如何用R的ggplot2绘制正则化逻辑回归的非线性决策边界

Got it, let's walk through how to recreate that decision boundary plot with ggplot2—since you already have your grid and prediction matrix ready, it's just a matter of getting your data into a format ggplot2 loves, then putting the pieces together.

Step 1: Reshape your grid data into a tidy data frame #

ggplot2 works best with long-format (tidy) data, so first we'll convert your u, v grid and z matrix into a single data frame. The key here is making sure the order of your z values matches the grid combinations:

# Create a data frame with all u-v grid pairs
grid_df <- expand.grid(u = u, v = v)
# Flatten the z matrix into a vector and bind it to the grid data frame
# as.vector() defaults to column-wise ordering, which matches expand.grid's output
grid_df$z <- as.vector(z)

Step 2: Build the ggplot2 plot #

Now we can use geom_contour to draw the p=0.5 decision boundary (which corresponds to z=0), then add your original data points with geom_point:

library(ggplot2)

ggplot() +
  # Draw the decision boundary contour (breaks = 0 targets the z=0 line)
  geom_contour(
    data = grid_df,
    aes(x = u, y = v, z = z),
    breaks = 0,  # This is equivalent to levels = 0 in base R contour()
    color = "darkred",
    size = 1.2
  ) +
  # Add your original test data points
  geom_point(
    data = data,
    aes(x = Test1, y = Test2),
    alpha = 0.7,  # Optional: make points slightly transparent
    color = "steelblue"
  ) +
  # Optional: polish the plot with labels and a clean theme
  labs(
    x = "Test 1",
    y = "Test 2",
    title = "Polynomial Regularized Logistic Regression Decision Boundary"
  ) +
  theme_minimal()

Quick notes to avoid issues: #

• Double-check that as.vector(z) is using the right order: base R's contour() reads z column-wise, and expand.grid() generates grid pairs in the same column-first order, so this should line up perfectly. If your contour looks messed up, you can try as.vector(t(z)) to transpose the matrix first (though that shouldn't be needed here).
• If you want to visualize the predicted probabilities across the grid (not just the boundary), you can use geom_tile(aes(fill = z)) to add a heatmap layer underneath the contour.

内容的提问来源于stack exchange，提问作者Melvin86