plot.lgspline: Plot Method for Lagrangian Multiplier Smoothing Spline Models

Description

Creates visualizations of fitted spline models, supporting both 1D line plots and 2D surface plots with optional formula annotations and customizable aesthetics. (Wrapper for internal plot method)

Usage

# S3 method for lgspline
plot(
  x,
  show_formulas = FALSE,
  digits = 4,
  legend_pos = "topright",
  custom_response_lab = "y",
  custom_predictor_lab = NULL,
  custom_predictor_lab1 = NULL,
  custom_predictor_lab2 = NULL,
  custom_formula_lab = NULL,
  custom_title = "Fitted Function",
  text_size_formula = NULL,
  legend_args = list(),
  new_predictors = NULL,
  xlim = NULL,
  ylim = NULL,
  color_function = NULL,
  add = FALSE,
  vars = c(),
  ...
)

Value

Returns

1D: Invisibly returns NULL (base R plot is drawn to device).
2D: Plotly object showing interactive surface plot.

Arguments

x

A fitted lgspline model object containing the model fit to be plotted

show_formulas

Logical; whether to display analytical formulas for each partition. Default FALSE

digits

Integer; Number of decimal places for coefficient display in formulas. Default 4

legend_pos

Character; Position of legend for 1D plots ("top", "bottom", "left", "right", "topleft", etc.). Default "topright"

custom_response_lab

Character; Label for response variable axis. Default "y"

custom_predictor_lab

Character; Label for predictor axis in 1D plots. If NULL (default), uses predictor column name

custom_predictor_lab1

Character; Label for first predictor axis (x1) in 2D plots. If NULL (default), uses first predictor column name

custom_predictor_lab2

Character; Label for second predictor axis (x2) in 2D plots. If NULL (default), uses second predictor column name

custom_formula_lab

Character; Label for fitted response on link function scale. If NULL (default), uses "link(E[custom_response_lab])" for non-Gaussian models with non-identity link, otherwise uses custom_response_lab

custom_title

Character; Main plot title. Default "Fitted Function"

text_size_formula

Numeric; Text size for formula display. Passed to cex in legend() for 1D plots and hover font size for 2D plots. If NULL (default), uses 0.8 for 1D and 8 for 2D

legend_args

List; Additional arguments passed to legend() for 1D plots

new_predictors

Matrix; Optional new predictor values for prediction. If NULL (default), uses original fitting data

xlim

Numeric vector; Optional x-axis limits for 1D plots. Default NULL

ylim

Numeric vector; Optional y-axis limits for 1D plots. Default NULL

color_function

Function; Returns colors for plotting by partition, must return K+1 vector of valid colors. Defaults to NULL, in which case grDevices::rainbow(K+1) is used for 1D and grDevices::colorRampPalette(RColorBrewer::brewer.pal(8, "Spectral"))(K+1) used in multiple.

add

Logical; If TRUE, adds to existing plot (1D only). Similar to add in hist. Default FALSE

vars

Numeric or character vector; Optional indices for selecting variables to plot. Can either be numeric (the column indices of "predictors" or "data") or character (the column names, if available from "predictors" or "data")

...

Additional arguments passed to underlying plot functions:

1D: Passed to plot
2D: Passed to plot_ly

Details

Produces different visualizations based on model dimensionality:

1D models: Line plot showing fitted function across partitions, with optional data points and formula annotations
2D models: Interactive 3D surface plot using plotly, with hover text showing predicted values and optional formula display

Partition boundaries are indicated by color changes in both 1D and 2D plots.

When plotting using "select_vars" option, it is recommended to use the "new_predictors" argument to set all terms not involved with plotting to 0 to avoid non-sensical results. But for some cases, it may be useful to set other predictors fixed at certain values. By default, observed values in the data set are used.

The function relies on linear expansions being present - if (for example) a user includes the argument "_1_" or "_2_" in "exclude_these_expansions", then this function will not be able to extract the predictors needed for plotting.

For this case, try constraining the effects of these terms to 0 instead using "constraint_vectors" and "constraint_values" argument, so they are kept in the expansions but their corresponding coefficients will be 0.

Examples

Run this code


## Generate example data
set.seed(1234)
t_data <- runif(1000, -10, 10)
y_data <- 2*sin(t_data) + -0.06*t_data^2 + rnorm(length(t_data))

## Fit model with 10 partitions
model_fit <- lgspline(t_data, y_data, K = 9)

## Basic plot
plot(model_fit)

## Customized plot with formulas
plot(model_fit,
     show_formulas = TRUE,         # Show partition formulas
     custom_response_lab = 'Price',  # Custom axis labels
     custom_predictor_lab = 'Size',
     custom_title = 'Price vs Size', # Custom title
     digits = 2,                    # Round coefficients
     legend_pos = 'bottom',         # Move legend
     text_size_formula = 0.375,     # Adjust formula text size
     pch = 16,                      # Point style
     cex.main = 1.25)               # Title size