Altair中拼接地图时transform_calculate关联类别图例排序异常修复

我尝试将多张通过字符串字典标注的分级统计图（choropleth）拼接，它们共享图例。但拼接后图例排序混乱，原本的首个类别变成了最后一个，单独绘制单张地图时代码运行正常。

原始代码 #

import altair as alt
import geopandas as gpd
import pickle
import requests


resp = requests.get("https://raw.githubusercontent.com/ccsuehara/cfi/main/mapping.pickle")
mapping = pickle.loads(resp.content)
srcs = gpd.read_file("https://raw.githubusercontent.com/ccsuehara/cfi/main/example.geojson")


base = alt.Chart(srcs).mark_geoshape().properties(
    width=400,
    height=400
).project(
    type='mercator'  
)

g_ = alt.concat(
    base.encode(
    color=alt.Color(
        'disp_2010:N',
        scale=alt.Scale(scheme='greens'),
        title = "% Change (winsor)",
        sort=alt.EncodingSortField('class_2010_win',  order='ascending')
       ),
    ).transform_filter(
        'isValid(datum.class_2010_win)'
    ).transform_calculate(
        disp_2010=f"{mapping}[datum.class_2010_win]"
    ).properties(
        title="Δ 2005 - 2010(%)"
    )    | base.encode(
    color=alt.Color(
        'disp_2015:N',
        scale=alt.Scale(scheme='greens'),
      title = "% Change (winsor)",
     #sort=alt.EncodingSortField('class_2010_win',  order='ascending')
      ),
    ).transform_filter(
        'isValid(datum.class_2015_win)'
    ).transform_calculate(
        disp_2015=f"{mapping}[datum.class_2015_win]",
    ).properties(
        title="Δ 2010 - 2015(%)"
    )
)

g_

问题原因 #

拼接图表共享图例时，Altair会合并两个图表的排序规则，但第二个图表未指定排序逻辑，导致整体排序被打乱。同时，transform_calculate生成的离散字段排序依赖各自的数值类字段（class_2010_win和class_2015_win），拼接时规则冲突引发排序异常。

修复方案 #

方案1：统一两个图表的排序依据 #

给第二个图表添加与第一个图表逻辑一致的排序参数，基于各自的数值类别字段：

# 修改第二个图表的color编码
color=alt.Color(
    'disp_2015:N',
    scale=alt.Scale(scheme='greens'),
    title = "% Change (winsor)",
    sort=alt.EncodingSortField('class_2015_win',  order='ascending')
),

方案2：提前定义固定排序顺序（更稳定） #

直接从mapping字典中提取固定的类别顺序，作为两个图表的排序基准，彻底避免字段依赖的不确定性：

# 从mapping字典获取按键排序后的标签列表
sorted_labels = [v for k, v in sorted(mapping.items())]

# 在两个图表的color编码中使用该固定顺序
color=alt.Color(
    'disp_2010:N',
    scale=alt.Scale(scheme='greens', domain=sorted_labels),
    title = "% Change (winsor)",
    sort=sorted_labels
),

完整修复代码 #

import altair as alt
import geopandas as gpd
import pickle
import requests


resp = requests.get("https://raw.githubusercontent.com/ccsuehara/cfi/main/mapping.pickle")
mapping = pickle.loads(resp.content)
# 提前生成固定排序的类别标签
sorted_labels = [v for k, v in sorted(mapping.items())]
srcs = gpd.read_file("https://raw.githubusercontent.com/ccsuehara/cfi/main/example.geojson")


base = alt.Chart(srcs).mark_geoshape().properties(
    width=400,
    height=400
).project(
    type='mercator'  
)

g_ = alt.concat(
    base.encode(
    color=alt.Color(
        'disp_2010:N',
        scale=alt.Scale(scheme='greens', domain=sorted_labels),
        title = "% Change (winsor)",
        sort=sorted_labels
       ),
    ).transform_filter(
        'isValid(datum.class_2010_win)'
    ).transform_calculate(
        disp_2010=f"{mapping}[datum.class_2010_win]"
    ).properties(
        title="Δ 2005 - 2010(%)"
    )    | base.encode(
    color=alt.Color(
        'disp_2015:N',
        scale=alt.Scale(scheme='greens', domain=sorted_labels),
        title = "% Change (winsor)",
        sort=sorted_labels
      ),
    ).transform_filter(
        'isValid(datum.class_2015_win)'
    ).transform_calculate(
        disp_2015=f"{mapping}[datum.class_2015_win]",
    ).properties(
        title="Δ 2010 - 2015(%)"
    )
)

g_

内容的提问来源于stack exchange，提问作者Carla S.