float() argument must be a string or a number, not 'Timestamp'

Question

I can't make scilearn work with a datetime series.

found this post but did not help me = Pandas : TypeError: float() argument must be a string or a number

the csv file has 2 date columns with a date, dates are in the following format: 2017-07-21 06:19:53 (string)

i converted the string to an datetime64[ns], so the date became a long value and i could do calculations on it. scilearn refuses this type and gives the error float() argument must be a string or a number, not 'Timestamp'

also tried with pandas.to_datetime() no luck.

the model i use in scilearn is the KMeans clustering model. when printing the dtypes this is the result:

ip                      int64
date           datetime64[ns]
succesFlag              int64
app                     int64
enddate        datetime64[ns]
user_userid             int64
dtype: object

Here is my code:

def getDataframe():
    df = pd.read_csv(filename)
    df['date']=df['date'].astype('datetime64[ns]',inplace=True)
    df['enddate']=df['enddate'].astype('datetime64[ns]',inplace=True)
    df['app']=df['app'].replace({
            "Azure": 0 ,
            "Peoplesoft":1,
            "Office":2 ,
            "DevOps":3 ,
            "Optima":4 ,
            "Ada-Tech": 5 
         },inplace=True)    
    df['ip']=df['ip'].apply(lambda x: int(ip4.ip_address(x))).to_frame('ip')
    print(df.dtypes)
    return df

the expectation was that KMeans clustering model would work with numerical values as i converted them but it did not.

what did i wrong ?

jezrael · Accepted Answer · 2018-12-22T15:27:26.170

I suggest change your solution - a but simplify also:

add parameter parse_dates for converting columns to datetimes and then to numeric unix datetimes
for converting remove inplace=True or use faster map - it also create NaNs for non matched values, so output is numeric too

def getDataframe():
    df = pd.read_csv(filename, parse_dates=['date','enddate'])
    df[['date','enddate']] = df[['date','enddate']].astype(np.int64) // 10**9

    df['app']=df['app'].map({
            "Azure": 0 ,
            "Peoplesoft":1,
            "Office":2 ,
            "DevOps":3 ,
            "Optima":4 ,
            "Ada-Tech": 5 
         })    
    df['ip']=df['ip'].apply(lambda x: int(ip4.ip_address(x))).to_frame('ip')
    print(df.dtypes)
    return df

float() argument must be a string or a number, not 'Timestamp'

1 Answers1

Linked