关于另一个 csv、pandas 上的操作（聚合最大值）的相应 csv 值

Question

我有两张均值表和标准值表，我想计算均值表某些列的最大值，并且我想得到标准值表的对应行。

mean_df = pd.read_csv(r'./csvs/mean.csv')

std_df = pd.read_csv(r'./csvs/std.csv')

#mean_df, std_df are of same size

grouped_df = mean_df.groupby(['alpha', 'beta'])

columns = ['val']
max_df = grouped_df[columns].agg(['max'])

# Here i want the corresponding std_max_df table for the max_df. i.e., for every max calculated from mean, i want the std of that max in a new table.

例如：

输入 mean_df 是

阿尔法贝塔伽马值

1 2 3 100

4 6 8 200

1 2 9 400

4 6 7 500

3 5 8 600

---

输入 std_df 是

阿尔法贝塔伽马值

1 2 3 300

4 6 8 500

1 2 9 100

4 6 7 700

3 5 8 900

---

输出将是

alpha beta gamma max_mean_val corresp_std_val

1 2 9 400 100

4 6 7 500 700

3 5 8 600 900

---

Answer 1

首先通过内连接或左连接将DataFrames合并在一起，然后通过DataFrameGroupBy.idxmax获取具有最大索引值的行：

df = mean_df.merge(std_df, on=['alpha', 'beta','gamma'], 
                           how='left', 
                           suffixes=('_mean','_std'))

df = df.loc[df.groupby(['alpha', 'beta'])['val_mean'].idxmax()]
print (df)
   alpha  beta  gamma  val_mean  val_std
2      1     2      9       400      100
4      3     5      8       600      900
3      4     6      7       500      700

对于多列一个可能的想法：

print (mean_df)

   alpha  beta  gamma  val  val1
0      1     2      3  100     5
1      4     6      8  200     9
2      1     2      9  400     3
3      4     6      7  500     5
4      3     5      8  600     2

print (std_df)
   alpha  beta  gamma  val  val1
0      1     2      3  300     5
1      4     6      8  500     7
2      1     2      9  100     9
3      4     6      7  700     2
4      3     5      8  900     1


df = mean_df.merge(std_df, on=['alpha', 'beta','gamma'], 
                           how='left', 
                           suffixes=('_mean','_std'))

cols = ['val','val1']

dfs = [df.loc[df.groupby(['alpha', 'beta'])[f'{col}_mean'].idxmax(), 
              [f'{col}_mean', f'{col}_std']].reset_index(drop=True) for col in cols]

df0 = df[['alpha', 'beta','gamma']].drop_duplicates(['alpha','beta']).reset_index(drop=True)
df = pd.concat([df0] + dfs, axis=1)
print (df)
   alpha  beta  gamma  val_mean  val_std  val1_mean  val1_std
0      1     2      3       400      100          5         5
1      4     6      8       600      900          2         1
2      3     5      8       500      700          9         7