Pandas read_csv（）：将0保留为0（不转换为NaN）

Question

我试图读取一个csv文件，其中有一个示例：

datetime,check,lat,lon,co_alpha,atn,status,bc
2012-10-27 15:00:59,2,0,0,2.427,,,
2012-10-27 15:01:00,2,0,0,2.407,,,
2012-10-27 15:02:49,2,0,0,2.207,-17.358,0,-16162
2012-10-27 15:02:50,2,0,0,2.207,-17.354,0,8192
2012-10-27 15:02:51,1,0,0,2.207,-17.358,0,-8152
2012-10-27 15:02:52,1,0,0,2.207,-17.358,0,648
2012-10-27 15:06:03,0,51.195076,4.444407,2.349,-17.289,0,4909
2012-10-27 15:06:04,0,51.195182,4.44427,2.344,-17.289,0,587
2012-12-05 09:21:34,,,,,42.960,1,16430
2012-12-05 09:21:35,,,,,42.962,1,3597

我遇到的问题是，在只有int的列中，0被转换为NaN（例如列'check'和'status'，这些列只包含int，但该列被读取为float，因为有实际的缺失值）。但我只希望将空值转换为NaN，而不是0。在

我得到的是：

^{pr2}$

因此，在'check'和'status'列中，有许多NaN's。在'lat'和'lon'列中，0不被转换成NaN's

• 使用na_values=''和{}没有帮助。有没有办法指定不将int 0转换为NaN？或者这是个虫子？

• 我可以用dtype关键字将特定列的数据类型指定为int。这将0保持为0，但问题是这些列也包含真正的NaN（空值）。所以，在这个例子中，这些值也被转换成0，就像在int列中不能有NaN一样。

---

编辑：升级到pandas 0.10.1后，即使不指定keep_default_na和na_values，它也能正常工作：

>>> pd.read_clipboard(sep=',', parse_dates=True, index_col=0)
                     check        lat       lon  co_alpha     atn  status     bc
datetime                                                                        
2012-10-27 15:00:59      2   0.000000  0.000000     2.427     NaN     NaN    NaN
2012-10-27 15:01:00      2   0.000000  0.000000     2.407     NaN     NaN    NaN
2012-10-27 15:02:49      2   0.000000  0.000000     2.207 -17.358       0 -16162
2012-10-27 15:02:50      2   0.000000  0.000000     2.207 -17.354       0   8192
2012-10-27 15:02:51      1   0.000000  0.000000     2.207 -17.358       0  -8152
2012-10-27 15:02:52      1   0.000000  0.000000     2.207 -17.358       0    648
2012-10-27 15:06:03      0  51.195076  4.444407     2.349 -17.289       0   4909
2012-10-27 15:06:04      0  51.195182  4.444270     2.344 -17.289       0    587
2012-12-05 09:21:34    NaN        NaN       NaN       NaN  42.960       1  16430
2012-12-05 09:21:35    NaN        NaN       NaN       NaN  42.962       1   3597

Answer 1

您必须首先将keep_default_na设置为False：

df = pd.read_clipboard(sep=',', index_col=0, keep_default_na=False, na_values='')

In [2]: df
Out[2]: 
                     check        lat       lon  co_alpha     atn  status     bc
datetime                                                                        
2012-10-27 15:00:59      2   0.000000  0.000000     2.427     NaN     NaN    NaN
2012-10-27 15:01:00      2   0.000000  0.000000     2.407     NaN     NaN    NaN
2012-10-27 15:02:49      2   0.000000  0.000000     2.207 -17.358       0 -16162
2012-10-27 15:02:50      2   0.000000  0.000000     2.207 -17.354       0   8192
2012-10-27 15:02:51      1   0.000000  0.000000     2.207 -17.358       0  -8152
2012-10-27 15:02:52      1   0.000000  0.000000     2.207 -17.358       0    648
2012-10-27 15:06:03      0  51.195076  4.444407     2.349 -17.289       0   4909
2012-10-27 15:06:04      0  51.195182  4.444270     2.344 -17.289       0    587
2012-12-05 09:21:34    NaN        NaN       NaN       NaN  42.960       1  16430
2012-12-05 09:21:35    NaN        NaN       NaN       NaN  42.962       1   3597

从^{}的文档字符串：

keep_default_na : bool, default True
     If na_values are specified and keep_default_na is False the default NaN
    values are overridden, otherwise they're appended to

na_values : list-like or dict, default None
Additional strings to recognize as NA/NaN. If dict passed, specific per-column NA values