Question

I have a df:

df:
               a      b       c
date
2012-01-01   0.50    1.2    0.70 
2012-01-01   0.45    1.2    0.65
2012-01-01   0.65    1.2    0.63
2012-01-01   0.75    1.2    0.29
2012-01-01   -0.25   1.2    -0.68

I want to calculate: a/b – c

I run:

new = df['a']/df['b'] - df['c']

This returns
new:

date
2012-01-01   -0.2833 
2012-01-01   -0.2750
2012-01-01   -0.0883
2012-01-01   0.3350
2012-01-01   0.4717

First error:
If I go:

new.ix[0][0]
TypeError: 'NumericType' object is unsubscriptable

So I assume it’s a series.

So I change it to:

new = pd.DataFrame(new)
type(new.ix[0][0])

<type 'NumericType'>

This is the weird part:

new.ix[0][0]
-0.2833
new.ix[0][0]/2
-0.141650
new.ix[0][0]/2.0
0

What is going on with the number type? How can I change it to float? What is the best practice here?

Thank you.

Answer 1

The reason for the differences in behaviour is when new is a Series vs. a DataFrame.

The first new you create is a Series:

In [11]: s = new = df['a'] / df['b'] - df['c']

In [12]: type(s)
Out[12]: pandas.core.series.Series

This has a value in position 0 of -0.2833:

In [13]: s.ix[0]
Out[13]: -0.28333333333333327

(You can’t __getitem__ on a float e.g. 2.0[0] gives you a similar error.)

However, when you then force it to a DataFrame:

In [14]: df_new = new = pd.DataFrame(s)

In [15]: df_new
Out[15]: 
          0
0 -0.283333
1 -0.275000
2 -0.088333
3  0.335000
4  0.471667

In [16]: type(df_new)
Out[16]: pandas.core.frame.DataFrame

this time the 0th index is a Series (row):

In [17]: df_new.ix[0]
Out[17]: 
0   -0.283333
Name: 0

In [18]: df_new.ix[0][0]
Out[18]: -0.28333333333333327

Update:

To ensure that all your DataFrames values are floats (numpy.float64) you can applymap (which works here since every column is being converted to a float):

df = df.applymap(float)

I don’t know how you got NumericType object, I assume the divide by 2 vs 2.0 derives from this.