| Re: Negative Variance? [message #65767] |
Thu, 19 March 2009 20:11 |
Homeyer
Messages: 9
Registered: March 2009
|
Junior Member |
|
|
On Mar 19, 5:00 pm, "anniebry...@gmail.com" <anniebry...@gmail.com>
wrote:
> I am using the VARIANCE comand in IDL and am getting negative values.
> The description in the HELP menu for IDL doesn't provide much
> information, but from what I know, I thought variance was either an
> absolute value or a deviation from a squared number, neither of which
> could yield a negative value.
>
> I am looking at the variance for 2 different bands in an AVIRIS
> image. I use the same .pro, just change the band I am looking at.
> The results for Band 1 (lets call it) make sense: all positive
> values. For Band 2, I get a lot of negatives.?! How is this
> possible.
>
> These are the values for Band 1:
>
> integer mean std dev minimum maximum n_elements
> 2633.3 2059.3 0.0000 9367.0 (160686) =
> 160686
>
> These are the values for Band2:
>
> integer mean std dev minimum maximum n_elements
> 1167.5 18269. -32768. 32767. (160686) =
> 160686
I am not sure how you are doing your calculations.. but I think you
may have a conversion wrong in some process. If for some reason you
are converting a variance (should be a FLOAT or DOUBLE) to an INT,
which by the way maximizes at 32767 and minimizes at -32768, of course
you will have negative values. If you are converting to INTs, that is
your first problem. Variance should always be a FLOAT or DOUBLE.
Cheers,
Cameron Homeyer
|
|
|
|
| Re: Negative Variance? [message #65768 is a reply to message #65767] |
Thu, 19 March 2009 19:43  |
wlandsman
Messages: 743
Registered: June 2000
|
Senior Member |
|
|
On Mar 19, 6:00 pm, "anniebry...@gmail.com" <anniebry...@gmail.com>
wrote:
> I am using the VARIANCE comand in IDL and am getting negative values.
I would check that you not getting round-off problems by using double
precision or the /DOUBLE keyword to VARIANCE. As discussed in
http://en.wikipedia.org/wiki/Algorithms_for_calculating_vari ance
round-off error can accumulate in the two-pass variance calculation
(used by IDL) when most values are similar but some are wildly
different.
I had a similar headscratcher a few years back when I found that
SMOOTHing a non-negative array could yield negative numbers. This
was a roundoff problem that was fixed by using double precision.
--Wayne
|
|
|
|
| Re: Negative Variance? [message #65769 is a reply to message #65768] |
Thu, 19 March 2009 18:47 |
anniebryant@gmail.com
Messages: 16
Registered: March 2009
|
Junior Member |
|
|
On Mar 19, 4:35 pm, "Jean H." <jghas...@DELTHIS.ucalgary.ANDTHIS.ca>
wrote:
> anniebry...@gmail.com wrote:
>> On Mar 19, 4:08 pm, Paolo <pgri...@gmail.com> wrote:
>>> anniebry...@gmail.com wrote:
>>>> I am using the VARIANCE comand in IDL and am getting negative values.
>>>> The description in the HELP menu for IDL doesn't provide much
>>>> information, but from what I know, I thought variance was either an
>>>> absolute value or a deviation from a squared number, neither of which
>>>> could yield a negative value.
>>>> I am looking at the variance for 2 different bands in an AVIRIS
>>>> image. I use the same .pro, just change the band I am looking at.
>>>> The results for Band 1 (lets call it) make sense: all positive
>>>> values. For Band 2, I get a lot of negatives.?! How is this
>>>> possible.
>>>> These are the values for Band 1:
>>>> integer mean std dev minimum maximum n_elements
>>>> 2633.3 2059.3 0.0000 9367.0 (160686) =
>>>> 160686
>>>> These are the values for Band2:
>>>> integer mean std dev minimum maximum n_elements
>>>> 1167.5 18269. -32768. 32767. (160686) =
>>>> 160686
>>> I don't see any variance here. The only negative number is the
>>> minimum.
>
>>> Ciao,
>>> Paolo
>
>> Those are the values of the vector AFTER the variance was computed, it
>> should read Variance_Band1 and Variance_Band2.
>
> How can you get multiple variances for a band?
