How to convert datatype:object to float64 in python?

Question

I am going around in circles and tried so many different ways so I guess my core understanding is wrong. I would be grateful for help in understanding my encoding/decoding issues.

I import the dataframe from SQL and it seems that some datatypes:float64 are converted to Object. Thus, I cannot do any calculation. I fail to convert the Object back to float64.

df.head()

Date        WD  Manpower 2nd     CTR    2ndU    T1  　　T2    　　T3    　　T4 

2013/4/6    6   NaN     2,645   5.27%   0.29    407     533     454     368
2013/4/7    7   NaN     2,118   5.89%   0.31    257     659     583     369
2013/4/13   6   NaN     2,470   5.38%   0.29    354     531     473 　　383
2013/4/14   7   NaN     2,033   6.77%   0.37    396     748     681     458
2013/4/20   6   NaN     2,690   5.38%   0.29    361     528     541     381

df.dtypes

WD             float64
Manpower       float64
2nd             object
CTR             object
2ndU           float64
T1              object
T2              object
T3              object
T4              object
T5              object

dtype: object

SQL table:

Answer 1

You can convert most of the columns by just calling convert_objects:

In [36]:

df = df.convert_objects(convert_numeric=True)
df.dtypes
Out[36]:
Date         object
WD            int64
Manpower    float64
2nd          object
CTR          object
2ndU        float64
T1            int64
T2          int64
T3           int64
T4        float64
dtype: object

For column '2nd' and 'CTR' we can call the vectorised str methods to replace the thousands separator and remove the '%' sign and then astype to convert:

In [39]:

df['2nd'] = df['2nd'].str.replace(',','').astype(int)
df['CTR'] = df['CTR'].str.replace('%','').astype(np.float64)
df.dtypes
Out[39]:
Date         object
WD            int64
Manpower    float64
2nd           int32
CTR         float64
2ndU        float64
T1            int64
T2            int64
T3            int64
T4           object
dtype: object
In [40]:

df.head()
Out[40]:
        Date  WD  Manpower   2nd   CTR  2ndU   T1  　　T2   T3     T4
0   2013/4/6   6       NaN  2645  5.27  0.29  407   533  454    368
1   2013/4/7   7       NaN  2118  5.89  0.31  257   659  583    369
2  2013/4/13   6       NaN  2470  5.38  0.29  354   531  473  　　383
3  2013/4/14   7       NaN  2033  6.77  0.37  396   748  681    458
4  2013/4/20   6       NaN  2690  5.38  0.29  361   528  541    381

Or you can do the string handling operations above without the call to astype and then call convert_objects to convert everything in one go.

UPDATE

Since version 0.17.0 convert_objects is deprecated and there isn't a top-level function to do this so you need to do:

df.apply(lambda col:pd.to_numeric(col, errors='coerce'))

See the docs and this related question: pandas: to_numeric for multiple columns

Answer 2

convert_objects is deprecated.

For pandas >= 0.17.0, use pd.to_numeric

df["2nd"] = pd.to_numeric(df["2nd"])

Answer 3

I had this problem in a DataFrame (df) created from an Excel-sheet with several internal header rows.

After cleaning out the internal header rows from df, the columns' values were of "non-null object" type (DataFrame.info()).

This code converted all numerical values of multiple columns to int64 and float64 in one go:

for i in range(0, len(df.columns)):
    df.iloc[:,i] = pd.to_numeric(df.iloc[:,i], errors='ignore')
    # errors='ignore' lets strings remain as 'non-null objects'

Answer 4

You can try this:

df['2nd'] = pd.to_numeric(df['2nd'].str.replace(',', ''))
df['CTR'] = pd.to_numeric(df['CTR'].str.replace('%', ''))

Answer 5

Or you can use regular expression to handle multiple items as the general case of this issue,

df['2nd'] = pd.to_numeric(df['2nd'].str.replace(r'[,.%]','')) 
df['CTR'] = pd.to_numeric(df['CTR'].str.replace(r'[^\d%]',''))

Answer 6

X = np.array(X, dtype=float)

You can use this to convert to array of float in python 3.7.6