[postgis-devel] raster: ST_Histogram - Not sure this is a bug or I just don't understand the numbers

[Bborie Park]

On Sun, Dec 4, 2011 at 5:29 PM, Paragon Corporation <lr at pcorp.us> wrote:
> I was testing out the ST_Histogram and I was expecting my percentage to dd
> up to 1 in both cases.
>
> for example:
> This query:
> SELECT (st_histogram(rast,2,6, ARRAY[5,10,20,50,100,2000])).*
> FROM  postgis_analysis_20
> WHERE descrip = 'dbox3 2011-11-25';
>
> -- gives me this which doesn't seem to jive with the counts (for example I
> would expect the 15-25 bucket to be 0.95
>
>  min | max  | count |        percent
> -----+------+-------+-----------------------
>  10 |   15 |    31 |  0.000563175583613407
>  15 |   25 | 10557 |    0.0958942683259152
>  25 |   45 |   220 |  0.000999182487056045
>  45 |   95 |   135 |  0.000245253883186484
>  95 |  195 |    48 | 4.36006903442638e-005
>  195 | 2195 |    18 | 8.17512943954946e-007
>

I think you're seeing the effect of specifying the bin-widths.  When
the widths are specified, the percentage is not count / total but
rather count / total / width, so is proportionate.  In the link below,
look at Data by Proportion.

http://en.wikipedia.org/wiki/Histogram#Examples

I added that ability for when bin-widths are specified, but don't know
if it should remain due to the confusion...

>
> -- this query looks like what I would expect - counts and percents agree
>
> SELECT (st_histogram(rast,2,6)).*
> FROM   postgis_analysis_20
> WHERE descrip = 'dbox3 2011-11-25';
>
>
>       min        |       max        | count |       percent
> ------------------+------------------+-------+----------------------
>               10 | 307.333333333333 | 11003 |    0.999454991370697
>  307.333333333333 | 604.666666666667 |     4 | 0.000363339086202198
>  604.666666666667 |              902 |     0 |                    0
>              902 | 1199.33333333333 |     0 |                    0
>  1199.33333333333 | 1496.66666666667 |     0 |                    0
>  1496.66666666667 |             1794 |     2 | 0.000181669543101099
>
>
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-devel