删除多指标为nan（Pandas）的行

Question

I have the following table with a multindex

                       value
userid     date

NaN        2014-06-12   42799
           2014-06-13   47673
           2014-06-14   47042
           2014-06-15   48079
           2014-06-16   44873
           2014-06-17   46586
           2014-06-18   44575
1000000021 2014-06-17   0
1000000024 2014-06-22   20
1000000043 2014-06-12   14
           2014-06-14   22
          .
          .
          .
          .

I would like to drop the row where the userid is Nan. If I wanted to drop another row I could do

data = data.drop(1000000021)

but

data = data.drop('NaN')
data = data.drop(np.nan)

and other attempts all return errors of differing varieties. Is there a way to drop the row without having to reindex?

Answer 1

You could identify the rows whose index is NaN using df.index.labels[0] == -1 , and select the other rows using df.loc :

In [48]: df.loc[~(df.index.labels[0] == -1)]
Out[48]: 
                       value
userid     date             
1000000021 2014-06-17      0
1000000024 2014-06-22     20
1000000043 2014-06-12     14
           2014-06-14     22

When using a boolean index, df[...] , df.loc[...] , and df.iloc[...] all behave the same. df[...] is commonly used to select columns, however, so you might want to avoid using df[...] for also selecting rows, as done above. That leaves df.loc and df.iloc as viable choices. Since df.iloc was created mainly for selecting by integer index, you might want to use df.loc[...] for selecting by label and by boolean mask. But this is just my convention -- Pandas allows all three.

Answer 2

Easier to reset and drop from the frame, then set the index.

In [3]: df =  DataFrame(np.random.randint(0,10,size=16).reshape(-1,1),columns=['value'],index=pd.MultiIndex.from_product([[np.nan,1,2,3],pd.date_range('20130101',periods=4)],names=['first','second']))

In [4]: df
Out[4]: 
                  value
first second           
NaN   2013-01-01      0
      2013-01-02      2
      2013-01-03      9
      2013-01-04      3
1     2013-01-01      8
      2013-01-02      8
      2013-01-03      5
      2013-01-04      3
2     2013-01-01      4
      2013-01-02      1
      2013-01-03      2
      2013-01-04      7
3     2013-01-01      3
      2013-01-02      9
      2013-01-03      3
      2013-01-04      4

In [5]: df.reset_index().dropna(subset=['first']).set_index(['first','second'])
Out[5]: 
                  value
first second           
1     2013-01-01      8
      2013-01-02      8
      2013-01-03      5
      2013-01-04      3
2     2013-01-01      4
      2013-01-02      1
      2013-01-03      2
      2013-01-04      7
3     2013-01-01      3
      2013-01-02      9
      2013-01-03      3
      2013-01-04      4