扩展数据框中的每一行

Question

Consider this simple example

data = pd.DataFrame({'mydate' : [pd.to_datetime('2016-06-06'),
                                 pd.to_datetime('2016-06-02')],
                     'value' : [1, 2]})

data.set_index('mydate', inplace = True)

data
Out[260]: 
            value
mydate           
2016-06-06      1
2016-06-02      2

I want to iterate over each row so that the dataframe gets "enlarged" by a couple days (2 days before, 2 days after) around each index value (which is a date) for the current row.

For instance, if you consider the first row, I want to tell Pandas to add 4 more rows, corresponding to the days 2016-06-04 , 2016-06-05 , 2016-06-07 and 2016-06-07 . The value for these extra rows can just be whathever is in value for that row (in this case: 1). This logic is applied every row and the final dataframe is the concatenation of all these enlarged dataframes.

I have tried the following function in an apply(., axis = 1) :

def expand_onerow(df, ndaysback = 2, nhdaysfwd = 2):

    new_index = pd.date_range(pd.to_datetime(df.name) - pd.Timedelta(days=ndaysback), 
                              pd.to_datetime(df.name) + pd.Timedelta(days=nhdaysfwd), 
                              freq='D')

    newdf = df.reindex(index=new_index, method='nearest')     #New df with expanded index
    return newdf

But unfortunately I running data.apply(lambda x: expand_onerow(x), axis = 1) gives:

  File "pandas/_libs/tslib.pyx", line 1165, in pandas._libs.tslib._Timestamp.__richcmp__

TypeError: ("Cannot compare type 'Timestamp' with type 'str'", 'occurred at index 2016-06-06 00:00:00')

Another approach I tried is the following: I first reset the index,

data.reset_index(inplace = True)
data
Out[339]: 
      mydate  value
0 2016-06-06      1
1 2016-06-02      2

Then I use a slight modification of my function

def expand_onerow_alt(df, ndaysback = 2, nhdaysfwd = 2):

    new_index = pd.date_range(pd.to_datetime(df.mydate) - pd.Timedelta(days=ndaysback), 
                              pd.to_datetime(df.mydate) + pd.Timedelta(days=nhdaysfwd), 
                              freq='D')
    newdf = pd.Series(df).reindex(index = new_index).T  #New df with expanded index
    return newdf

which gives

data.apply(lambda x: expand_onerow_alt(x), axis = 1)
Out[338]: 
   2016-05-31  2016-06-01  2016-06-02  2016-06-03  2016-06-04  2016-06-05  2016-06-06  2016-06-07  2016-06-08
0         nan         nan         nan         nan         nan         nan         nan         nan         nan
1         nan         nan         nan         nan         nan         nan         nan         nan         nan

closer but not there yet...

I do not understand what is wrong here. What am I missing? I am looking for the most Pandonic approach here.

Thanks!

Answer 1

I modify little bit of your function

def expand_onerow(df, ndaysback = 2, nhdaysfwd = 2):

    new_index = pd.date_range(pd.to_datetime(df.index[0]) - pd.Timedelta(days=ndaysback),
                              pd.to_datetime(df.index[0]) + pd.Timedelta(days=nhdaysfwd),
                              freq='D')

    newdf = df.reindex(index=new_index, method='nearest')     #New df with expanded index
    return newdf

pd.concat([expand_onerow(data.loc[[x],:], ndaysback = 2, nhdaysfwd = 2) for x ,_ in data.iterrows()])


Out[455]: 
            value
2016-05-31      2
2016-06-01      2
2016-06-02      2
2016-06-03      2
2016-06-04      2
2016-06-04      1
2016-06-05      1
2016-06-06      1
2016-06-07      1
2016-06-08      1

More info

Basically that one line equal to

l=[]
for x ,_ in data.iterrows():

    l.append(expand_onerow(data.loc[[x],:], ndaysback = 2, nhdaysfwd = 2))# query out each row by using their index(x is the index for each row) and append then into a empty list


pd.concat(l)# concat the list to one df at the end