如何优化加州住房数据集PCA双标图（Biplot）的可读性？

Got it, let's fix that messy biplot you're getting from biplot(housingpr, scale = 0). Here are actionable, R-specific tweaks to make your plot clear and informative:

• Add meaningful axis labels with variance explained
  Default axis labels just say "PC1" and "PC2"—that doesn't tell you how much variance each component captures. First calculate the variance percentages:

  variance <- summary(housingpr)$importance[2, 1:2]
  pc1_label <- paste0("Principal Component 1 (", round(variance[1]*100, 1), "% Variance)")
  pc2_label <- paste0("Principal Component 2 (", round(variance[2]*100, 1), "% Variance)")
  

  Then pass these to your biplot:

  biplot(housingpr, scale = 0, xlab = pc1_label, ylab = pc2_label, main = "PCA Biplot - California Housing")
  

• Fix overlapping variable labels with custom plotting
  The base R biplot() function often crams variable labels together. Instead, build the plot manually to control arrow and label positions:

  # Extract PCA results
  pca_results <- prcomp(your_housing_data, scale. = TRUE) # Replace with your raw data frame
  pc_scores <- pca_results$x[, 1:2]
  loadings <- pca_results$rotation[, 1:2]
  
  # Plot observations with low opacity to reduce clutter
  plot(pc_scores, pch = 16, col = rgb(0,0,0, alpha = 0.3), xlab = pc1_label, ylab = pc2_label)
  
  # Add scaled arrows for variables
  arrow_scale <- 2 # Adjust this to make arrows longer/shorter
  arrows(0, 0, loadings[,1]*arrow_scale, loadings[,2]*arrow_scale, length = 0.1, col = "darkred")
  
  # Place labels slightly outside arrows to avoid overlap
  text(loadings[,1]*arrow_scale*1.1, loadings[,2]*arrow_scale*1.1, rownames(loadings), col = "darkred", cex = 0.8)
  

• Use the ggbiplot package for flexible, clean visuals
  If you're open to ggplot2-based tools, ggbiplot is a game-changer for PCA plots. It handles label placement better and supports grouping:

  # Install if needed: install.packages("ggbiplot")
  library(ggbiplot)
  
  # Plot with optimized labels and optional grouping (e.g., by ocean proximity)
  ggbiplot(housingpr, obs.scale = 1, var.scale = 1, 
           groups = your_housing_data$ocean_proximity, # Replace with your categorical variable
           ellipse = TRUE, circle = TRUE) +
    theme_minimal() +
    theme(legend.position = "bottom",
          panel.grid = element_blank()) +
    labs(title = "PCA Biplot - California Housing Dataset",
         x = pc1_label, y = pc2_label)
  

• Tweak element sizes to reduce clutter
  In base R, use the cex parameter to shrink labels and points—critical for your 20k+ observations:

  biplot(housingpr, scale = 0, cex = 0.7, xlab = pc1_label, ylab = pc2_label)
  

  Smaller text and points mean less overlap without losing key information.

• Remove unnecessary grid lines
  Grid lines can make a dense plot look busier. Turn them off in base R with grid = FALSE:

  biplot(housingpr, scale = 0, grid = FALSE, cex = 0.7, xlab = pc1_label, ylab = pc2_label)
  

内容的提问来源于stack exchange，提问作者user13601395