This component identifies the five best-fitting continuous probability distributions for numeric columns using Python’s Fitter library.
It supports 80 different continuos probability distributions and plots the best five distributions and probabily plot based on the least sum of squared errors.
The component configuration window allows to select the input colomn(s) to fit the distribution. The Python's library behind the component is able to handle datasets of max 10K rows. Larger datasets will be downsampled to 10K rows.
The result of the component comprises a table with the five best fit distributions, probability plots, and a table with descriptive stats for the selected column(s).
Which kind of extension are necessary to load the Continuous_Distribution?
Are CONDA installation enough to install and run this component?
Thanks for answering.
The component has an important limitation: it doesn’t report the parameter of the five best fitting distribution. Therefore, once I’ve identified the best probability distribution, I’ll need to perform a further targeted fitting operation using the identified distribution to extract the distribution parameters. I solved this problem, see my next post!!!
I modified the main component’s Python script to report the parameters of the fitted distributions in a tuple. I also streamlined the component by eliminating the input statistics for all unnecessary dataset variables: I just reported the descriptive statistics of the selected varible used in fitting. If anyone is interested, I’ll share the modified component.