在R语言中利用散点图坐标提取直线两侧的数据框列

Got it, let's break this down step by step. You already have your PCA biplot where each point represents a column from your original data (since the labels are the row names of prin_comp$rotation—those map directly to your original columns). Here's how to pull out the column names on either side of your line:

Step 1: Tidy up your rotation data #

First, let's convert the rotation matrix into a easy-to-work-with data frame, and explicitly save your original column names as a separate column (instead of just keeping them as row names):

# Extract PC1 and PC2 values, plus add original column names as a column
rot_df <- as.data.frame(prin_comp$rotation[, c(1, 2)])
rot_df$original_col <- rownames(rot_df)

Step 2: Define your line's mathematical equation #

You know the line you drew—let's write it in the form y = m*x + b, where:

• y = PC2 (your plot's y-axis)
• x = PC1 (your plot's x-axis)
• m = the slope of your line
• b = the y-intercept of your line

For example, if your line was something like PC2 = 0.8PC1 + 0.1*, then m = 0.8 and b = 0.1.

Step 3: Filter column names by their position relative to the line #

Use a simple inequality to sort points into left/right sides. The direction of the inequality depends on which side you consider "left" or "right"—test with a point you can visually confirm to get the direction right.

Get left-side column names

Suppose points where PC2 < m*PC1 + b are on the left side of your line:

# Replace m and b with your line's actual slope and intercept
m <- 0.8
b <- 0.1

left_cols <- rot_df$original_col[rot_df$PC2 < (m * rot_df$PC1) + b]

Get right-side column names

Just flip the inequality to grab the other side:

right_cols <- rot_df$original_col[rot_df$PC2 > (m * rot_df$PC1) + b]

Optional: Verify with a colored plot #

If you want to double-check you're picking the right points, add a "side" label to your data and redraw the plot:

# Add a column to mark which side each point is on
rot_df$side <- ifelse(rot_df$PC2 < (m * rot_df$PC1) + b, "Left", "Right")

# Redraw with colored points and your dashed line
qplot(x = PC1, y = PC2, data = rot_df, color = side) +
  geom_text(aes(label = original_col)) +
  geom_abline(slope = m, intercept = b, linetype = "dashed") +
  ggtitle(my_title)

This will let you visually confirm that your code is grouping points exactly as you want.

The core idea is translating your drawn line into a mathematical rule, then using that rule to filter your rotation data—once you have that, pulling the column names is super straightforward!

内容的提问来源于stack exchange，提问作者Glubbdrubb