Re: Histogram

[maxgacode]

Il 24/03/2020 01:55, Peter Johansson ha scritto:
> Hi Max,
>
> I think the current behavior is intuitive. I think it would be 
> unexpected if the count of a bin depend on whether it's a bin in the 
> middle and the last bin.

I have a set of data points and I'm embedding the histogram in a C++ 
class. So I create the histogram starting from a std:vector<double> in 
the class constructor or in a setting function.

In the class the minimum and maximum vales of the data are used to set 
the range

gsl_histogram_set_ranges_uniform(gsl_histogram * h, double xmin, double 
xmax)

then the histogram is filled with

gsl_histogram_increment(gsl_histogram * h, double x)


Once the class is initialized (with data) there is a  member function 
creating a graphical plot from the histogram. It is also plotting some 
statistical properties of the data.

So if I have 100 data-points the shown number of data points may be 100 
or 99 or 98 or...who knows!!! It seems that the plot has lost data points!

It depends from a simple fact. The latest bin may or may not contains 
some of the elements. If x is equal to xmax

xmin + (n - 1) d <= x < xmax

fails to increment the bin.

It may happen if x and xmax are exactly representable.

It's a known problem. At the moment I'm using the following workaround

If xmax and xmin are the minimum and maximum values of data points of 
the vector of data I'm setting the range with

gsl_histogram_set_ranges_uniform(h, double xmin, double nextMax)

where

nextMax = std::nextafter(xmax,std::numeric_limits<double>::max());

So I'm usig the "next" floating point greater than xmax as the maximum 
value for the range. It works...(hopefully!) and no data poinst are missed

Thank your your reply

Massimo

--
PGP key: wwwkeys.pgp.net: 0EBF4A07