Visualize Model (Unified 1D/2D)
VisualizerModel.RdThis class provides a unified interface for visualizing machine learning models on both 1D and 2D tasks. It automatically detects the dimensionality and creates appropriate visualizations using ggplot2.
Super class
vistool::Visualizer -> VisualizerModel
Public fields
task(`mlr3::Task`)
Task used to train the model.learner(`mlr3::Learner`)
Learner used to train the model.
Methods
Method new()
Creates a new instance of this [R6][R6::R6Class] class.
Usage
VisualizerModel$new(
task,
learner,
x1_limits = NULL,
x2_limits = NULL,
padding = 0,
n_points = 100L,
hypothesis = NULL,
domain = NULL,
retrain = TRUE
)Arguments
task([mlr3::Task])
The task to train the model on.learner(`mlr3::Learner`)
The learner to train the model with.x1_limits(`numeric(2)`)
Limits for the first feature axis. For 1D tasks, this controls the x-axis range. If NULL, will be determined from task data.x2_limits(`numeric(2)`)
Limits for the second feature axis (2D tasks only). Ignored for 1D tasks. If NULL, will be determined from task data.padding(`numeric(1)`)
A margin that is added to x1limits and x2limits. The x1 margin is calculated by `max(x1lmits) - min(x1limits) * padding`.padding(`numeric(1)`)
A margin that is added to x1limits and x2limits. The x1 margin is calculated by `max(x1lmits) - min(x1limits) * padding`.n_points(`integer(1)`)
The number of generated point per dimension. Note that a grid of `npoints^2` values is generated and evaluated by `objective$eval(x)` to plot the surface.n_points(`integer(1)`)
The number of generated point per dimension. Note that a grid of `npoints^2` values is generated and evaluated by `objective$eval(x)` to plot the surface.hypothesis(`Hypothesis`|`NULL`) Optional hypothesis object that provides predictions instead of a learner. Supply either `learner` or `hypothesis`, but not both.
domain(`list`|`NULL`) Named list giving axis limits per predictor when no Task data is available, e.g., `list(x = c(-1, 1))` for 1D or `list(x = c(-1,1), y = c(-1,1))` for 2D.
retrain(`logical(1)`) Whether to (re)train the supplied learner on the task (default `TRUE`). Set to `FALSE` to reuse an already trained learner. If set to `FALSE` but the learner has not yet been trained (`learner$model` is `NULL`), a warning is emitted and training is performed to ensure predictions are available.
Method add_points()
Add points with optional residual loss geometry (1D regression only for now). Extends base add_points() with a `loss` argument to visualize residuals.
Usage
VisualizerModel$add_points(
points,
color = "auto",
size = NULL,
shape = 19,
alpha = NULL,
annotations = NULL,
annotation_size = NULL,
ordered = FALSE,
arrow_color = NULL,
arrow_size = 0.3,
loss = NULL,
loss_params = list(),
loss_fill = "auto",
loss_alpha = NULL,
loss_color = NA,
loss_linetype = "solid"
)Arguments
points(`data.frame`|`matrix`|`list`) Points to add. For 1D these should contain columns `x` and `y` (observed y required for residuals).
color(`character(1)`) Color of the points or "auto". Default "auto".
size(`numeric(1)`|`NULL`) Point size. If NULL uses theme default.
shape(`integer(1)`|`character(1)`) Point shape. Default 19.
alpha(`numeric(1)`|`NULL`) Alpha transparency for points.
annotations(`character()`|`NULL`) Optional text annotations for points.
annotation_size(`numeric(1)`|`NULL`) Size of annotation text.
ordered(`logical(1)`) Whether points should be connected in order (arrows). Default FALSE.
arrow_color(`character(1)`|`NULL`) Arrow color when ordered = TRUE.
arrow_size(`numeric(1)`) Arrow size when ordered = TRUE. Default 0.3.
loss(`character(1)`|`NULL`) One of `"l2_se"` (aliases: `"l2"`, `"se"`), `"l1_ae"` (aliases: `"l1"`, `"abs"`, `"mae"`), or `NULL` (no geometry).
loss_params(`list()`) Reserved for future loss-specific parameters (e.g. Huber delta).
loss_fill(`character(1)`) Fill color for L2 squares. "auto" draws from palette.
loss_alpha(`numeric(1)`|`NULL`) Fill alpha for L2 squares (defaults to theme$alpha * 0.4).
loss_color(`character(1)`|`NA`) Color for residual segment / square border. If `NA`, derived from fill.
loss_linetype(`character(1)`) Line type for residual segment. Default "solid".
Method add_training_data()
Adds the training data to the plot.
Usage
VisualizerModel$add_training_data(
color = "auto",
size = NULL,
shape = 19,
alpha = NULL,
show_labels = FALSE,
label_size = NULL
)Arguments
color(`character(1)` or named `character`)
Color of the points. For classification tasks: - `character(1)`: A single color for all points (e.g., `"blue"`) or `"auto"` for automatic color assignment. - `named character`: A vector mapping class labels to colors (e.g., `c(pos = "red", neg = "blue")`). For regression tasks, only single colors are supported. Default is `"auto"`.size(`numeric(1)`)
Size of the points. If NULL, uses theme$point_size. Default is NULL.shape(`numeric(1)` or named `numeric`)
Shape of the points. For classification tasks, can be a named vector mapping class labels to shapes. Default is 19 (filled circle).alpha(`numeric(1)`)
Alpha transparency of the points. If NULL, uses theme$alpha. Default is NULL.show_labels(`logical(1)`)
Whether to show data point labels. Default is FALSE.label_size(`numeric(1)`)
Size of data point labels. If NULL, defaults to smaller text.
Method add_boundary()
Adds boundary lines/contours to the plot.
Usage
VisualizerModel$add_boundary(
values = NULL,
color = "black",
linetype = "dashed",
linewidth = NULL,
alpha = NULL
)Arguments
values(`numeric()`)
Values at which to draw boundaries. For 1D: horizontal lines (y-values). For 2D: contour lines (z-values). If NULL, uses sensible defaults based on prediction type.color(`character(1)`)
Color of the boundary lines. Default is "black".linetype(`character(1)`)
Line type for boundaries. For 1D: ggplot2 linetypes. For 2D: contour line types. Default is "dashed".linewidth(`numeric(1)`)
Width of boundary lines. If NULL, uses theme$line_width.alpha(`numeric(1)`)
Alpha transparency of boundary lines. If NULL, uses theme$alpha.
Method plot()
Create and return the ggplot2 plot with model-specific layers.