Using Pandas to_numeric() in Azure Machine Learning Studio

Question

I am facing an issue that Azure Machine Learning Studio fails to find the to_numeric method in pandas.

After reading a .csv in AMLS I try to process it in a python script. The line that is throwing me an error is:

dataframe1['Monthly Debt'] = pd.to_numeric(dataframe1['Monthly Debt'])

pd of course is pandas, dataframe1 is my working dataframe. The error thrown is:

AttributeError: 'module' object has no attribute 'to_numeric'

Of course everything works on my local python. Do you have any idea what AMLS might be talking about?

Answer 1

What’s new in Pandas v0.17.0

DataFrame.convert_objects has been deprecated in favor of type-specific functions pd.to_datetime, pd.to_timestamp and pd.to_numeric (new in 0.17.0) (GH11133).

So for Pandas versions < 0.17.0 you can and should use: df.convert_objects(convert_numeric=True)

Demo:

In [213]: x = pd.DataFrame({'a':['11', 'aaa', '0', np.nan, '123']})

In [214]: x
Out[214]:
     a
0   11
1  aaa
2    0
3  NaN
4  123

In [215]: x.dtypes
Out[215]:
a    object
dtype: object

In [216]: x = x.convert_objects(convert_numeric=True)

In [217]: x
Out[217]:
       a
0   11.0
1    NaN
2    0.0
3    NaN
4  123.0

In [218]: x.dtypes
Out[218]:
a    float64
dtype: object

Answer 2

OK this is an issue with Azure Machine Learning Studio. I just confirmed this with one of their data scientists.

I was using the Anaconda 2.0/Python 2.7.7 python version. Here, for some unknown reason the error will appear. If you just use Anaconda 4.0/Python 2.7.11 it will work as intended.