Interacting with pandas is quite easy. In fact, with pandas being built upon NumPy, arrays can easily be extracted from DataFrame objects, and they can be transformed into DataFrames themselves.

First, let's upload some data into a DataFrame. The BostonHouse example we downloaded in the previous chapter from the ML repository is suitable:

In: import pandas as pd
    import numpy as np
    housing_filename = 'regression-datasets-housing.csv'
    housing = pd.read_csv(housing_filename, header=None)

As demonstrated in the Heterogeneous lists section, at this point, the .values method will extract an array of a type that accommodates all the different types that are present in the DataFrame:

In: housing_array = housing.values
    housing_array.dtype

Out: dtype('float64')  

In such a case, the selected type is float64 because the float type prevails over the int type:

In: housing.dtypes

Out:  0     float64
      1       int64
      2     float64
      3       int64
      4     float64
      5     float64
      6     float64
      7     float64
      8       int64
      9       int64
      10      int64
      11    float64
      12    float64
      13    float64
      dtype: object

Asking for the types used by the DataFrame object before extracting your NumPy array by using the .dtypes method on the DataFrame allows you to anticipate the dtype of the resulting array. Consequently, it allows you to decide whether to transform or change the type of the variables in the DataFrame object before proceeding (please consult the Working with categorical and textual data section of this chapter).