使用熊猫添加缺失的行

Question

这个问题与有关，但有点复杂。

我有一个 table 这样的：

ID    DEGREE    TERM    STATUS GRADTERM
1     Bachelors 20111   1
1     Bachelors 20116   1
2     Bachelors 20126   1
2     Bachelors 20131   1
2     Bachelors 20141   1
3     Bachelors 20106   1
3     Bachelors 20111   1       20116
3     Masters   20116   1
3     Masters   20121   1
3     Masters   20131   1       20136

我想把它变成这样（当运行为 20151 学期时）：

ID    DEGREE    TERM    STATUS
1     Bachelors 20111   1
1     Bachelors 20116   1
1     Bachelors 20121   0
1     Bachelors 20126   0
1     Bachelors 20131   0
1     Bachelors 20136   0
1     Bachelors 20141   0
1     Bachelors 20146   0
1     Bachelors 20151   0
2     Bachelors 20126   1
2     Bachelors 20131   1
2     Bachelors 20136   0
2     Bachelors 20141   1
2     Bachelors 20146   0
2     Bachelors 20151   0
3     Bachelors 20106   1
3     Bachelors 20111   1
3     Bachelors 20116   2
3     Bachelors 20121   2
3     Bachelors 20126   2
3     Bachelors 20131   2
3     Bachelors 20136   2
3     Bachelors 20141   2
3     Bachelors 20146   2
3     Bachelors 20151   2
3     Masters   20116   1
3     Masters   20121   1
3     Masters   20126   0
3     Masters   20131   1
3     Masters   20136   2
3     Masters   20141   2
3     Masters   20146   2
3     Masters   20151   2

在每个 table 中，状态为 0 - 未注册、1 - 已注册和 2 - 已毕业。 TERM 字段是年份，后跟 1 或 6 表示 spring 或秋季。

应该为每个人在他们的第一条记录和当前学期（在本例中为 20151）之间添加缺失的 TERM 记录。对于每条添加的记录，分配 STATUS 为 0，除非最后一个现有记录的 STATUS 为 2（携带）。也就是说，一个人已注册 (STATUS=1) 或未注册（STATUS=0 或 2）。

我在 Python 中使用 pandas，但我是 Python 的新手。我一直在试图弄清楚 DataFrame 的索引是如何工作的，但在这一点上这完全是个谜。任何指导将不胜感激。

Answer 1

你可以这样做。

import pandas as pd
# python 3.4 used
import io

# just try to replicate your data. Use your own csv file instead
# =========================================================
csv = 'ID,DEGREE,TERM,STATUS,GRADTERM\n1,Bachelors,20111,1,\n1,Bachelors,20116,1,\n2,Bachelors,20126,1,\n2,Bachelors,20131,1,\n2,Bachelors,20141,1,\n3,Bachelors,20106,1,\n3,Bachelors,20111,1,20116.0\n3,Masters,20116,1,\n3,Masters,20121,1,\n3,Masters,20131,1,20136.0\n'

df = pd.read_csv(io.StringIO(csv)).set_index('ID')
print(df)

       DEGREE   TERM  STATUS  GRADTERM
ID                                    
1   Bachelors  20111       1       NaN
1   Bachelors  20116       1       NaN
2   Bachelors  20126       1       NaN
2   Bachelors  20131       1       NaN
2   Bachelors  20141       1       NaN
3   Bachelors  20106       1       NaN
3   Bachelors  20111       1     20116
3     Masters  20116       1       NaN
3     Masters  20121       1       NaN
3     Masters  20131       1     20136


# two helper functions
# =========================================================

def build_year_term_range(start_term, current_term):
    # assumes start_term current_term in format '20151' alike
    start_year = int(start_term[:4])  # first four are year
    start_term = int(start_term[-1])  # last four is term
    current_year = int(current_term[:4])
    current_term = int(current_term[-1])
    # build a range
    year_rng = np.repeat(np.arange(start_year, current_year+1), 2)
    term_rng = [1, 6] * int(len(year_rng) / 2)
    year_term_rng = [int(str(year) + str(term)) for year, term in zip(year_rng, term_rng)]
    # check whether need to trim the first and last
    if start_term == 6:  # remove the first
        year_term_rng = year_term_rng[1:]
    if current_term == 1:  # remove the last
        year_term_rng = year_term_rng[:-1]

    return year_term_rng

def my_apply_func(group, current_year_term=current_year_term):
    # start of the record 
    start_year_term = str(group['TERM'].iloc[0])  # gives 2001
    year_term_rng = build_year_term_range(start_year_term, current_year_term)
    # manipulate the group
    group = group.reset_index().set_index('TERM')
    # use reindex to populate missing rows
    group = group.reindex(year_term_rng)
    # fillna ID/DEGREE same as previous
    group[['ID', 'DEGREE']] = group[['ID', 'DEGREE']].fillna(method='ffill')  
    # fillna by 0 not enrolled (for now)
    group['STATUS'] = group['STATUS'].fillna(0)
    # shift GRADTERM 1 slot forward, because GRADTERM and TERM are not aligned
    group['GRADTERM'] = group['GRADTERM'].shift(1)
    # check whether has been graduate, convert to int, use cumsum to carry that non-zero entry forward, convert back to boolean
    # might seems non-trivial at first place :)
    group.loc[group['GRADTERM'].notnull().astype(int).cumsum().astype(bool), 'STATUS'] = 2
    # return only relevant columns
    return group['STATUS']


# start processing
# ============================================================
# move ID from index to a normal column
df = df.reset_index()
# please specify the current year term in string
current_year_term = '20151'
# assume ID is your index column
result = df.groupby(['ID', 'DEGREE']).apply(my_apply_func).reset_index()

Out[163]: 
    ID     DEGREE   TERM  STATUS
0    1  Bachelors  20111       1
1    1  Bachelors  20116       1
2    1  Bachelors  20121       0
3    1  Bachelors  20126       0
4    1  Bachelors  20131       0
5    1  Bachelors  20136       0
6    1  Bachelors  20141       0
7    1  Bachelors  20146       0
8    1  Bachelors  20151       0
9    2  Bachelors  20126       1
10   2  Bachelors  20131       1
11   2  Bachelors  20136       0
12   2  Bachelors  20141       1
13   2  Bachelors  20146       0
14   2  Bachelors  20151       0
15   3  Bachelors  20106       1
16   3  Bachelors  20111       1
17   3  Bachelors  20116       2
18   3  Bachelors  20121       2
19   3  Bachelors  20126       2
20   3  Bachelors  20131       2
21   3  Bachelors  20136       2
22   3  Bachelors  20141       2
23   3  Bachelors  20146       2
24   3  Bachelors  20151       2
25   3    Masters  20116       1
26   3    Masters  20121       1
27   3    Masters  20126       0
28   3    Masters  20131       1
29   3    Masters  20136       2
30   3    Masters  20141       2
31   3    Masters  20146       2
32   3    Masters  20151       2

使用熊猫添加缺失的行

Adding missing rows using panda

python

append

pandas