根据另一列出现百分比从数据框中选择行

Question

我有一个这样的数据集

1A1HI_R071_PH_INSPECT_VIS_1_2_201231_025816.JPG 1A  1A1HI
1A1JK_R071_PH_INSPECT_VIS_1_2_210115_121554.JPG 1A  1A1JK
1P3G6_R071_PH_INSPECT_VIS_2_2_201231_034741.JPG 1P  1P3G6
1P3GC_R071_PH_INSPECT_VIS_3_2_201107_140047.JPG 1P  1P3GC
M10L0_R071_PH_INSPECT_VIS_6_2_201121_071741.JPG M1  M10L0
M10L4_R071_PH_INSPECT_VIS_8_2_201201_142646.JPG M1  M10L4
S5148_R071_PH_INSPECT_VIS_1_2_201127_042210.JPG S5  S5148
S516U_R071_PH_INSPECT_VIS_5_2_201222_074443.JPG S5  S516U
V929S_R071_PH_INSPECT_VIS_8_2_201120_144633.JPG V9  V929S
V92B0_R071_PH_INSPECT_VIS_4_2_201121_095537.JPG V9  V92B0
V92B0_R071_PH_INSPECT_VIS_4_2_201121_095539.JPG V9  V92B0
V92EM_R071_PH_INSPECT_VIS_2_2_210105_133406.JPG V9  V92EM
W405K_R071_PH_INSPECT_VIS_11_2_201021_230940.JPG    W4  
W405O_R071_PH_INSPECT_VIS_2_2_201206_095433.JPG W4  W405O
W40EW_R071_PH_INSPECT_VIS_3_3_201219_120634.JPG W4  W40EW
W40EW_R071_PH_INSPECT_VIS_5_3_201220_072010.JPG W4  W40EW
W40EW_R071_PH_INSPECT_VIS_5_3_201220_072019.JPG W4  W40EW
X103K_R071_PH_INSPECT_VIS_2_3_210112_185054.JPG X1  X103K
1A1HI_R071_PH_INSPECT_VIS_1_4_201231_025833.JPG 1A  1A1HI
1A1RE_R071_PH_INSPECT_VIS_1_4_201227_153637.JPG 1A  1A1RE
1P3G6_R071_PH_INSPECT_VIS_2_4_201231_034806.JPG 1P  1P3G6
1P3GC_R071_PH_INSPECT_VIS_3_4_201107_140102.JPG 1P  1P3GC
1P3HO_R071_PH_INSPECT_VIS_6_4_201214_113511.JPG 1P  1P3HO
1P3HQ_R071_PH_INSPECT_VIS_5_4_201207_191653.JPG 1P  1P3HQ
5X6X6_R071_PH_INSPECT_VIS_3_4_201211_142453.JPG 5X  5X6X6
A70NG_R071_PH_INSPECT_VIS_5_4_201025_182537.JPG A7  A70NG
M10L0_R071_PH_INSPECT_VIS_6_4_201121_071750.JPG M1  M10L0
M10L4_R071_PH_INSPECT_VIS_8_4_201201_142701.JPG M1  M10L4
V929S_R071_PH_INSPECT_VIS_8_4_201120_144651.JPG V9  V929S
V92EM_R071_PH_INSPECT_VIS_2_4_210105_133438.JPG V9  V92EM
W405O_R071_PH_INSPECT_VIS_2_4_201206_095500.JPG W4  W405O
W4078_R071_PH_INSPECT_VIS_5_4_201215_153919.JPG W4  W4078
W40BK_R071_PH_INSPECT_VIS_2_4_210113_175802.JPG W4  W40BK
W40EW_R071_PH_INSPECT_VIS_5_4_201220_072024.JPG W4  W40EW
1A1HI_R071_PH_INSPECT_VIS_1_5_201231_025836.JPG 1A  1A1HI
1A1JK_R071_PH_INSPECT_VIS_1_5_210115_121617.JPG 1A  1A1JK
1A1RE_R071_PH_INSPECT_VIS_1_5_201227_153639.JPG 1A  1A1RE
1P3G6_R071_PH_INSPECT_VIS_2_5_201231_034809.JPG 1P  1P3G6
1P3GC_R071_PH_INSPECT_VIS_3_5_201107_140105.JPG 1P  1P3GC

第一列是图像名称，第二列是产品名称

数据集中有这么多产品。

如何根据第二列出现百分比选择图像（第1列）

例如，我需要选择第1列中包含1P的170个随机行（图像），第2列中包含1A的156个随机行，以此类推，以获得每个产品类别中20%的图像来构建训练集

Answer 1

您可以使用lambda对每个组进行采样，如下所示：

import pandas as pd
df = pd.read_csv("example.csv")
df.columns = ["image", "product"]

grouped = df.groupby("product")
result = grouped.apply(lambda x: x.sample(frac=0.2))

Answer 2

可以使用DataFrameGroupBy.sample对每个类别中的行进行采样

n = 0.2  # 20% per category
# Sample dataframe
df = pandas.DataFrame({
        'image_id': [1,2,3,4,5,6,7],
        'product_category': ['A', 'A', 'A', 'A', 'A', 'B', 'B']
})

df.groupby('product_category').sample(frac=n)

但是，请注意，如果抽样计数低于1，则某些类别可能不会返回任何行