python - 为什么 pandas 数据帧将整数转换为 float 据类型

Question

我有一个 csv 文件

Wed Dec 04 11:30:04 GMT+05:30 2019,20,35.0,143455434,0
Wed Dec 04 11:30:13 GMT+05:30 2019,40,25.5,null,

我想将其加载到 pandas 中并将各个列转换为我各自的数据类型。我就是这样做的

raw_df = pd.read_csv('raw.csv', dtype=str)
raw_df = raw_df.replace({'null':None, pd.np.nan: None})

这是我要转换的函数

def df_function(row):
    row['timestamp'] = parse(row['timestamp'])
    row['odometer'] = float(row['odometer']) + 1
    row['speed'] = float(row['speed'])

    if row['id'] is not None:
        row['id'] = str(row['id'])

    if row['error_code'] is not None:
        row['error_code'] = int(row['error_code'])

    return row

raw_df = raw_df.apply(df_function, axis=1)

当您打印列的数据类型时，您会发现

timestamp     datetime64[ns, tzoffset(None, -19800)]
odometer                                     float64
speed                                        float64
id                                            object
error_code                                   float64
dtype: object

error_code 是 float64，虽然它应该是 int64，这里有什么问题

Answer 1

正如 pandas documents 中提到的

The Integer NA support currently uses the capitalized dtype version, e.g. Int8 as compared to the traditional int8. This may be changed at a future date

您需要将列更改为 Int8

df = pd.DataFrame({"error_code":[1,2,5,np.nan]}) 
print(df.dtypes)

# error_code    float64
# dtype: object

df["error_code"] = df["error_code"].astype("Int8") 
print(df.dtypes)

输出:

error_code    Int8
dtype: object