使用factoextra绘制PCA分组椭圆时3个点报错，求解析所用椭圆函数

Great question! Let's break this down clearly:

What ellipse function does factoextra's fviz series use? #

Factoextra's fviz_pca_ind(), fviz_pca_biplot(), and related visualization functions rely on ggpubr::stat_ellipse() under the hood when you set ellipse = TRUE.

This function is a wrapper around ggplot2::stat_ellipse(), but with one key difference that's causing your error:

The root cause of your error #

ggplot2::stat_ellipse() only throws the "Too few points" error when a group has fewer than 3 points (since it needs at least 3 to calculate the covariance matrix for an ellipse).

However, ggpubr::stat_ellipse() (which factoextra uses by default) has a default min_n parameter set to 4. That's why your 3-point groups trigger the error in factoextra, but work fine in raw ggplot2.

How to fix it and draw ellipses for 3-point groups in factoextra #

You have two straightforward options:

Option 1: Add ggplot2's native stat_ellipse layer manually #

Generate your base PCA plot with factoextra, then layer on ggplot2's ellipse function directly:

# Example code
library(FactoMineR)
library(factoextra)
library(ggplot2)
library(dplyr)

# Create a small dataset with 3 points per group
small_iris <- iris %>% 
  group_by(Species) %>% 
  slice(1:3) %>% 
  ungroup()

# Run PCA
pca_small <- PCA(small_iris[,1:4], scale.unit = TRUE, graph = FALSE)

# Plot with factoextra + ggplot2's ellipses
fviz_pca_ind(pca_small, habillage = small_iris$Species) +
  stat_ellipse(aes(color = Species), type = "norm")

Option 2: Override ggpubr's min_n threshold #

You can modify the min_n parameter in ggpubr::stat_ellipse() to allow 3-point groups. You can either pass this directly via factoextra's ellipse-related parameters, or add the layer explicitly:

# Using factoextra's ellipse parameters (if supported)
fviz_pca_ind(pca_small, habillage = small_iris$Species, 
             ellipse = TRUE, ellipse.min_n = 3)

# Or explicitly add the ggpubr layer with adjusted min_n
fviz_pca_ind(pca_small, habillage = small_iris$Species) +
  ggpubr::stat_ellipse(aes(color = Species), min_n = 3)

Either approach will let you draw ellipses for your 3-point groups, matching the behavior you see in raw ggplot2.

内容的提问来源于stack exchange，提问作者J.Con