使用R中lm()与bs()绘制样条遇行数不匹配错误及相关咨询

错误原因分析 #

你遇到的'newdata' has 38 rows, but the variables found have 700 rows错误，根源有两个：

1. 构建模型时直接用了data_clean$visits、data_clean$discount这种硬编码列引用，而非数据框内的变量名。这会让模型把这些向量当作全局变量，而非数据框的列，导致predict无法匹配newdata里的变量。
2. 模型包含discount变量，但调用predict时的newdata只提供了visits，缺少discount的取值，模型无法完成预测。
   另外，bs()里同时设置df=3和knots=3存在冲突：df=3的样条默认只需要2个内部节点，手动设knots=3会导致自由度不匹配，建议只保留df参数。

疑问解答 #

1. 如何从样条模型生成可用于绘图的曲线？ #

按以下步骤操作：

• 修正模型定义：使用变量名而非硬编码列引用，让模型正确关联数据框变量：

  spline_model <- lm(sales ~ bs(visits, df = 3) + discount, data = data_clean)
  

• 生成连续预测序列：为visits生成覆盖全范围的连续序列，同时给discount指定固定值（比如均值，或你想观察的特定折扣率）：

  visits_seq <- seq(min(data_clean$visits), max(data_clean$visits), length.out = 100) # 生成更平滑的序列
  # 固定discount为均值，也可设为特定值如0.1
  new_data <- data.frame(visits = visits_seq, discount = mean(data_clean$discount))
  

• 预测拟合值：

  spline_pred <- predict(spline_model, newdata = new_data)
  

• 绘图：用基础绘图或ggplot2绘制原始点+拟合曲线：

  # 基础绘图
  plot(data_clean$visits, data_clean$sales, main = "观测值 vs 拟合值", xlab = "访问量", ylab = "销售额")
  lines(visits_seq, spline_pred, col = "red", lwd = 3)
  
  # ggplot2版本
  library(ggplot2)
  ggplot(data_clean, aes(x = visits, y = sales)) +
    geom_point(alpha = 0.5) +
    geom_line(data = new_data, aes(y = spline_pred), color = "red", linewidth = 1) +
    labs(title = "观测值 vs 样条拟合曲线", x = "访问量", y = "销售额")
  

2. 为什么需要用seq生成序列+predict绘图，而非直接用原模型？ #

核心原因有两点：

• 原始数据离散且分布不均：你的原始visits数据可能存在重复值、稀疏区间，直接用模型对原始数据的拟合值绘图，得到的是零散的点而非平滑曲线，无法直观展示变量间的整体趋势。
• 样条模型拟合连续关系：样条的核心是捕捉变量间的非线性连续趋势，生成覆盖全范围的连续visits序列，能让模型预测出每个点的拟合值，从而绘制出平滑的趋势曲线，完整展示变量间的非线性关联。

3. 如何绘制3D图展示sales、visits、discount的关联？ #

可以用plotly包绘制交互式3D曲面图，直观展示三个变量的关系：

library(plotly)

# 生成visits和discount的网格数据
visits_grid <- seq(min(data_clean$visits), max(data_clean$visits), length.out = 50)
discount_grid <- seq(min(data_clean$discount), max(data_clean$discount), length.out = 50)
grid_data <- expand.grid(visits = visits_grid, discount = discount_grid)

# 预测所有网格点的sales值
grid_data$sales_pred <- predict(spline_model, newdata = grid_data)

# 绘制3D曲面图
plot_ly(grid_data, x = ~visits, y = ~discount, z = ~sales_pred) %>%
  add_surface(colorscale = "Viridis") %>%
  add_markers(data = data_clean, x = ~visits, y = ~discount, z = ~sales, 
              marker = list(color = "black", size = 3)) %>%
  layout(scene = list(xaxis = list(title = "访问量"), 
                      yaxis = list(title = "折扣率"), 
                      zaxis = list(title = "销售额")))

这个图会显示样条模型拟合的3D曲面，同时叠加原始数据点，清晰展示两个自变量对因变量的联合影响。

完整修正代码 #

library(dplyr)
library(splines)
library(ggplot2)
library(plotly)

# 加载数据
data_path <- "https://raw.githubusercontent.com/juanitorduz/website_projects/master/data/sales_dag.csv"
data <- read.csv(data_path)
data_clean <- data %>% select(visits, discount, sales)

# 修正后的样条模型
spline_model <- lm(sales ~ bs(visits, df = 3) + discount, data = data_clean)

# 生成预测序列并绘图（2D）
visits_seq <- seq(min(data_clean$visits), max(data_clean$visits), length.out = 100)
new_data <- data.frame(visits = visits_seq, discount = mean(data_clean$discount))
spline_pred <- predict(spline_model, newdata = new_data)

ggplot(data_clean, aes(x = visits, y = sales)) +
  geom_point(alpha = 0.5) +
  geom_line(data = new_data, aes(y = spline_pred), color = "red", linewidth = 1) +
  labs(title = "观测销售额 vs 样条拟合曲线", x = "网站访问量", y = "销售额")

# 绘制3D关联图
visits_grid <- seq(min(data_clean$visits), max(data_clean$visits), length.out = 50)
discount_grid <- seq(min(data_clean$discount), max(data_clean$discount), length.out = 50)
grid_data <- expand.grid(visits = visits_grid, discount = discount_grid)
grid_data$sales_pred <- predict(spline_model, newdata = grid_data)

plot_ly(grid_data, x = ~visits, y = ~discount, z = ~sales_pred) %>%
  add_surface(colorscale = "Viridis") %>%
  add_markers(data = data_clean, x = ~visits, y = ~discount, z = ~sales, 
              marker = list(color = "black", size = 3)) %>%
  layout(scene = list(xaxis = list(title = "访问量"), 
                      yaxis = list(title = "折扣率"), 
                      zaxis = list(title = "销售额")))

内容的提问来源于stack exchange，提问作者tt33tt