[Tutor] From Numpy Import *

[Michael H. Goldwasser]

On Wednesday November 7, 2007, Dinesh B Vadhia wrote: 

>    Hello!  The standard Python practice for importing modules is, for example:
>    
>    import sys
>    import os
>    etc.
>    
>    In NumPy (and SciPy) the 'book' suggests using:
>    
>    from numpy import *
>    from scipy import *
>    
>    However, when I instead use 'import numpy' it causes all sorts of errors in my existing code.

The issue is the following.  The numpy module includes many definitions, for
example a class named array.   When you use the syntax,

   from numpy import *

That takes all definitions from the module and places them into your
current namespace.  At this point, it would be fine to use a command
such as 

  values = array([1.0, 2.0, 3.0])

which instantiates a (numpy) array.


If you instead use the syntax

   import numpy

things brings that module as a whole into your namespace, but to
access definitions from that module you have to give a qualified
name, for example as

  values = numpy.array([1.0, 2.0, 3.0])

You cannot simply use the word array as in the first scenario.  This
would explain why your existing code would no longer work with the
change.

>    What do you suggest?

The advantage of the "from numpy import *" syntax is mostly
convenience.   However, the better style is "import numpy" precisely
becuase it does not automatically introduce many other definitions
into your current namespace.

If you were using some other package that also defined an "array" and
then you were to use the "from numpy import *", the new definition
would override the other definition.  The use of qualified names helps
to avoid these collisions and makes clear where those definitions are
coming from.

With regard,
Michael