[BUG] Distributions.Empirical.create produces strange propabilities

[fslaborg]

[BUG] Distributions.Empirical.create produces strange propabilities

[bvenn]

Thanks for the report.

You are right, the creation of the probability mass function (PMF) is incorrect.
Since the PMF must sum up to one, all bin sizes are divided by the total count of observations. This area calculation seems to be faulty.

If empty bins are present in the data, the area is overestimated (acc + (abs (x1 - x0)) * ((y0 + y1) / 2.)).

It is sufficient to divide the bin counts by the total number of items. A fix is prepared and will be pushed soon.

[bvenn]

Additional comment:

There are functions to normalize the Map generated by Empirical.create. I think the normalization should be incorporated into the create function to avoid confusion about probabilities that do not reach 100%.

[bvenn]

closed by 75f5262