[Numpy-discussion] How do I make a diagonal matrix?

Travis Oliphant oliphant at ee.byu.edu
Fri Jun 23 16:01:09 CDT 2006


Alan G Isaac wrote:

>>Alan G Isaac wrote: 
>>    
>>
>>>Why is a.flat not the same as a.A.flat? 
>>>      
>>>
>
>
>On Fri, 23 Jun 2006, Travis Oliphant apparently wrote: 
>  
>
>>It is the same object except for the pointer to the 
>>underlying array.  When asarray(a.flat) get's called it 
>>looks to the underlying array to get the sub-class and 
>>constructs that sub-class (and matrices can never be 1-d).  
>>Thus, it's a "feature" 
>>    
>>
>
>
>I doubt I will prove the only one to stumble over this.
>
>I can roughly understand why a.ravel() returns a matrix;
>but is there a good reason to forbid truly flattening the matrix?
>  
>
Because matrices are never 1-d.  This is actually pretty consistent 
behavior.

>My instincts are that a flatiter object should not have this 
>hidden "feature": flatiter objects should produce 
>a consistent behavior in all settings, regardless of the 
>underlying array. Anything else will prove too surprising.
>  
>
I think you are right that this is a bug, though.  Because __array__() 
(which is where the behavior comes from) should return a base-class 
array (not a sub-class).

-Travis





More information about the Numpy-discussion mailing list