> Jean
I created a moving kernel inside which to calculate the variance. So
the question is. What is the variance within a 17pixel x 17pixel
moving window. That is what I am calculating the variance for.
|
|
|
|
| Re: Negative Variance? [message #65770 is a reply to message #65769] |
Thu, 19 March 2009 15:35 |
Jean H.
Messages: 472
Registered: July 2006
|
Senior Member |
|
|
anniebryant@gmail.com wrote:
> On Mar 19, 4:08 pm, Paolo <pgri...@gmail.com> wrote:
>> anniebry...@gmail.com wrote:
>>> I am using the VARIANCE comand in IDL and am getting negative values.
>>> The description in the HELP menu for IDL doesn't provide much
>>> information, but from what I know, I thought variance was either an
>>> absolute value or a deviation from a squared number, neither of which
>>> could yield a negative value.
>>> I am looking at the variance for 2 different bands in an AVIRIS
>>> image. I use the same .pro, just change the band I am looking at.
>>> The results for Band 1 (lets call it) make sense: all positive
>>> values. For Band 2, I get a lot of negatives.?! How is this
>>> possible.
>>> These are the values for Band 1:
>>> integer mean std dev minimum maximum n_elements
>>> 2633.3 2059.3 0.0000 9367.0 (160686) =
>>> 160686
>>> These are the values for Band2:
>>> integer mean std dev minimum maximum n_elements
>>> 1167.5 18269. -32768. 32767. (160686) =
>>> 160686
>> I don't see any variance here. The only negative number is the
>> minimum.
>>
>> Ciao,
>> Paolo
>
> Those are the values of the vector AFTER the variance was computed, it
> should read Variance_Band1 and Variance_Band2.
How can you get multiple variances for a band?
Jean
|
|
|
|
| Re: Negative Variance? [message #65771 is a reply to message #65770] |
Thu, 19 March 2009 15:10 |
anniebryant@gmail.com
Messages: 16
Registered: March 2009
|
Junior Member |
|
|
On Mar 19, 4:08 pm, Paolo <pgri...@gmail.com> wrote:
> anniebry...@gmail.com wrote:
>> I am using the VARIANCE comand in IDL and am getting negative values.
>> The description in the HELP menu for IDL doesn't provide much
>> information, but from what I know, I thought variance was either an
>> absolute value or a deviation from a squared number, neither of which
>> could yield a negative value.
>
>> I am looking at the variance for 2 different bands in an AVIRIS
>> image. I use the same .pro, just change the band I am looking at.
>> The results for Band 1 (lets call it) make sense: all positive
>> values. For Band 2, I get a lot of negatives.?! How is this
>> possible.
>
>> These are the values for Band 1:
>
>> integer mean std dev minimum maximum n_elements
>> 2633.3 2059.3 0.0000 9367.0 (160686) =
>> 160686
>
>> These are the values for Band2:
>
>> integer mean std dev minimum maximum n_elements
>> 1167.5 18269. -32768. 32767. (160686) =
>> 160686
>
> I don't see any variance here. The only negative number is the
> minimum.
>
> Ciao,
> Paolo
Those are the values of the vector AFTER the variance was computed, it
should read Variance_Band1 and Variance_Band2.
|
|
|
|
| Re: Negative Variance? [message #65773 is a reply to message #65771] |
Thu, 19 March 2009 15:08 |
pgrigis
Messages: 436
Registered: September 2007
|
Senior Member |
|
|
anniebryant@gmail.com wrote:
> I am using the VARIANCE comand in IDL and am getting negative values.
> The description in the HELP menu for IDL doesn't provide much
> information, but from what I know, I thought variance was either an
> absolute value or a deviation from a squared number, neither of which
> could yield a negative value.
>
> I am looking at the variance for 2 different bands in an AVIRIS
> image. I use the same .pro, just change the band I am looking at.
> The results for Band 1 (lets call it) make sense: all positive
> values. For Band 2, I get a lot of negatives.?! How is this
> possible.
>
> These are the values for Band 1:
>
> integer mean std dev minimum maximum n_elements
> 2633.3 2059.3 0.0000 9367.0 (160686) =
> 160686
>
> These are the values for Band2:
>
> integer mean std dev minimum maximum n_elements
> 1167.5 18269. -32768. 32767. (160686) =
> 160686
I don't see any variance here. The only negative number is the
minimum.
Ciao,
Paolo
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
