Thread Subject: Interesting INTERP1 behavior

Hi all,

I found some interesting behavior in INTERP1:

%--------------------
 >> interp1([-inf 10 20 inf],[30 30 40 40],0:5:30)

ans =

    NaN NaN 30 35 40 40 40
%--------------------

Ideally, I would want this to produce [30 30 30 35 40 40 40]. However,
I'd probably be okay if it produced an error along the lines of "inputs
must be finite".

I reported this as a bug (albeit minor) to The Mathworks, but thought
I'd share here.

Jordan

>From the documentation for interp1

'For the 'nearest', 'linear', and 'v5cubic' methods,
interp1(x,Y,xi,method) returns NaN for any element of xi that is
outside the interval spanned by x. For all other methods, interp1
performs extrapolation for out of range values.'

The default method is linear, so the behavior seems to be according to
spec

Bob,

Actually, no, it doesn't.

If it really worked that way, then the NaNs would have appeared on
"both" sides of the interval I gave it. That is, it would have produced:

   [NaN NaN 30 35 40 NaN NaN]

instead of what I got which was:

   [NaN NaN 30 35 40 40 40]

Jordan

Bob Alvarez wrote:
>>From the documentation for interp1
>
> 'For the 'nearest', 'linear', and 'v5cubic' methods,
> interp1(x,Y,xi,method) returns NaN for any element of xi that is
> outside the interval spanned by x. For all other methods, interp1
> performs extrapolation for out of range values.'
>
> The default method is linear, so the behavior seems to be according to
> spec
>

Jordan Rosenthal wrote:
> Hi all,
>
> I found some interesting behavior in INTERP1:
>
> %--------------------
> >> interp1([-inf 10 20 inf],[30 30 40 40],0:5:30)
>
> ans =
>
> NaN NaN 30 35 40 40 40
> %--------------------
>
> Ideally, I would want this to produce [30 30 30 35 40 40 40]. However,
> I'd probably be okay if it produced an error along the lines of "inputs
> must be finite".
>
> I reported this as a bug (albeit minor) to The Mathworks, but thought
> I'd share here.
>
> Jordan

Indeed interesting. I tried and noticed that

>> interp1([-realmax 10 20 inf],[30 30 40 40],0:5:30)
ans =
    30 30 30 35 40 40 40

works fine. Let us know what TMW has to say.

Regards,
-rajeev-

[top posting repaired]

In article <MQZwe.98$75.91@llslave.llan.ll.mit.edu>,
 Jordan Rosenthal <jr@ll.mit.edu> wrote:

> Bob Alvarez wrote:
> >>From the documentation for interp1
> >
> > 'For the 'nearest', 'linear', and 'v5cubic' methods,
> > interp1(x,Y,xi,method) returns NaN for any element of xi that is
> > outside the interval spanned by x. For all other methods, interp1
> > performs extrapolation for out of range values.'
> >
> > The default method is linear, so the behavior seems to be according to
> > spec
> >
> Bob,
>
> Actually, no, it doesn't.
>
> If it really worked that way, then the NaNs would have appeared on
> "both" sides of the interval I gave it. That is, it would have produced:
>
> [NaN NaN 30 35 40 NaN NaN]
>
> instead of what I got which was:
>
> [NaN NaN 30 35 40 40 40]
>
> Jordan


Hi Jordan (we haven't communicated in quite a while)!

Actually, there are no points outside the x interval since the minimum x
is -inf and the maximum x is +inf and every number I can think of is
between those two extremes. :-)

I say that your example should work and indeed, when you replace inf
with realmax it does work -- at least in the ancient version 5.2 that I
have.

--
Doug Schwarz
dmschwarz&urgrad,rochester,edu
Make obvious changes to get real email address.

Hi Doug,

Yep, your right; I screwed up that last post.

The Mathworks did reply to my bug submission. They thanked me for my
suggestions to improve the INTERP1 function, and suggested a few
alternatives.

Jordan

PS I don't get to read the group as regularly as I used to. Now, I'm
mostly relegated to browsing every once and a while. Ah...I miss the
days of being a graduate student. :)

Doug Schwarz wrote:

> [top posting repaired]
>
> In article <MQZwe.98$75.91@llslave.llan.ll.mit.edu>,
> Jordan Rosenthal <jr@ll.mit.edu> wrote:
>
>
>>Bob Alvarez wrote:
>>
>>>>From the documentation for interp1
>>>
>>>'For the 'nearest', 'linear', and 'v5cubic' methods,
>>>interp1(x,Y,xi,method) returns NaN for any element of xi that is
>>>outside the interval spanned by x. For all other methods, interp1
>>>performs extrapolation for out of range values.'
>>>
>>>The default method is linear, so the behavior seems to be according to
>>>spec
>>>
>>
>>Bob,
>>
>>Actually, no, it doesn't.
>>
>>If it really worked that way, then the NaNs would have appeared on
>>"both" sides of the interval I gave it. That is, it would have produced:
>>
>> [NaN NaN 30 35 40 NaN NaN]
>>
>>instead of what I got which was:
>>
>> [NaN NaN 30 35 40 40 40]
>>
>>Jordan
>
>
>
> Hi Jordan (we haven't communicated in quite a while)!
>
> Actually, there are no points outside the x interval since the minimum x
> is -inf and the maximum x is +inf and every number I can think of is
> between those two extremes. :-)
>
> I say that your example should work and indeed, when you replace inf
> with realmax it does work -- at least in the ancient version 5.2 that I
> have.
>

In article <O9axe.99$75.84@llslave.llan.ll.mit.edu>,
 Jordan Rosenthal <jr@ll.mit.edu> wrote:

> Ah...I miss the days of being a graduate student. :)

But not the paycheck, I'll bet!

--
Doug Schwarz
dmschwarz&urgrad,rochester,edu
Make obvious changes to get real email address.