Pandas DataFrame应用样式后round方法失效的问题排查

问题场景 #

给DataFrame添加行高亮和表格样式后，尝试对start_hyp_minus_ref、dur_hyp_minus_ref列保留两位小数，使用.style.format()或DataFrame.round()方法均无效；但移除样式代码后，小数格式化功能正常。

原代码 #

import pandas as pd

def printSegments(segments_ref, segments_hyp, timestamps_ref, timestamps_hyp):
    def highlight_unequal_rows(row):
        if len(row['reference'].split()) != len(row['hypothesis'].split()):
            return ['background-color: red'] * len(row)
        return [''] * len(row)
    
    start_diff = [th["start"] - tr["start"] for th, tr in zip(timestamps_hyp, timestamps_ref)]
    dur_diff = [th["dur"] - tr["dur"] for th, tr in zip(timestamps_hyp, timestamps_ref)]
    df = pd.DataFrame(dict(reference=segments_ref, hypothesis=segments_hyp, start_hyp_minus_ref=start_diff, dur_hyp_minus_ref=dur_diff))
    pd.set_option('display.max_colwidth', None)

    df = df.style.apply(highlight_unequal_rows, axis=1)
    df = df.set_table_styles(
        [
            {'selector': 'td', 'props': [('direction', 'ltr'), ('text-align', 'left')]},
            {'selector': 'th', 'props': [('direction', 'ltr'), ('text-align', 'left')]}
        ],
        overwrite=False
    )
    return df

尝试过的无效方法 #

1. 错误调用Styler格式化方法（未对已有Styler对象操作）：

f = {'start_hyp_minus_ref':'{:.2f}'}
df.style.format(f).bar(subset='start_hyp_minus_ref')

2. 转换为Styler后尝试用DataFrame方法修改列值：

df['start_hyp_minus_ref'] = df['start_hyp_minus_ref'].round(2)
df['dur_hyp_minus_ref'] = df['dur_hyp_minus_ref'].round(2)

解决方案 #

问题核心是操作顺序错误：一旦将DataFrame转换为Styler对象，就无法再用DataFrame的方法修改数据；且格式化操作需要在样式、高亮之前执行，或直接在Styler对象上链式调用格式化方法。

修改后的代码：

import pandas as pd

def printSegments(segments_ref, segments_hyp, timestamps_ref, timestamps_hyp):
    def highlight_unequal_rows(row):
        if len(row['reference'].split()) != len(row['hypothesis'].split()):
            return ['background-color: red'] * len(row)
        return [''] * len(row)
    
    start_diff = [th["start"] - tr["start"] for th, tr in zip(timestamps_hyp, timestamps_ref)]
    dur_diff = [th["dur"] - tr["dur"] for th, tr in zip(timestamps_hyp, timestamps_ref)]
    df = pd.DataFrame(dict(reference=segments_ref, hypothesis=segments_hyp, start_hyp_minus_ref=start_diff, dur_hyp_minus_ref=dur_diff))
    pd.set_option('display.max_colwidth', None)

    # 先执行小数格式化，再应用高亮和表格样式
    styler = df.style.format({
        'start_hyp_minus_ref': '{:.2f}',
        'dur_hyp_minus_ref': '{:.2f}'
    })
    styler = styler.apply(highlight_unequal_rows, axis=1)
    styler = styler.set_table_styles(
        [
            {'selector': 'td', 'props': [('direction', 'ltr'), ('text-align', 'left')]},
            {'selector': 'th', 'props': [('direction', 'ltr'), ('text-align', 'left')]}
        ],
        overwrite=False
    )
    return styler

也可以用链式调用简化写法：

# 链式调用版本
styler = df.style.format({
            'start_hyp_minus_ref': '{:.2f}',
            'dur_hyp_minus_ref': '{:.2f}'
        }).apply(highlight_unequal_rows, axis=1).set_table_styles(
            [
                {'selector': 'td', 'props': [('direction', 'ltr'), ('text-align', 'left')]},
                {'selector': 'th', 'props': [('direction', 'ltr'), ('text-align', 'left')]}
            ],
            overwrite=False
        )

测试用例 #

printSegments(["I went to the school"], ["I want to the school"], [{"start": 0.69, "dur": 1.5}], [{"start": 0.79, "dur": 1.3}])

运行后，start_hyp_minus_ref会显示0.10，dur_hyp_minus_ref显示-0.20，同时因两行单词数量一致，不会触发红色高亮。

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