[Numpy-discussion] [numpy] argmin in multidimensional arrays

[Glen W. Mabey]

On Sat, Mar 03, 2007 at 03:09:35PM -0600, Robert Kern wrote:
> Glen W. Mabey wrote:
> > Does anyone else find this behavior counter-intuitive?  
> > 
> > It seems to me that one of the great design features of numpy is the
> > n-dim generality it provides, and argmin is one function in this breaks
> > down, IMHO.
> 
> Not at all. It consistently applies the simple rule: if the method operates over
> an axis, it takes an axis= keyword argument. The default for the axis= argument
> is None, which means that it operates over the flattened array. Other axes need
> to be specified explicitly.
> 
> See .sum(), .mean(), .var(), .repeat(), .min(), etc.

Okay, I see the reasoning.  I am glad that functions like sum, max, and
min return a single value, unless axis is specified.

However, it seems to me that the arg* functions are inherently
different.  After all, it is an index that is sought instead of a value, 
and when this result cannot be directly applied to the original array, 
then at first glance it appears to be ... less than intuitive.

Thanks for your reply.

Glen