Python如何基于固定阈值对pandas DataFrame数据实现分箱处理

你的需求属于动态贪心分箱逻辑，pandas内置的cut和qcut确实不直接适配这类场景：cut是基于预定义的固定区间分箱，qcut是基于分位数区间分箱，而你的分箱起始点由前一个分箱的结束位置动态决定，需要自定义逻辑实现。

以下是可直接运行的实现方案，输出完全符合你的预期：

import pandas as pd

# 示例数据
df = pd.DataFrame({'Marker': ['m1','m2','m3','m4','m5','m6','m7','m8','m9','m10'],
                  'Position': [0,0.5,0.6,2,5,7,8.4,15,16,17]})

threshold = 5
# 初始化分箱参数
current_start = df['Position'].iloc[0]
bin_id = 1
bin_ids = []

for pos in df['Position']:
    # 当前位置和分箱起始位置差值超过阈值，就新开一个分箱
    if pos - current_start > threshold:
        current_start = pos
        bin_id += 1
    bin_ids.append(bin_id)

# 给原表增加分箱编号列
df['bin_id'] = bin_ids

# 按分箱分组提取Marker列表
bin_result = df.groupby('bin_id')['Marker'].apply(list).to_dict()

# 打印输出
for bin_num, markers in bin_result.items():
    print(f"bin{bin_num} = {markers}")

运行输出：

bin1 = ['m1', 'm2', 'm3', 'm4', 'm5']
bin2 = ['m6', 'm7']
bin3 = ['m8', 'm9', 'm10']

如果处理的数据量较大，想避免循环提升效率，可以用向量化操作实现：

import pandas as pd
import numpy as np

threshold = 5
# 生成动态分箱编号
df['bin_id'] = 1
df['bin_id'] = df['bin_id'].where(df['Position'] - df['Position'].shift(1).where(df['Position'].diff() > threshold).ffill().fillna(df['Position'].iloc[0]) <= threshold, df['bin_id'].cumsum())
# 分组提取结果
bin_result = df.groupby('bin_id')['Marker'].apply(list).to_dict()

内容的提问来源于stack exchange，提问作者Amilovsky