black") +
  theme_minimal() +
  coord_equal()

Try six more different places, some from the axis parallel places etc.

ggplot() +
  geom_point(data = final_bounds$perturbs, mapping = aes(x = x, y = y, color = preds)) +
  rule_rect_layers(final_bounds$final_anchor) +
  geom_point(data = train_data[21, ], mapping = aes(x = x, y = y), size = 2, shape = 19, fill = "black") +
  theme_minimal()