Question

我正在尝试使用科学记数法中包含数字的列来读取.csv。不管我做什么，最终都会把它们读成字符串：

def readData(path, cols):
    types  = [str, str, str, str, np.float32]
    t_dict = {key: value for (key, value) in zip(c, types)}

    df = pd.read_csv(path, header=0, sep=';', encoding='latin1', usecols=cols, dtype=t_dict, chunksize=5000)

    return df

c = [3, 6, 7, 9, 16]
df2017_chunks = readData('Data/2017.csv', c)

def preProcess(df, f):    
    df.columns = f
    df['id_client'] = df['id_client'].apply(lambda x: str(int(float(x))))

    return df

f = ['issue_date', 'channel', 'product', 'issue', 'id_client']

df = pd.DataFrame(columns=f)
for chunk in df2017_chunks:
    aux = preProcess(chunk, f)
    df = pd.concat([df, aux])

如何正确阅读此数据？

Answer 1

您的预处理函数在应用其他函数后应用字符串转换。这是预期的行为吗？

你可以尝试一下：

df = pd.read_csv(path, header=0, sep=';', encoding='latin1', usecols=cols, chunksize=5000)
df["id_client"] = pd.to_numeric(df["id_client"])

Answer 2

示例数据框：

df = pd.DataFrame({'issue_date': [1920,1921,1922,1923,1924,1925,1926],
    'name': ['jon doe1','jon doe2','jon doe3','jon doe4','jon doe5','jon doe6','jon doe7'],
    'id_cleint': ['18.61', '17.60', '18.27', '16.18', '16.81', '16.37', '67.07']})

您可以使用以下命令检查数据框的类型

print df.dtypes

输出：

id_client     object
issue_date     int64
name          object
dtype: object

使用以下命令将df['id_client'] dtype从object转换为float64：

df['id_client'] =  pd.to_numeric(df['id_client'], errors='coerce')

无法转换项目时，

errors='coerce'将导致NaN。使用命令
print df.dtypes会产生以下输出：

id_client     float64
issue_date      int64
name           object
dtype: object

科学记数法在熊猫中被读作字符串

2 个答案: