Fitting distribution for discrete/categorical data

[erdogant]

Fitting distribution for discrete/categorical data

[erdogant]

Have you seen this issue #4 about integer values? It sounds similar to yours.
Let me know whether this helps.

[ogreyesp]

Hi,

Thank you for your response. Yes, I have previously seen this answer... However, I suppose that when I will subsequently try to sample some random numbers from the best fitted distribution, the random numbers will be not exactly integers, but floats, isn´t it?

Following with your example, if we do

best_distr["distr"].rvs(*best_distr["params"], size=10)

the result is:

array([42.43419208, 43.9193906 , 39.83836361, 46.84268631, 6.40314481,
6.14620859, 39.51184439, 8.06064685, 5.75366875, 27.2441456 ])

The only thing that I can do right now is to round these numbers and convert it to integers. However, certainly the distribution is not a discrete one, it is still a continuos one.

Have you planned to extend distfit library to include the discrete distributions available in scipy.stats?

[erdogant]

Good point. Im going to look into this.

[wgmueller1]

The R package fitdistrplus has an option where a user can specify that the distribution is discrete. I believe they support binomial, poisson, and negative binomial. If you are planning on implementing discrete distributions it might nice to include a discrete option in the fit function or the instantiation of the distfit object.

[erdogant]

I worked on the implementation of discrete fitting using the binomial distribution. This is now possible in the latest version:

# Update to latest version
pip install -U distfit
import distfit
print(distfit.__version__)

This should be >= 1.2.7

Examples for discrete fitting can be found in the readme, sphinx pages and colab notebook:
https://colab.research.google.com/github/erdogant/distfit/blob/master/notebooks/distfit.ipynb
https://erdogant.github.io/distfit/pages/html/Discrete.html