Density Function Probability BIGGER than 1

[ThalesArakawa]

Hi!

I'm having an problem with Fitter. First of all, i'm trying to fit under all distributions, getting a bad result in all. But when i'm looking to the plot, i see this.

An DFP with values bigger than 1.

I try rounding the values, truncate, take absolute values and nothing... In my dataset with more than 150 features i have a couple of relative features (values close to zero) and this problem repeat with this features, the other's is ok.

Here a sample of data

problematic_data.txt

[cokelaer]

@ThalesArakawa . First of all, thanks for using fitter.
Concerning the tile (density function greater than 1), I agree it is not intuitive but correct. What will change the values of the y-axis is the step/bins used in the histgram. Indeed, to compute the final density, this is taken into account. For instance, play with the binning in the histogram (parameter bin or bins) and see the impact. Yet, at the end, sum of the X_i over i times dX (dX is your step, or width of each bar) should be 1.

You data has a large peak at 0 indeed. and then another one at -1. This is clearly not going to fit well with any continuous distribution.

Fitter uses standard distribution from scipy. check their shapes on scipy documentation but I doubt you will find that fit your data. Now, it you can get rid of the peak at 0 and -1, maybe the underlying data would be more interesting. depends on your scientific problem really. You may also consider adding +1 to all the data since your distribution looks like it starts at -1.

I see you have also outliers, maybe you can try to remove some.

best