[Numpy-discussion] question about standalone small software and teaching

Gael Varoquaux gael.varoquaux at normalesup.org
Thu Apr 5 13:39:18 EDT 2007


On Thu, Apr 05, 2007 at 01:36:28PM -0400, Alan G Isaac wrote:
> On Thu, 05 Apr 2007, Christopher Barker apparently wrote: 
> > I think, in a situation when you are intending the class 
> > attribute to be over-ridden by an instance attribute, then 
> > it's clearer to define it as an instance attribute in the 
> > first place: 

> This is true of course.
> BUT I am with Gael on this one:
> when **introducing** newbies to their first class(es),
> it can be useful not to introduce with the __init__ function.

> Remember, for many perfectly intelligent people,
> the phrases 'class object' and 'instance object'
> are no more than 'n o i s e  object'.

Actually I do it the other way around nowadays. I don't talk about class
object, instance object... I simply tell them to create attributes by
putting them in the __init__ . That way I simply avoid the explaination
until they actually need class attributes. By the time they need these
they are flying with their own wings.

When I would do it the other way around they would trip on the issue,
eventually, and than I would have to give the explaination, and this has
always been a difficult moment.

Gaël



More information about the NumPy-Discussion mailing list