Python Pandas：保持列值最高的行

Question

Say I have a dataframe of test scores of students, where each student studies different subjects. Each student can take the test for each subject multiple times, and only the highest score (out of 100) will be retained. For instance, say I have a dataframe of all test records:

| student_name | subject | test_number | score | 
|--------------|---------|-------------|-------|
| sarah        | maths   | test1       | 78    |
| sarah        | maths   | test2       | 71    |
| sarah        | maths   | test3       | 83    |
| sarah        | physics | test1       | 91    |
| sarah        | physics | test2       | 97    |
| sarah        | history | test1       | 83    |
| sarah        | history | test2       | 87    |
| joan         | maths   | test1       | 83    |
| joan         | maths   | test2       | 88    |

(1) How do I keep only the test records (rows) with the maximum score? That is,

| student_name | subject | test_number | score | 
|--------------|---------|-------------|-------|
| sarah        | maths   | test1       | 78    |
| sarah        | maths   | test2       | 71    |
| sarah        | maths   | test3       | 83    |
| sarah        | physics | test1       | 91    |

(2) How would I keep the average of all tests taken for the same subject, for the same student? That is:

| student_name | subject | test_number | ave_score | 
|--------------|---------|-------------|-----------|
| sarah        | maths   | na          | 77.333    |
| sarah        | maths   | na          | 94        |
| sarah        | maths   | na          | 85        |
| sarah        | physics | na          | 85.5      |

I've tried various combinations of df.sort_values() and df.drop_duplicates(subset=..., keep=...) , to no avail.

Actual Data

| query | target   | pct-similarity | p-val | aln_length | bit-score |
|-------|----------|----------------|-------|------------|-----------|
| EV239 | B/Fw6/623 | 99.23         | 0.966 |  832       | 356       |
| EV239 | B/Fw6/623 | 97.34         | 0.982 |  1022      | 739       |
| EV239 | MMS-alpha | 92.23         | 0.997 |  838       | 384       |
| EV239 | MMS-alpha | 93.49         | 0.993 |  1402      | 829       |
| EV380 | B/Fw6/623 | 94.32         | 0.951 |  324       | 423       |
| EV380 | B/Fw6/623 | 95.27         | 0.932 |  1245      | 938       |
| EV380 | MMS-alpha | 99.23         | 0.927 |  723       | 522       |
| EV380 | MMS-alpha | 99.15         | 0.903 |  948       | 1092      |

After aggregation function is applied, only the column pct-similarity will be of interest.

(1) Drop duplicate query+target rows, by choosing the maximum aln_length . Retain the pct-similarity value that belongs to the row with maximum aln_length .

(2) Aggregate duplicate query+target rows by choosing the row with maximum aln_length , and computing the average pct-similarity for that set of duplicate rows. The other numerical columns aren't necessary and will be dropped eventually, so I really don't care what aggregation function (max or mean) is applied to them.

Answer 1

Just use max() to each group of student/subject:

df.groupby(["student_name","subject"], as_index=False).max()


    student_name    subject         test_number     score
0   joan            maths           test2           88
1   sarah           history         test2           87
2   sarah           maths           test3           83
3   sarah           physics         test2           97

---

For the average, this use mean() instead:

df.groupby(["student_name","subject"], as_index=False).mean()

    student_name    subject     score
0   joan            maths       85.500000
1   sarah           history     85.000000
2   sarah           maths       77.333333
3   sarah           physics     94.000000

Answer 2

Most likely describe can

df.groupby(["student_name","subject"]).score.describe()
Out[15]: 
                          count       mean       std   min    25%   50%  \
student_name   subject                                                    
 joan           maths       2.0  85.500000  3.535534  83.0  84.25  85.5   
 sarah          history     2.0  85.000000  2.828427  83.0  84.00  85.0   
                maths       3.0  77.333333  6.027714  71.0  74.50  78.0   
                physics     2.0  94.000000  4.242641  91.0  92.50  94.0   
                            75%   max  
student_name   subject                 
 joan           maths     86.75  88.0  
 sarah          history   86.00  87.0  
                maths     80.50  83.0  
                physics   95.50  97.0  

And with drop_duplicates

df.sort_values('score').drop_duplicates(["student_name","subject"],keep='last')
Out[22]: 
     student_name    subject    test_number  score
2   sarah           maths      test3            83
6   sarah           history    test2            87
8   joan            maths      test2            88
4   sarah           physics    test2            97

For mean value with reindex

df.groupby(["student_name","subject"], as_index=False).mean().reindex(columns=df.columns)
Out[24]: 
     student_name    subject  test_number      score
0   joan            maths             NaN  85.500000
1   sarah           history           NaN  85.000000
2   sarah           maths             NaN  77.333333
3   sarah           physics           NaN  94.000000

Answer 3

We can use agg on a groupby to get 'idxmax' and 'mean' .
With that we can perform an inner join to get both the correct rows and means.

df.join(
    df.groupby(['student_name', 'subject'])
      .score.agg(['idxmax', 'mean']).set_index('idxmax'),
    how='inner'
)

  student_name  subject test_number  score       mean
2        sarah    maths       test3     83  77.333333
4        sarah  physics       test2     97  94.000000
6        sarah  history       test2     87  85.000000
8         joan    maths       test2     88  85.500000