[GRASS-dev] d.histogram problem [bug #1977]

[Michael Barton]

I can probably write GUI code to make a nice histogram, using output from
r.stats. But Glynn's point about having GRASS tools available for scripting
is well taken. Also, it's a lot easier to simply call d.histogram than to
manually write the code to create a histogram.

Since you folks are looking at the guts of d.histogram, I should point out
that primitive as it is, it can create a couple of statistical graphs
(histogram and pie graph based on internal binning calculations), a bar
graph (what the histogram is actually drawing), and a pie graph. If it can
draw a bar graph, it can probably draw a line graph and scatter plot with
little additional work.

This could make it a fairly comprehensive internal graphing utility, with
limited but useful statistical capabilities, that could be applied to a
variety of tasks (e.g., displays from r.regression, several image processing
functions). 

I don't know if it is worth the effort or not, but maybe it is worth
thinking about for a few minutes at least.

Michael


On 5/15/07 3:04 PM, "Paul Kelly" <paul-grass at stjohnspoint.co.uk> wrote:

> On Tue, 15 May 2007, Paul Kelly wrote:
> 
>> On Tue, 15 May 2007, Hamish wrote:
>> 
>>> ok, so bug #1977 still matters. (axes tags somewhat broken for FP maps)
>>>  http://intevation.de/rt/webrt?serial_num=1977
>> 
>> I wonder is there a bit more too it that just the axes labelling. E.g. I
>> tried comparing the output of "r.stats -c input=slope nsteps=10" to what
>> d.histogram draws for nsteps=10. r.stats says:
>> 0-5.252012 81333
>> 5.252012-10.504024 86135
>> 10.504024-15.756036 60593
>> 15.756036-21.008047 37301
>> 21.008047-26.260059 16918
>> 26.260059-31.512071 6275
>> 31.512071-36.764083 1415
>> 36.764083-42.016095 139
>> 42.016095-47.268107 13
>> 47.268107-52.520119 1
>> * 12929
>> 
>> i.e. it seems obvious the first bar should cover the range 0-5.25012 and so
>> on (although I think it's fine that the x-axis label shows whole numbers; it
>> doesn't need to line up with the bar/bin edges), but in the d.histogram
>> output it seems quite clear that the first bar is drawn between 5 and 10
>> (approximately) on the x-axis.
> 
> Actually, to put that another way you could say the numbers on the x-axis
> are in completely the wrong places. To me, it seems the numbers shown
> under the ticks refer to the floor'ed value (i.e. probably a double cast
> to an int) of the maximum value included in the bin to the right of the
> tickmark, if that makes sense. I.e. the numbers are one-off to the left
> of the ticks they represent, but even if shown under the correct tick they
> would still be wrong because they have been truncated to integers.
> 
> (Looking at the output of r.stats -c xxx nsteps=xx next to the histogram
> plot is really helpful for understanding this.)
> 
> I think the ticks should have no relation to the position of the borders
> between the bars. The left-hand edge of the first bar should start at the
> minimum value of the data and the right hand edge of the last bar should
> end at the maximum value of the data. I think round numbers in between
> should be marked with ticks (spacing worked out somehow from the
> hard-coded maximum 40 bins between numbered ticks).
> 
> That's probably easier to fix, than to put a more comprehensive fix in
> that would do the binning within d.histogram rather than relying on
> r.stats to do it all. Such a solution would enable the number of bins for
> integer maps to be specified too. At present this really doesn't seem
> possible without changing the structure of either d.histogram or r.stats.
> 
> Paul
> 
> 
>> 
>> Also it seems strange that it doesn't allow you to set the number of bins for
>> non-floating point maps.
>> 
>> Also, 255 bins seems not a very good default to me. It is a bit high and the
>> result doesn't really look like a histogram.
>> 
>> I'll try and look into it more if I have time. Always remember doing maths at
>> school and drawing a proper histogram being something that was really easy to
>> get wrong!
>> 
>> Paul
>> 
>> _______________________________________________
>> grass-dev mailing list
>> grass-dev at grass.itc.it
>> http://grass.itc.it/mailman/listinfo/grass-dev
>> 
> 
> 

__________________________________________
Michael Barton, Professor of Anthropology
School of Human Evolution & Social Change
Center for Social Dynamics & Complexity
Arizona State University

phone: 480-965-6213
fax: 480-965-7671
www: http://www.public.asu.edu/~cmbarton