如何仅用ggplot实现预测模型与观测值的分离图例？

要实现你想要的分离图例可视化，无需借助其他绘图工具，以下两种纯ggplot方案可以达成目标：

方案一：整合数据格式生成分离图例 #

先把观测值和预测模型数据统一整理成长格式，通过区分color（映射预测模型）和shape/linetype（映射观测值）的美学映射，让ggplot自动生成两个独立图例：

library(dplyr)
library(tidyr)
library(ggplot2)

n = 5

df = data.frame(    
    time = c(1 : n),
    prediction_model1 = rnorm(n) %>% sort,
    prediction_model2 = rnorm(n) %>% sort,
    observed_values = rnorm(n) %>% sort    
)

# 将所有序列转成长格式，新增字段区分预测模型和观测值
df_plot = df %>% 
    pivot_longer(cols = everything(), names_to = "series", values_to = "happyness") %>%
    mutate(
        group_type = case_when(
            grepl("prediction", series) ~ "预测模型",
            series == "observed_values" ~ "观测值"
        ),
        model_name = ifelse(group_type == "预测模型", sub("prediction_", "", series), NA)
    )

ggplot(df_plot) +
    # 绘制预测模型的点和线，用颜色区分模型
    geom_point(aes(x = time, y = happyness, color = model_name), size = 2, data = . %>% filter(group_type == "预测模型")) +
    geom_line(aes(x = time, y = happyness, color = model_name), data = . %>% filter(group_type == "预测模型")) +
    # 绘制观测值的点和线，用形状和线型区分
    geom_point(aes(x = time, y = happyness, shape = group_type), size = 4, data = . %>% filter(group_type == "观测值")) +
    geom_line(aes(x = time, y = happyness, linetype = group_type), data = . %>% filter(group_type == "观测值")) +
    # 配置图例标签与样式
    labs(color = "预测模型", shape = "", linetype = "") +
    scale_shape_manual(values = c(5)) +
    scale_linetype_manual(values = c("solid")) +
    theme(
        legend.position = "bottom",
        legend.box = "vertical",
        legend.spacing.y = unit(0.2, "cm")
    )

方案二：用ggnewscale添加独立图例 #

如果不想调整原始数据结构，使用ggnewscale包可以为观测值单独创建一个图例，和预测模型的图例分开：

library(dplyr)
library(tidyr)
library(ggplot2)
library(ggnewscale)

n = 5

df = data.frame(    
    time = c(1 : n),
    prediction_model1 = rnorm(n) %>% sort,
    prediction_model2 = rnorm(n) %>% sort,
    observed_values = rnorm(n) %>% sort    
)

df_plot = df %>% 
    pivot_longer(cols = c(prediction_model1, prediction_model2), names_prefix = "prediction_", values_to = "happyness")

ggplot(df_plot) +
    # 绘制预测模型的点和线，生成第一个颜色图例
    geom_point(aes(x = time, y = happyness, color = name), size = 2) +
    geom_line(aes(x = time, y = happyness, color = name)) +
    labs(color = "预测模型") +
    # 启动新的图例上下文，用于观测值
    new_scale("color") +
    new_scale("shape") +
    # 绘制观测值的点和线，生成第二个独立图例
    geom_point(aes(x = time, y = observed_values, color = "观测值", shape = "观测值"), size = 4) +
    geom_line(aes(x = time, y = observed_values, color = "观测值")) +
    # 设置观测值的图例样式
    scale_color_manual(name = "", values = c("观测值" = "black")) +
    scale_shape_manual(name = "", values = c("观测值" = 5)) +
    theme(
        legend.position = "bottom",
        legend.box = "vertical",
        legend.spacing.y = unit(0.2, "cm")
    )

两种方案都能实现目标图表的分离图例效果，可根据需求选择。

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