如何对Y变量做对数变换并设置动态值域适配多图表？

核心需求回顾 #

• 对因变量virus执行10对数变换，优化数据可视化效果
• Y轴值域设为当前数据的±10%，支持筛选后动态调整
• 保留Loess回归线，且与散点在对数刻度上对齐
• 解决仅用log scale时，筛选数据导致图表消失的问题

原代码问题分析 #

1. 存在两个独立的transform数组，Vega-Lite仅会执行最后一个，导致第一个对数变换逻辑被覆盖
2. 动态计算的是对数变换后的值域，但Y轴仍使用原始virus字段，逻辑不匹配
3. Loess回归基于原始virus值计算，未与对数变换对齐，导致回归线偏离

修正后的完整代码 #

{
  "data": {"name": "dataset"},
  "transform": [
    // 计算virus的10对数变换值
    {"calculate": "log10(datum.virus)", "as": "log_virus"},
    // 全局计算log_virus的最值
    {
      "joinaggregate": [
        {"op": "max", "field": "log_virus", "as": "log_max"},
        {"op": "min", "field": "log_virus", "as": "log_min"}
      ]
    },
    // 计算对应原始值±10%的对数值域：log10(原始值*1.1)=log10(原始值)+log10(1.1)
    {"calculate": "datum.log_max + log10(1.1)", "as": "domain_max"},
    {"calculate": "datum.log_min - log10(1.1)", "as": "domain_min"}
  ],
  "params": [
    {"name": "y_domain", "expr": "[data('dataset')[0]['domain_min'], data('dataset')[0]['domain_max']]"}
  ],
  "layer": [
    // 散点图层：使用原始virus值，Y轴启用log刻度+动态值域
    {
      "mark": {"type": "point", "tooltip": true, "filled": true},
      "encoding": {
        "x": {"field": "date", "title": "Date", "type": "temporal"},
        "y": {
          "field": "virus",
          "title": "Normalized Levels (N1/PMMOV Concentrations)",
          "type": "quantitative",
          "scale": {
            "type": "log",
            "domain": {"expr": "y_domain"},
            "base": 10
          },
          "axis": {"format": ".0f"} // 显示原始数值刻度，保证可读性
        },
        "tooltip": [
          {"field": "virus", "type": "quantitative"},
          {"field": "date", "type": "temporal", "format": "%B %d %Y"},
          {"field": "siteno", "type": "nominal"},
          {"field": "virus_o", "type": "nominal"}
        ]
      }
    },
    // Loess回归图层：基于对数变换值计算，再转换回原始值匹配Y轴
    {
      "name": "Regression Line",
      "transform": [
        {
          "loess": "log_virus",
          "on": "date",
          "bandwidth": 0.25,
          "as": ["date", "predicted_log_virus"]
        },
        {"calculate": "pow(10, datum.predicted_log_virus)", "as": "predicted_virus"}
      ],
      "mark": {"type": "line", "color": "#D64550"},
      "encoding": {
        "x": {
          "field": "date",
          "type": "temporal",
          "axis": {"format": "MMMM yyyy", "formatType": "pbiFormat"}
        },
        "y": {
          "field": "predicted_virus",
          "type": "quantitative",
          "scale": {
            "type": "log",
            "domain": {"expr": "y_domain"},
            "base": 10
          }
        }
      }
    }
  ]
}

关键修改说明 #

1. 合并transform执行流：将原本独立的两个transform数组合并，确保对数变换和值域计算逻辑依次生效
2. 精确对齐值域逻辑：基于对数变换后的值计算±10%原始值对应的对数范围，避免直接缩放对数值导致的偏差
3. 修正Loess回归逻辑：先对对数变换后的值做回归，再转换回原始值，保证回归线与散点在同一对数刻度上对齐
4. 统一Y轴配置：散点和回归线的Y轴均启用log刻度和动态值域，同时保留原始数值刻度显示，平衡可视化效果与可读性
5. 简化参数定义：用单个参数存储值域数组，避免重复定义

可选优化方案 #

如果需要严格基于原始值的±10%计算值域，可以替换transform中的值域计算逻辑为：

{
  "joinaggregate": [
    {"op": "max", "field": "virus", "as": "raw_max"},
    {"op": "min", "field": "virus", "as": "raw_min"}
  ],
  {"calculate": "log10(datum.raw_max * 1.1)", "as": "domain_max"},
  {"calculate": "log10(datum.raw_min * 0.9)", "as": "domain_min"}
}

内容的提问来源于stack exchange，提问作者Chumbi