[Numpy-discussion] where

[Travis Oliphant]

Ryan Krauss wrote:

>Does where return a mask?
>  
>
Only in the second use case...

>If I do
>myvect=where((f > 19.5) & (phase > 0),f,phase)
>myvect is the same length as f and phase and there is some
>modification of the values where the condition is met, but what that
>modification is is unclear to me.
>  
>

The behavior of

where(condition, for_true, for_false)

is to return an array of the same shape as condition with elements of 
for_true where condition is true and
for_false where condition is false.

Thus myvect will contain elements of f where the condition is met and 
elements of phase otherwise.

>If I do
>myind=where((f > 19.5) & (phase > 0))
>I seem to get the indices of the points where both conditions are met.
>  
>
Yes.  That is correct.   It is a different use-case... Note, however, 
that in the current SVN version of NumPy, this use-case will always 
return a tuple of indices (use the nonzero function instead for behavior 
that will stay constant).  For your 1-d example (I'm guessing it's 1-d)  
where will return a length-1 tuple.

>I am using version 0.9.5.2043.  I see those kinds of errors about
>truth testing an array often, but not in this case.
>  
>
That is strange.   What are the sizes of f and phase?

-Travis